The interaction of economic and political inequality in Latin America: a view from Colombia

Abstract

We investigate how economic inequality can persist in Latin America in the context of radical falls in political inequality in the last decades. Using data from Colombia, we focus on a critical facet of democratization—the entry of new politicians. We show that initial levels of inequality play a significant role in determining the impact of political entry on local institutions, policy and development outcomes, which can impact future inequality. A vicious circle emerges whereby policies that reduce inequality are less likely to be adopted and implemented in places with relatively high inequality. We present evidence that this is caused both by the capture of new politicians and barriers to institution and state capacity building and by the fact that politicians committed to redistribution are less likely to win in relatively unequal places. Our results, therefore, help to reconcile the persistence of economic inequality with the new political context.

inequality, political entry, public policy, development

Introduction

Over the past 200 years, Latin America has diverged economically from North America. At the time of independence, some parts of the sub-continent, like Argentina and Cuba, had income per-capita levels close to those of the USA (Coatsworth 1993; Engerman and Sokoloff 1997). Today, on average, Latin American countries have around 15 % of US levels of prosperity.

However, while historic levels of development might have been similar in some dimensions, the different parts of the Americas differed quite radically in at least two key others. The first is inequality of incomes and assets. The second is the distribution of political power. Though there is an intense historical debate on the extent of comparable economic inequality in the Americas 200 years ago,¹ a great deal of evidence suggests that it was high. For example, the evidence for greater inequality in human capital in Latin America is incontrovertible. The evidence on the inequality of political rights is similarly clear (Engerman and Sokoloff (2005) and Acemoglu and Robinson (2008a) on both of these). The preponderance of research suggests these are causally related, with the inequality of political power often seen as the key forcing variable creating institutions and policies that lead to economic inequality. For example, García-Jimeno and Robinson (2011) argue that the greater political power inequality in nineteenth-century Latin America led to the very skewed distribution of land in frontier expansion. This concentration of land, a form of economic inequality, feeds back and reinforces the initial political inequality that created it. Relatedly, the theoretical framework in Acemoglu et al. (2005) captures this feedback, where policies are deliberately chosen to sustain political and economic inequality, leading to under-provision of public goods, such as education, and consequentially low rates of economic growth and economic divergence (see Engerman and Sokoloff (1997); Bourguignon and Verdier (2000); Acemoglu and Robinson (2012) and Acemoglu (2008); Dell (2010) and Ferraz et al. (2020) for econometric evidence). Eslava and Valencia Caicedo (2023) survey the literature and overview the historical roots of inequality in Latin America and its expression along several dimensions, noting the deep institutional roots of persistent inequality.²

These observations motivate an essential question for contemporary Latin America. Why does economic inequality remain high and exceedingly inertial, while the second factor above, inequality of political power, has changed dramatically? This transformation is manifested most obviously in the democratization of Latin America. Figure 1 plots some standard measures of the extent of democracy from the VDEM project. No matter the definition, there have been dramatic improvements in democracy since the collapse of military dictatorships in the 1980s and the arrival of democracy in Central America for almost the first time. Compared to other world regions, only Eastern Europe has experienced a comparably large and fast improvement in democracy (Fig. 2)³.

Figure 1

Long-term democracy trends in Latin America and the Caribbean. Notes: Evolution of democracy indices in Latin America and the Caribbean (1800–2000). Countries included in the sample: Argentina, Brazil, Bolivia, Chile, Colombia, Costa Rica, Cuba, Dominican Republic, Ecuador, El Salvador, Guatemala, Honduras, Mexico, Nicaragua, Panama, Paraguay, Peru and Uruguay. Indices definition (defined on a scale of 0 to 1, where 1 is a perfect score): (a) Electoral democracy index—to what extent is the ideal of electoral democracy in its fullest sense achieved? (b) Liberal democracy index—to what extent is the ideal of liberal democracy achieved? (c) Participatory democracy index—to what extent is the ideal of participatory democracy achieved? (d) Egalitarian democracy index—to what extent is the ideal of egalitarian democracy achieved. Egalitarian democracy is achieved when rights and freedoms of individuals are protected equally across all social groups, and resources are distributed equally across all social groups; three groups and individuals enjoy equal access to power. Source: V-Dem project- Coppedge et al. (2021).

Figure 2

Democracy trends for different world regions. Notes: Evolution of democracy indices for different world regions (1870–2000). Electoral democracy index—to what extent is the ideal of electoral democracy achieved in its fullest sense? Source: V-Dem project-Coppedge (2021).

This democratization also seems to have led to a genuine entry of new parties and peoples into the political system. A prominent example is the Workers’ Party of Brazil, whose first president, Luiz Inácio Lula da Silva, was a working-class former metal worker and trade unionist. Another example is the Movement Towards Socialism in Bolivia, whose first President, Evo Morales, was of indigenous origins, a former coca farmer, and ex-head of the Cocaleros trade union. People from clearly non-elite backgrounds have been or are presidents of Chile and Perú, and the president of Venezuela, Nicolás Maduro, was a former bus driver. These people and the parties they represent seem to have genuinely different interests, ideologies and policy preferences from the people and parties they replaced. These changes are particularly relevant in a world where politicians cannot commit to policy as in the citizen candidate model of political competition developed by Besley and Coate (1997).

If political inequality has been the driving force behind economic inequality and poverty in Latin America, why has this dramatic increase in political equality not manifested in new inequality and development dynamics? The patterns we have discussed rule out the simplest mechanism, related to the direct persistence of the same elites in democracy (Stone 1992).⁴

While the question we address has not been clearly articulated before, there are several arguments as to why greater political inequality might not have reduced inequality in Latin America.⁵ Most arguments emphasize that though political power has become more equal in some dimensions, this transformation has been imperfect and insufficiently comprehensive, or its consequences can be offset via other mechanisms.⁶ Therefore, power relations have not changed substantially; thus, neither have policies, economic institutions, opportunities, and, ultimately, economic inequality.⁷

We emphasize several mechanisms explaining why democratization may fail to foster inequality-ameliorating policies. All mechanisms emerge from the fact that a small subset of people controls the mass of economic resources when inequality is high. They also undoubtedly monopolize cultural capital and access to high-status activities and positions. This control allows elites to influence and capture new politicians and citizens. This capture may be voluntary as new elites adapt to existing values, hierarchy and status systems.

It is helpful to distinguish between two broad classes of mechanisms: incentives and selection. Concerning incentives, a first subset of mechanisms stems from the fact that the higher is inequality, the more threatening democracy is to elites (as in the analysis of Acemoglu and Robinson (2006)), and the greater their relative ability to influence policy (see Bénabou (1996)). This influence may manifest itself in different ways. The first and most obvious is outright corruption. Corrupting influence can occur either at the level of the politician or at the level of the voter. In the former case, elites can bribe politicians not to adopt particular policies or, if they are adopted, not to implement them. In the latter case, inequality may make clientelism and vote buying more attractive and effective, as in Robinson and Verdier (2013) and Stokes (2013).⁸ Second, suppose elites capture or control local institutions, such as state bureaucracies. In that case, they may block the implementation of policies designed to counter inequality (see Zamosc (1986) for a Colombian example in the context of agrarian reform). Third, elites may also have an incentive to directly undermine local state institutions that are a threat to them (as in the models of Besley and Persson (2009) or Acemoglu et al. (2011)) since this will guarantee that potentially threatening policies go unimplemented.

Another subset of incentive mechanisms comes from the fact that inequality may not just impact politics through the actions of pre-existing elites. It may also make rent-seeking by politicians highly attractive. Instead of being manipulated by old elites, politicians prey on them.⁹ It is also plausible that a municipality with high levels of inequality will result in high levels of inequality of resources amongst the candidates. If incumbents have far more resources than entrants, newcomers may have to use certain electoral strategies, like clientelism, to access power. Once in power, these strategies limit their ability to transform local society.

Concerning selection, high levels of inequality may influence who decides to run for political office. For example, suppose elite-dominated local societies dissuade more educated or reformist politicians from running in the first place. This may occur because the existing elite can better conceal corrupt practices in more unequal places. Interestingly, Garbiras-Díaz (2022) shows, using random audits in Brazil, that increasing the visibility of corruption can help overcome barriers to entry for outsider candidates. More broadly, inequality may reduce the pool of candidates with the resources to compete against those in power. Lastly, inequality can affect social cohesion, also potentially limiting the quality of politicians (Dal Bó and Finan 2018) when voters are willing to vote for less competent politicians so long as they match their identity, which may be class or income-based. Either way, selection becomes a channel for reproducing the status quo and inequality.¹⁰

To investigate these channels, we focus on mayoral elections in Colombian municipalities. This case is relevant for several reasons. Colombia faced a significant process of local-level democratization since the late 1980s. Majoritarian elections for local mayors and departmental governors replaced presidential appointments in 1988. Territorial autonomy and political decentralization also increased. Small local parties emerged, contested and some won in the first local elections. A new constitution in 1991 ratified local democracy and introduced other significant changes for underrepresented groups. It formally recognized that many groups had been historically left out of politics and committed to adopting policies for inclusion (Hoskin et al. 2003).¹¹ In short, replacing formerly appointed mayors with elected ones and extending the political arena to underrepresented groups were clear moves toward local democracy, producing, in particular, the entry of new voices into politics. A secondary, methodological reason why the democratization process of Colombian municipalities is helpful concerns having a large enough sample to evaluate the implications of such an entry empirically.

Thus, this democratic experiment in Colombia is an interesting case study to explore the microeconomic implications of the broad Latin American trends we observed in Figs 1 and 2. Just as Colombia experienced a local-level increase in democracy, the entire region was experiencing a wave of democratization unlike any other. Also, just like Colombia, the region’s improvements in inequality, while complex and multi-faceted, remain insufficient (for instance, Alvaredo (2023) document how many improvements in inequality proved short-lived, with most of the region experiencing an inverted U curve between the 1970s and the 2010s). While one must be careful not to generalize the Colombian experience to the entire region, the country is an interesting laboratory to study the connection between these phenomena.

To measure increased democracy at this level, we examined the entry of people without previous political experience into politics. Presumably, this is a crucial part of the mechanism via which democratization is supposed to change policy, enabling new voices to access political institutions and shape such policies. We compare places electing (by a small margin) a new politician to those where someone who is not new narrowly wins the mayoral election.¹²

The prime way democracy impacts inequality is via changes in policies, particularly redistributive policies and the provision of public goods. We examine three sets of dependent variables. The first captures local policies that mayors and the municipal government control and implement. Municipal governments in Colombia share a crucial responsibility in education, health care and other local public services. We compiled a dataset with various educational and health outcomes measures. These are all listed in Appendix 5. For example, we use information on primary and secondary school enrollment, the teacher/pupil ratio and national test scores for education. For health, we examine infant mortality, the percentage of underweight babies, the government’s subsidized health insurance coverage, and the number of health facilities per capita. Public services include the coverage of aqueducts, electricity, sewage, natural gas and internet penetration. In Colombia, perhaps the most basic public good is security. We measure this using data on thefts, kidnappings, murders and forced displacement, all normalized. Finally, in this set of outcomes, we examine economic development measures. We use mean nighttime light, municipal value-added per capita and the local employment rate.

Our second set of outcomes comes from information collected by the DANE, the Colombian National Statistics Institute (Departamento Administrativo Nacional de Estadística), on four indices of the functioning of local state institutions. The DANE compiled information on the efficiency and efficacy of local state institutions. Efficiency approximates the local public sector’s ‘total factor productivity,’ capturing the relationship between services provided and the inputs used. Efficacy measures the extent to which a particular action generates the desired outcome. The DANE also measured management practices, which captures the extent of bureaucratization and professionalization of the local state. Finally, they collected information on fiscal performance, which captures how effectively fiscal institutions work (whether the local budget is balanced, whether laws relating to expenditure limits are obeyed, and whether the debt is serviced correctly). Additionally, we examine measures of corruption accusations and prosecutions.

The DANE’s indices of fiscal performance and management practices, introduced above, are particularly interesting because they allow us to directly discuss the difference between policy changes and policy implementation. Most of our data are on policy outcomes (e.g. the pupil-teacher ratio in schools), so it is difficult to distinguish between the conceptually distinct situations where a policy is adopted and not implemented and those where a policy is not adopted in the first place. To some extent, fiscal performance and management practices capture policy changes under a mayor’s control. Thus, if one sees these indices improving but no change in other policy outcomes, this suggests a policy change whose implementation fails or is blocked. This pattern would not support some of the hypotheses above, for example, the idea that newcomers behave like incumbents to win and retain power or adapt themselves to some status quo systems of values or hierarchies.

Finally, the third set of outcomes covers different dimensions of corruption. We examine the number of formal complaints and sanctions issued against mayors, using information gathered from the SIRI platform of the “Procuraduría General de la Nación”, the national watchdog agency.

The basic prediction of the above mechanisms is that greater democracy and political equality may accompany unchanged redistributive policy and institutions and, ultimately, unchanged economic inequality on average. Nevertheless, the mechanisms are also considerably heterogeneous, which we exploit to test their relevance. First, in places with greater initial levels of inequality, the above arguments suggest that elites have more substantial incentives and the ability to influence policy choices. Hence, such places should have more political corruption and clientelism as elites make greater efforts to guarantee their interests are not threatened. Second, elites should have stronger incentives to block policy implementation in highly unequal places through their links to the local state. Third, we expect that in areas with high inequality, elites attempt to undermine local institutions to ensure they do not implement policies against their interests. Finally, through the selection channel, high inequality should also impact who runs for political office. For example, one might conjecture that this would lead to more moderate candidates since radicals would have little chance of winning in the face of entrenched elite interests and power.

We focus here on heterogeneity concerning initial levels of inequality because the paper focuses on the reproduction of inequality in the face of democratization and falling political inequality. Fergusson et al. (2023) investigate much more broadly the conditional effects of democratization and what they call the ‘Conditional Iron Law of Oligarchy.’ They develop the idea that increased democracy while leading to the entry of new political forces and people, has potentially diverging impacts depending on the context. In their paper, they show in particular that initial levels of order and state capacity influence whether or not democracy is transformative. The fact that it may not be is due, as in the sociological theories from which their title comes, to the reproduction of elites: new democratic entrants become new elites, replacing old ones with little change in policy or outcomes. Rather than a tool of political inclusion, democracy becomes more of a mechanism for elite circulation. Our main contribution is to use their framework to study initial levels of economic inequality as a contextual factor and show that their results apply there also.

Our evidence shows patterns consistent with many of the hypotheses we sketch above. First, little happens on average in the short run to most policy outcomes when a new politician comes to power. There is no improvement in economic development and corruption outcomes, and the only improvement in policy is for education. Health policy outcomes even deteriorate on average, while other public services are no different in municipalities where a newcomer wins, as opposed to one where a newcomer loses.

Second, concerning institutional performance, we find minor positive effects of newcomers on an index that aggregates all the measures from DANE, driven by an impact on its components of fiscal performance, efficacy and management practices. These results are consistent with democratization and greater political equality having little, or only a modest, effect on policy and inequality. These findings are interesting since, as we stressed, fiscal performance and management practices can be thought of as directly measuring policy adoption, independent of implementation. To the extent that efficacy also increases, there is some evidence that better policies are both chosen and applied.

Third, there is little evidence for an average decrease in corruption when a new politician comes to office. There is no decrease in the likelihood that a mayor will be warned, sanctioned, suspended, removed, disqualified from office, or imprisoned.

More interestingly, however, there is considerable heterogeneity in the impact of newcomers depending on the initial levels of inequality. We measure this level of inequality via the Gini coefficient calculated at the municipality level from the 1993 Colombian national census and Encuesta de Calidad de Vida. Municipalities with below-median inequality drive the positive effect on education. The negative impact on health policy goes away in these relatively more equal municipalities. The effects on institutional performance also vary between high- and low-inequality municipalities. Municipalities with relatively low levels of inequality account for all of the average improvements we mentioned above. Finally, for corruption, mayors of low-inequality municipalities are less likely to be disqualified for public office. At the same time, those in highly unequal places are more likely to be sanctioned and imprisoned for corruption.

This evidence suggests that newcomers have different policy preferences from existing political elites, but whether or not they choose or manage to implement different policies depends on the initial level of inequality. Indeed, if we consider fiscal performance and management practices as measures of policy adoption, these do not change in highly unequal municipalities when a new mayor is elected. This indicates that elites bribe such mayors not to choose different policies or that mayors do not change policies because they aspire to merge into the existing status hierarchy. Our results on corruption suggest that the first mechanism applies here.

These results also show that the fact that we do not find any impact of newcomers on policy outcomes in highly unequal municipalities does not simply reflect a lack of state capacity or elites influencing bureaucracies not to implement policy. If this were so, we would expect to see policy change without implementation, but this is not the pattern we observe. Since we do not have detailed information on manifestos or campaign promises, we do not know whether newcomers in highly unequal municipalities offer different policies and then fail to adopt them once elected (a well-documented phenomenon at the national level in Latin America; see Stokes (2001)), or whether they instead never promise different policies. Finally, our findings are potentially consistent with the idea that mayors in unequal municipalities turn predatory. The fact that policy does not change and there are increases in corruption could, in addition to reflecting elites bribing mayors, be consistent with new mayors turning on elites. A more benign interpretation would be that to win in such a context, one has to ‘fight fire with fire.’ Therefore, after winning, better policies might be hostage to repaying debts with state patronage and other means, making it impossible to improve policy or public goods outcomes.

These interpretations leave the issue of selection open. The lack of policy change in highly unequal municipalities could be caused by different types of newcomers entering. To win in such an environment, it is possible that only newcomers who promise not to ‘rock the boat’ stand a chance of winning. Indeed, newcomers differ from old politicians along several observable characteristics, including ideology, gender and age, even when restricting the comparison to close electoral races. Therefore, ‘selecting’ politicians with these different characteristics, beyond the newcomer condition per se, may partly explain the effects of newcomers. Using the methods in Torres (2023); Fergusson et al. (2023) find that the ‘pure’ newcomer effect on outcomes is generally more positive than when not correcting for the influence of such selection. In our context, we find that in many dimensions, there are no significant differences between newcomers and non-newcomers in municipalities with high or low inequality. But there do appear to be some interesting selection effects. In close elections, newcomers from left-wing parties tend to be less frequent in municipalities with high inequality. Moreover, they also tend to be poorer. The first finding is likely caused by left-wing parties finding it challenging to win in places with entrenched elites (Fergusson 2021b). This reduces the likelihood of inequality-reducing policies being adopted.

This is suggestive evidence that selection, as well as incentives, drive the differences in the impact of newcomers by levels of underlying inequality, and both sets of mechanisms help to perpetuate disparities even in the face of large falls in political inequality.

Since economic and political elites overlap, especially given Latin America’s institutional roots, this may complicate the interpretation of some of our results. For instance, they may reveal the ascent of new economic elites who transform their economic power into political power rather than the entry of new political voices. While this may be partly so, even such a phenomenon is interesting as part of the circulation of different economic elites that may be essential to processes of creative destruction that promote shared prosperity Schumpeter (1942). Moreover, while the overlap between economic and political elites is significant, they are not the same even in the Latin American context. In the context of Colombia, Chaves et al. (2015) present historical evidence of the overlap and lack thereof between political and economic elites, using archival records of landowners and access to formal political power in the department of Cundinamarca. They reveal substantial heterogeneity in the degree of overlap, with some places with ow overlap having significantly lower incidence of practices of electoral fraud and better economic outcomes. More broadly, historians of Latin America contest the simplified view that economic elites, historically landed elites, necessarily dominated politics (Brading 1973; Safford 1985; Schwartz 1996; Hora 2001). More recently, Rovira Kaltwasser (2018) documents patterns of both continuity and true circulation of political elites In Latin America.

The paper proceeds as follows. In Section 5, we examine the evidence of political change in Colombia over the past 30 years, documenting the extent to which there has recently been significant political entry. In Section 5, we discuss the data we use in the study and the identification strategy that we use to tackle the issue that it is endogenous where new politicians get elected. Section 5 newcomer incidence on policy and institutions by initial levels of inequality. Section 5 concludes.

Political change in Colombia

A brief history of Colombian democracy

In the Latin American context, Colombia stands out for its long and comparatively strong democratic tradition (see Posada-Carbó (2017) and Posada-Carbó (2020) for an introduction to the comparative historical evidence). Elections have been the salient path to political power since at least the Liberal Republic of the 1850s. Even if they were often boycotted and violent, clientelistic and fraudulent, the aspiration was always to return to elections, which were seen as ways to share power and reconcile different interests (Mazzuca and Robinson (2009)).

The result was a democracy that was quite restrictive (Bejarano and Pizarro 2005). After the 1850s, Liberal and Conservative parties dominated the political arena until the 2000s (Bushnell 1993). These parties competed in elections and frequent civil conflicts. The last of such disputes, commonly referred to as “La Violencia” (The Violence), occurred between 1946 and 1957, intensifying after Liberal presidential candidate Jorge Eliécer Gaitan was murdered in 1948.

Facing political instability, General Gustavo Rojas Pinilla overthrew conservative president Laureano Gómez in 1953 and created a military dictatorship. This short spell of dictatorship in Colombia’s otherwise democratic tradition ended in 1957 when the party elites signed a power-sharing deal between the Liberals and Conservatives, called “El Frente Nacional” (National Front) (see Hartlyn (1988) for an overview). The parties alternated executive power and divided seats equally in legislative bodies for the next 16 years. The National Front successfully ended “La Violencia”, but excluded other political forces and served as a power-preserving mechanism for different factions of economic elites (Bushnell 1993). Political exclusion and weak state presence in several territories led to the formation of left-wing guerrillas such as the FARC and the ELN. Right-wing militias or paramilitaries then joined the scene to attack the guerrillas, often spontaneously formed by local people but then subsequently financed by large landowners and druglords and often colluding with the Colombian army.

Colombia’s bipartisan dominance outlived the formal agreement until the late 1980s when the system opened substantially, and the country joined its version of the ‘Third Wave’ of Democratization (Huntington 1993). A particularly noteworthy change was the introduction of local elections for formerly appointed mayors of municipalities and governors of departments in 1988. This significant change was ratified with the new constitution of 1991, which, in addition, opened up the political arena by allowing citizens to run independently or create new parties by collecting signatures. It also reserved seats for indigenous and Afro-Colombian communities in the National Congress. Finally, it established public financing and media access for all political parties, including those newly formed.

Political competition increased, and underrepresented minorities increasingly entered politics through third parties. At the same time, splinter factions of the traditional parties also emerged and competed in elections, sometimes absorbing even smaller parties. Indeed, electoral rules favored smaller parties (Pizarro-Leongómez 2002) leading to what came to be known as “operación avispa” (literally, ‘operation wasp’), fragmenting political parties into small and personal electoral enterprises.

To counteract these trends, a new political reform in 2003 (Legislative Act number 01 of 2003) sought to strengthen political parties.¹³ The reform consolidated the relatively more popular new parties and hindered the growth or even annihilated many small parties and movements, including some ethnic parties. Although a constraint for underrepresented groups at the national level, it did not undermine their success locally.

Indeed, the number of new political parties registered in the country shows virtually no change for over a century and up to the reforms of the late 1980s and early 1990s (Fig. 3). Then, the number of officially registered parties exploded and reached almost |$1000$| in 2003. While the 2003 reform reduced this number substantially, traditional bi-partisanism was long gone: Colombia had tens of registered parties, and new ones kept emerging to reach hundreds again by the end of our period.

Figure 3

Entry of new political parties and movements in Colombia. Notes: Number of newly registered parties and movements in Colombia. Source: Cabra-Ruíz (2023) and Torres (2024).

The 2003 reform emerged largely as a consequence of the crisis of party institutionalization following the changes after the 1991 Constitution. These changes allowed new political forces to participate in electoral processes and increased political competition, but they also exacerbated a crisis of legitimacy and organization for traditional parties (Vélez et al. 2006). Traditionally from the Liberal Party, candidate Álvaro Uribe ran as an independent, partly to avoid the bad reputation of traditional parties. Aware of this problem of party de-institutionalization, he pushed (like other Presidents before him) for a change of the electoral system.

The 2003 reform hurt the electoral interests of several incumbent politicians in Congress, and similar attempts had failed before for this reason. The 2003 attempt was successful for several factors (for detailed analyses, see Shugart et al. (2007) and Vélez et al. (2006)). First, the reform occurred shortly after the landslide victory of Álvaro Uribe. His popularity and the threat of a comprehensive referendum (with a less fundamental electoral reform but other changes that many in Congress feared) motivated Congress to approve its own reform. Second, fragmentation inside Congress increased, with a fall in the number of legislators officially elected by the Liberal and Conservative parties (and a higher proportion of legislators that had promised loyalty to the president). Third, in some politicians’ calculations, defending the system that had given them access to Congress might create barriers over the long run. Finally, a compromise implied that parties could decide whether to include preferential voting in tier party lists, a ’loophole’ that would significantly undermine the reform’s ability to re-institutionalize parties.

In short, as elsewhere in Latin America, political inclusion has increased significantly in recent decades. This is not to deny the existing challenges of Colombian democracy. These include the persistent violent exclusion of some groups (Acemoglu et al. 2013; Fergusson 2021b) and the regional differences in political representation (Fergusson 2017). But, as noted regarding the ‘wasp operation,’ the increase in the number of parties partly represents the entry of new voices in the political arena and partly the old parties’ new strategies.

Documenting the recent transition

While our focus is on local office, some national-level patterns are helpful to highlight the broader relevance of political inclusion in Colombia over the last 30 years. Data from the National Congress helps illustrate one implication of increased political inclusion: fiercer competition. In Fig. 4, we observe that tenure in Congress decreases for every younger cohort. For instance, the ‘class’ of 2006 survived two consecutive periods at a rate of under |$20\%$| compared to close to |$25\%$| for those entering in 1991 and |$35\%$| for those entering in 1978. The 2006 rookies persisted three consecutive periods at a rate of under |$10\%$|⁠, compared to |$20\%$| for those first elected in 1978. In short, it has become increasingly challenging to remain in positions of political power.

Figure 4

Congressional survival rates by cohort of initial entry. Notes: This figure depicts, on the y-axis, the share of congress members elected in a particular year (1978, 1991 and 2006) who remained in Congress for at least the specified number of consecutive terms indicated on the x-axis. Source: Torres (2024) and Torres (2024).

We also have direct evidence that some of these entrants differ from traditional political players. The left panel in Fig. 7 shows that new identity groups who were previously excluded and improved their political clout since the 1991 Constitution are more active in politics. The right panel shows one dimension of inclusion that is easy to map across all offices (executive and legislative) and levels (local and national): female participation. The share of female candidates since the 1980s follows an increasing trend for every office.

For our empirical analysis, we will focus on a broad indicator of entry: newcomers. For each election, we look at all national, regional and local elections and define a newcomer as someone who has never previously held public office via democratic elections.¹⁴

With the political transformations since the late 1980s, people who never held public office increasingly competed and won at the local, regional and national levels. Figure 5 shows that in the Upper and Lower Houses of Congress, the share of newcomers competing doubled after the constitution. Besides the remarkable trend, the levels are quite important in some elections. For instance, almost 90% of the municipal council candidates were newcomers in 2019.

Figure 5

Evolution of the share of newcomer candidates in Colombia by type of election. Notes: Share of newcomers competing in elections, 1974–2022. A newcomer is an individual who has not been elected for any public office. Series are plotted after at least ten years of public elections for that specific public office were held for the first time. Source: Torres (2024).

In short, groups with different identities and actors without previous political involvement acquired political power. Their inclusion also coincided with a meaningful increase in political competition. These figures suggest a substantial change in Colombian politics over the last few decades. On the one hand, they paint an optimistic picture. Previously excluded groups with new identities and ideas accessed power. Indeed, the democratic opening in Colombia coincided with a significant improvement in poverty levels, public service provision, health coverage and infant mortality rates. These achievements are commonly attributed to the new constitution (Vallejo 2014). On the other hand, Colombia remains highly unequal and poor and has been unable to catch up with more affluent nations. How can we explain that the fall in political inequality does seem to have brought some benefits but has failed to be transformational more broadly? In what follows, we will rely on the richness of local variation to examine the causal connection between newcomers and development outcomes more closely and explore its limits.

Research design

The data

The politicians

We collect information from a variety of sources.¹⁵ We rely on Colombia’s Election Data Archive, compiled from official records from the Registaduría Nacional del Estado Civil by Torres (2024). This dataset contains the votes and winners of all local, regional and national elections held in Colombia. Our primary focus is on mayoral municipal elections. Municipal contests occur in October for terms starting in January of the following year. Between 1988 and 1994, we do not have vote counts for losers, only for winners and the aggregate. Regression Discontinuity analysis requires data for the runner-ups and thus relies on information from 1997 to 2019.¹⁶ The 1997 and 2000 elections are for 3-year periods (starting in 1998 and 2001, respectively), whereas later elections are for 4-year terms.

To identify candidate characteristics and, specifically, whether they are newcomers, we use data on the political trajectories and personal attributes of 455,000 candidates compiled by Torres (2024). Due to a non-disclosure agreement with the Departamento Nacional de Planeación, politician attributes from the Sistema de Identificacion de Potenciales beneficiarios de Programas Sociales (SISBEN IV) are not available in the public version of our dataset. To establish parties’ ideology, political scope and whether they represent identity groups, we use data from Cabra-Ruíz (2023), which characterizes all political parties, movements and coalitions from 1958 to 2022.

Measuring public policies, policy and development outcomes

We examine the effect on several dimensions of public services and economic development, testing whether newcomer politicians increase public goods provision, education quality and coverage, health provision quality and coverage, public services, security and finally, economic performance.

The data on these public policy outcomes comes from CEDE’s (Centro de Estudios sobre Desarrollo Económico at the Facultad de Economía de la Universidad de los Andes) municipal panel compiled by Acevedo and Bornacelly (2014) and the national planning office information contained in the “Terridata” server. We also extract municipality-level geographic and historical social characteristics from these sources to use as controls. Additionally, we use nightlight intensity information from the National Oceanic and Atmospheric Administration, following the extensive literature in development economics that utilizes this measure as a proxy of local economic activity (Henderson et al. 2012).¹⁷

Given the many outcomes, we group variables into thematic indices (and indices of each topic into a single development index) following Kling et al. (2007). We do this not only for ease of interpretation but also to avoid cherry-picking and multiple hypothesis-testing biases.

To construct the indices, we normalize and standardize specific components (relative to the mean and standard deviation of the control group) and aggregate them using a simple average. A brief discussion on the construction of these indices can be found in Appendix B1. We nonetheless report effects on each component for transparency, with the caveat that there is an increased probability of asserting false positives. The indices’ components were selected following a reliability and coverage criterion.¹⁸ Table B1 describes the composition of the indices.

The second set of indices is from DANE, also listed in Table A1, which allows us to measure local institutional quality and efficiency and distinguish between policy adoption and implementation.

Finally, as a proxy for corruption, we use data on the number of formal complaints and sanctions issued against mayors, using information gathered from the SIRI platform of the “Procuraduría General de la Nación”, the national watchdog agency. This is an imperfect measure of corruption since much may go undetected or sanctioned. Moreover, accusations of corruption may be used as a political tactic unrelated to corrupt practices or activities. Nevertheless, the “Procuraduría” enjoys a reputation of independence and autonomy, and it is the best systematic data available to examine this critical issue. See section G of Table A1.

Identification strategy

Model

We follow an identification strategy centering on tightly contested mayoral elections featuring newcomers and incumbents within a Regression Discontinuity Design framework. Specifically, it focuses on municipalities where a newcomer narrowly wins over a non-newcomer (the ’treated’ group) and compares them to those where a newcomer barely loses to a non-newcomer (the ’control’ group). By limiting our analysis to these electoral contests, it is possible to estimate the effects of newcomer victories on outcomes while controlling for other municipal-level characteristics that could potentially confound the results. There were numerous newcomers during our period of study. Figure 5 shows |$59.3\%$| of mayoral candidates have been newcomers since |$2000$|⁠. Furthermore, |$55.8\%$| of the elected mayors (e.g. election winners) during this period are also newcomers.

Furthermore, we concentrate on a newcomer’s influence in a particular outcome relative to the previous electoral term. To establish a benchmark, we consider the average disparities in outcomes between the treated and control groups as observed in the period leading up to the election. Formally, let |$E_{i}$| denote an electoral term. The average improvement between two consecutive terms |$E_{i}$| and |$E_{i-1}$| for an outcome variable |$Y$| in municipality |$m$| is as follows:

The benefit of this transformation is that it only captures changes that the incoming politicians initiated, not the initial circumstances they faced. As a result, studying relative differences rather than absolute changes better reflects the desired experimental setting to the extent that we account for significant disparities in baseline outcomes across municipalities.

To construct the running variables and define the election terms, we employed available information on all ordinary mayoral and council elections held in Colombia between |$1997$| and |$2019$|⁠. This enables us to construct a panel dataset at the municipality-election level. Another important aspect is that office terms were modified in this study period. For instance, mayors elected in 1997 and 2000 (and who came into power in 1998 and 2001, respectively) had three-year terms. In contrast, winners in 2003, 2007, 2011, 2015 and 2019 (with start dates of 2004, 2008, 2012, 2016 and 2020 respectively) lasted 4 years in office.

We then estimate the model:

$$ \begin{align}& \Delta Y_{m,E_{i}}=\rho\, D_{m,E_{i}}+f(Z_{m,E_{i}})+\delta_{E_{i}}+ \gamma^{T} W_{m}+ \varepsilon_{m,E_{i}}\end{align} $$

(1)

The running variable is |$Z_{m, E_{i}}$|⁠, which contains the vote margin between the best-ranked newcomer and the non-newcomer candidate with the most votes in the elections for period |$E_{i}$|⁠. We restrict the analysis to elections where the winner and the runner-up are not simultaneously newcomers or non-newcomers. The treatment |$D_{m,E_{i}}=\mathbbm{1}[Z_{m,E_{i}}>0]$| is a dummy variable describing whether a newcomer candidate won the race or not. |$f(\cdot )$| is some unknown smooth function. We include election fixed effects (⁠|$\delta _{E_{i}}$|⁠) and municipality-level controls as in Fergusson (2021b) to improve precision.

Validation

We now discuss the plausibility of the underlying identification assumptions in our empirical strategies. The RD design assumes that the unobserved conditional expectation functions of potential outcomes are continuous at the threshold. We can indirectly examine how credible this assumption is by exploring if other covariates vary smoothly at the threshold (Lee and Lemieux 2010). Following de la Cuesta and Imai (2016), Table C1 estimates |$\rho $| with model (1) for a rich set of different municipal possible confounders. We find no statistically significant differences at the threshold.

Actors should also have no precise control of election outcomes when vote margins are tight for our causal interpretation to be valid. One implication is that the probability density function of the running variable should be continuous at the threshold. Figure C4 presents tests by McCrary (2008) and Cattaneo et al. (2020) that fail to reject the hypothesis of no sorting at the threshold.

We additionally conduct validation tests to assess the validity of the heterogeneous effects contemplated in our study. Specifically, we verify the balance of municipal covariates around the threshold when restricted to elections within the high- and low-inequality groups, as detailed in Table C2. Moreover, we find no manipulation or sorting within these subsamples, as depicted in Fig. C5.

Table C2 also shows that high- and low-inequality municipalities differ along dimensions other than inequality (e.g. some geographic characteristics and population size). Such differences may partly explain the heterogeneous effects we find, which is a general caution to bear in mind with heterogeneous effect analyses. Nevertheless, as we show below, the bulk of the evidence we present, guided by our theoretical discussion, is consistent with inequality being a critical mediator of these differences.

Newcomers’ impact on policy and institutions

In this section, we study the extent to which newcomers shaped different policy and institutional outcomes, focusing specifically on the differential influence exerted by newcomers in municipalities characterized by high and low levels of inequality. We categorize the municipalities into two groups based on the 1993 GINI index: those with above-median and below-median inequality levels.¹⁹

Newcomer policy preferences

An essential assumption underlying our analysis is that newcomer politicians bring new voices and policy preferences to the political arena. As noted in the introduction, it is well established that Colombia’s introduction of local elections in the late 1980s and the legislation that followed the 1991 Constitution facilitating entry into electoral politics effectively implied that these new politicians largely represented formerly excluded groups and brought new perspectives and preferences. Previous analyses have demonstrated the impact of these changes on allowing ‘third forces’ (different from the two traditional parties and their subsequent multiple factions) to successfully contest elections and access power (Shugart et al. 2007). Looking specifically at local elections for mayor, García Sánchez (2000) underscores the heterogeneity in different municipalities, yet a common trend where newcomers tend to represent previously sidelined groups like left-wing (a phenomenon also analyzed in Fergusson (2021b)) and ethnic parties.

To investigate whether these new newcomer politicians bring new preferences and ideas to policy, we further undertake an empirical exercise looking at national legislators. In Table 1, we systematically review the topics of legislation presented in Congress between 2006 and 2018. We use the repository of legal initiatives from Congreso Visible (2022), which contains the topics and politician-level vote records for each legal project.

Table 1

Distinctive legislation topics in Congress: newcomers vs. non-newcomers

	Topics of legislation
	Distinctively newcomer	Non-distinct	Distincitively non-newcomer
1	Environment	Labor	Celebrations, honors and monuments
2	Social security and health	Notarization and registry	Administrative affairs
3	Welfare and poverty	Control organisms and ministries	International humanitarian rights
4	Public administration	Education, science and technology	Public contracts taxation

Notes: Distinctive legislation topic in Congress following the |$\chi ^{2}$| statistic proposed by Gentzkow and Shapiro (2010). The procedure identifies which topics from a given list identify two known groups. To accomplish this, the statistic discriminates that categories seem overrepresented in a particular group while not appearing much in the other and which topics seem equally likely to be seen in either group. The analysis considers the universe of law projects debated by Congressmen between 2006 and 2018 (four legislative periods) and their authors. Newcomers are defined as elected congress members who are representatives for the first time in Congress following Torres (2024). Source: Congreso Visible (2022).

We then use standard text analysis to identify distinctively newcomer and non-newcomer topics. Specifically, we use a |$\chi ^{2}$| statistic Gentzkow and Shapiro (2010) that ranks words or groups of words (in our case, topics from the corpus of legislative proposals) according to how frequently two groups use them. With this statistic, we can identify distinctively newcomer topics as common for newcomers and uncommon for non-newcomers, and vice versa. In contrast, some topics are either frequent or infrequent for both groups, making it difficult to distinguish between them. Table 1 shows that newcomers are more likely to propose bills about the environment, social security, health, welfare and poverty, and, finally, public administration. These topics resonate with an increased importance to previously neglected groups or issues.

On the other hand, the most distinctively non-newcomer topic is ‘celebrations, honors, and monuments,’ a well-known indication of clientelistic exchanges. These politicians are also concerned with ‘Administrative affairs,’ perhaps an indicator of their condition of insiders. Finally, the third and fourth most distinctly non-newcomer topics are ‘International humanitarian rights’ and ‘Public contracts and taxation.’ In short, non-newcomers topics broadly reflect the operation of politics and policies, seemingly from an insider condition. Newcomer topics instead focus on broad social issues of concern. Such a stark contrast in the types of topics for newcomers and non-newcomers confirms that the former represent new policy voices.²⁰

Local development outcomes

Figure 6 presents our findings related to development outcomes. For convenience, we portray three sets of coefficient estimates on each diagram. In particular, we show the average newcomer effect (depicted in green) and then decompose it into high (dark blue) and low (pink) inequality municipalities. Positive coefficients indicate that for a particular outcome, newcomer mayors do better than non-newcomers (in close elections). Negative values indicate the opposite—that newcomer mayors do worse. The bars on the figure around the estimated coefficients indicate the 10% confidence interval. Our analysis reveals that, on average, the influence of newcomers is minimal across almost every considered dimension. Focusing on the average (green) effects first, there is no difference between the development index, which combines all the sub-indices between municipalities where a newcomer just won and one where they just lost. The only instance of an improvement on average is for educational outcomes. This is counterbalanced, however, by a deterioration in the health index, contributing to no significant difference in the aggregate development index. The other indices, like the economic index (capturing night-time light and value-added per capita), the public services index (measuring aspects of basic infrastructure), or the security index, are not different when a new politician wins power.

Figure 6

Effects of newcomers’ victories on development outcomes by levels of initial inequality. Notes: This figure reports RDD coefficients corresponding to the effect that a close victory of a newcomer in a mayoral election impacts different development indices and their components. Black illustrates the average effect estimated for newcomers. Dark grey represents the effect estimated in municipalities with above-median GINI, as measured in 1993. Light grey depicts the effect estimated in municipalities with below-median GINI, as measured in 1993. Indices are built following Kling et al. (2007). Bias-corrected RD estimates are reported with municipality-level clustered standard errors in parenthesis following Calonico et al. (2014a). 90% confidence intervals

Figure 7

New political groups entering politics. Notes: Left panel number of newly registered parties and movements in Colombia for particular identity groups. Source: Cabra-Ruíz (2023). Right panel share of female candidates competing in elections: 1980–2020. Source: Torres (2024).

Nevertheless, pulling apart the estimated effects with relatively high (dark blue) and relatively low (pink) levels of inequality, an interesting picture emerges. Though the aggregate development index again does not differ, the positive educational effect is driven entirely by municipalities with relatively low levels of inequality. Here, the education index has a |$0.87$| standard deviation increase. Moreover, the differential effect between low- and high-inequality municipalities appears statistically significant at a |$10\%$| level, as evidenced by the non-overlapping confidence intervals. In contrast, when new politicians come to power in close elections in municipalities with relatively high levels of inequality, there are no significant improvements in any policy or development outcome compared to when such a mayor loses.

State institutions

We now turn to the impact on local state institutions using our data from DANE. As mentioned above, this is particularly useful in distinguishing between policy adoption and implementation. For example, we do not know whether the many null effects in the last section are due to new mayors not choosing different policies from old politicians or whether they did adopt such policies but could not implement them. Of course, the different educational outcomes in low-inequality municipalities suggest that new policies are chosen and implemented at least in these contexts.

Figure 8 illustrates the varying influence of new political actors on local institutions, policy adopted and performance. As in the last section, we depict three different outcomes: the average newcomer effect (green) and the impact in high (dark blue) and low (pink) inequality municipalities. There is some improvement in the overall performance index (RD estimate: |$0.31$| SDs, SE: |$0.13$|⁠), which appears to result from improvements in management practices, fiscal performance and efficacy. In light of our previous discussion, this last finding is important since it indicates that newcomers adopted different policies—they improved management practices and fiscal performance, leading to improved efficacy and overall performance. Better management practices and budgetary performance might indicate less patrimonial practices inside state institutions and the use of more bureaucratic procedures.

Figure 8

Effect of newcomers' victories on performance by initial levels of inequality. Notes: This figure reports RDD coefficients corresponding to the effect of a newcomer's close victory in a mayoral election in management practices indices and municipalities characteristics. Black illustrates the average effect estimated for newcomers. Dark grey represents the effect estimated in municipalities with above-median GINI, as measured in 1993. Light grey depicts the effect estimated in municipalities with below-median GINI, as measured in 1993. Indices are built following Kling et al. (2007). Bias-corrected RD estimates are reported with municipality-level clustered standard errors in parenthesis following Calonico et al. (2014a). 90% confidence intervals.

Moving to the heterogeneity by initial inequality levels, municipalities with relatively low levels of inequality seem to drive the overall improvement in performance. Indeed, the point estimate is bigger for less unequal municipalities (RD estimate: |$0.54$| SDs, SE: |$0.19$|⁠) than for highly unequal municipalities (RD estimate: |$0.24$| SDs, SE: |$0.11$|⁠), where the confidence interval almost includes zero. Newcomers in municipalities at the lower end of the inequality distribution also drive improvements in fiscal performance and efficacy (RD estimates |$0.82$| and |$0.87$| SDs, respectively). The point estimates of newcomers’ effect exceed the average and remain significant at |$10$|%.

These results seem consistent with those for the development outcomes. On average, the election of a new mayor leads to some benefits, but in places with relatively low levels of inequality, one sees much more systematic improvements.

Corruption

Finally, we turn to corruption. In Fig. 9, we present the results of our findings in the same way we have done so far. Mirroring the results so far, on average (green), sanction frequency appears similar for newcomers and non-newcomers. The Regression Discontinuity point estimate is |$0.03$| standard deviations, with a standard error of |$0.09$|⁠. Looking at the different types of sanctions, we again see little difference. Newcomers are as likely to be disqualified from public office, suspended, removed from office, imprisoned and issued a written warning or fined. There is little evidence here that the ‘political selection’ of newcomers leads to better outcomes concerning corruption.

Figure 9

Probability mayor receives a sanction after a tight newcomer victory by initial level of inequality. Notes: This figure reports RDD coefficients corresponding to the effect of a newcomer’s close victory in a mayoral election in the probability a mayor receives a sanction. Black illustrates the average effect estimated for newcomers. Dark grey represents the effect estimated in municipalities with above-median GINI, as measured in 1993. Light grey depicts the effect estimated in municipalities with below-median GINI, as measured in 1993. All dependent variables are standardized. Bias-corrected RD estimates are reported with municipality-level clustered standard errors in parenthesis following Calonico et al. (2014a). 90% confidence intervals.

Yet, switching to the analysis of heterogeneity, some interesting patterns emerge. In municipalities with above-median inequality, newcomers seem to be sanctioned more frequently (RD estimate: |$0.25$| SDs, SE: |$0.13$|⁠). Moreover, this increase is statistically significant at a |$10\%$| level. In contrast, in low-inequality municipalities, the point estimate suggests they are sanctioned slightly less often (RD estimate: |$-0.05$| SDs, SE: |$0.14$|⁠). When examining the types of sanctions applied, the elevated sanction rate in highly unequal municipalities seems to be driven by increased rates of imprisonment (RD estimate: |$0.06$| SDs, SE: |$0.02$|⁠) and warnings/fines (⁠|$0.10$| SDs, SE: |$0.06$|⁠). Conversely, the lower sanction frequency in low-inequality municipalities is due to fewer newcomer mayors being disqualified from public office (RD estimate: |$-0.27$| SDs, SE: |$0.13$|⁠), removed from office (RD estimate: |$-0.08$| SDs, SE: |$0.05$|⁠) or imprisoned (RD estimate: |$-0.15$| SDs, SE: |$0.12$|⁠).

Combined with the evidence in the last two sub-sections, a systematic picture appears. In relatively equal municipalities, institutional and some public policy outcomes improve, and the newly elected mayors also seem less corrupt. A natural interpretation is that new mayors have distinct policy preferences and favor more redistributive policies. Still, they only choose and implement these policies in more equal and likely less elite-dominated municipalities.

The role of selection

We now turn to the issue of selection. New mayors, especially those who win close elections, likely differ in other ways from old politicians. These additional differences could confound the results we discussed above. Newcomers often differ from non-newcomers in various observable aspects, potentially explaining their varying performance abilities. Following this line of thought, we investigate whether the degree of underlying inequality influences the types of newcomers attracted to politics, thus potentially explaining the heterogeneity in the performances observed in these places.

Panel A of Fig. 10 investigates the differences between all newcomer and non-newcomer candidates (not simply the winners in close or any election) from 2007 to 2019 based on the historical level of inequality. For simplicity, we again use pink for low inequality and dark blue for high inequality. Coefficients indicate the newcomer effect. For example, in the bottom row, the bin [|$18,30$|] indicates that the mayor’s age is between |$18$| and |$30$|⁠. The positive number shows that newcomer mayors are about |$5\%$| points more likely to be in this bin than non-newcomers. Non-newcomers, on the other hand, are more likely to be over |$50$|⁠. Newcomers are generally younger, more educated and more likely to be female than non-newcomers. However, these characteristics do not exhibit significant systematic differences between high- and low-inequality municipalities. The two exceptions are education and political party affiliation. Newcomer candidates in high-inequality municipalities are more likely to hold a postgraduate degree than their counterparts in low-inequality municipalities. While this difference is significant, it affects only a handful of municipalities, given that only |$7\%$| of mayors have a postgraduate degree. Additionally, on average, newcomers are more likely to run for left-wing parties than right-wing or other parties. Nevertheless, newcomers are disproportionately more likely to run for left-wing parties in low-inequality municipalities than elsewhere. Conversely, this pattern is less pronounced in high-inequality areas.

Figure 10

Newcomer characteristics by initial level of inequality. Notes: This figure illustrates the differences in observable traits between newcomers and non-newcomers, contingent on the level of inequality. (A) shows these differences for all candidates. (B) depicts the differences between winners. Each coefficient represents the difference in percentage points between newcomers’ and non-newcomers’ candidates who possess the trait indicated on the y-axis. The analysis is limited to elections from 2007 to 2019, ensuring consistency with the estimation sample used in other exercises. The inequality distribution is based on the 1993 GINI coefficient. The results are presented with |$95\%$| confidence intervals and standard errors clustered at the municipality level.

Panel B of Fig. 10 focuses instead on the differences between newcomers and non-newcomers among elected mayors. Interestingly, the education advantage of newcomers in municipalities with high inequality versus municipalities with low inequality evens out when only winners are considered. This suggests that highly qualified newcomer candidates win at similar rates in high- and low-inequality areas despite being a more common characteristic of newcomers in the former. Likewise, while newcomer candidates are less likely to run for non-left-wing parties, among winners, successful candidates in highly unequal areas seem to slightly favor right-wing or non-strongly ideological parties, which often represent traditional or large national parties. Newcomer winners are less likely to come from left-wing parties in highly unequal areas, even though newcomers are disproportionately left-wing in these municipalities. Conversely, newcomer winners are more likely to come from left-wing parties in low-inequality municipalities, consistent with the pattern observed when examining all candidates.

Finally, we do a similar comparison in Fig. 11 by examining narrowly elected newcomer and narrowly elected non-newcomers. For ease of comparison with our earlier figures, we now add the average effect (in green) back into the figure. On average, newcomers who win close elections tend to be younger and poorer, and they tend to be from a left-wing party. Interestingly, both latter effects seem to be driven by high-inequality municipalities. Indeed, when we distinguish between these two sub-sets, running for a left-wing party is significantly lower in a high-inequality municipality but not in a low-inequality municipality, and the same is true for the result that new mayors are poor. When limited to this sample, newcomers no longer have relevant differences with non-newcomers regarding completed tertiary education.

Figure 11

Narrowly elected newcomer candidates characteristics by initial levels of inequality. Notes: This figure reports RDD coefficients corresponding to the difference in traits of close newcomer winners to close non-newcomer winners. Green illustrates the average difference estimated for newcomers. Blue represents the effect estimated in municipalities with above-median GINI, as measured in 1993. Pink depicts the effect estimated in municipalities with below-median GINI, as measured in 1993. All dependent variables are standardized. Bias-corrected RD estimates are reported with municipality-level clustered standard errors in parenthesis following Calonico et al. (2014a). 90% confidence intervals. Source: Torres (2024) and SISBEN IV.

The findings in this section fortify the tentative conclusions we reached at the end of the last one. While in high-inequality municipalities corruption seems to blunt the impetus to change and implement redistributive policies, it also appears that candidates who are likely to have redistributive platforms in the first place are less likely to both run and win. No doubt many mechanisms may deter them, perhaps the threat of violence, which has been particularly targeted at left-wing candidates in Colombia, but there also may be cultural barriers that prevent left-wing people from contesting in high-inequality places.

Robustness to politician-characteristic regression discontinuity bias

By highlighting that newcomers and non-newcomers differ on several dimensions, the preceding subsection evokes several issues raised in Marshall (2024) concerning politician-characteristic RD (PCRD) designs. These designs compare places where politicians with trait |$x$| win with those where such politicians lose, helping achieve balance in municipal features among closely contested contests. However, since trait |$x$| is correlated with other traits and may affect politicians’ popularity, the RD estimate may reflect the influence of a complex bundle of correlated traits and not simply the trait of interest.

Moreover, conditioning on close elections and verifying balance on other observable characteristics does not address this problem. The reason is that correlated traits may precisely adjust or ‘compensate’ among close contests (e.g. newcomers contesting close elections may be older or from certain parties, precisely if this helps compensate for some disadvantages of being a newcomer). This complicates the interpretation of the results in most applications since the RD estimates now capture a bundled effect of the trait of interest plus the impact of all remaining potential ‘compensated differentials.’

This subsection addresses these concerns using the methods developed in Torres (2023). Torres (2023) builds on Marshall’s discussion and insights to develop several econometric techniques to assess and correct estimation bias in PCRD designs. In particular, we implement his covariate-adjusted estimator that corrects the portion of the estimation bias resulting from observed politician-level confounders.²¹

As a first step, Fig. C3 illustrates that newcomers narrow winners are disproportionally younger, poorer and less likely to be affiliated with a left-wing party. Table C3 shows that implementing the covariate-adjusted estimator does not change our primary conclusions. Indeed, newcomers appear to have had an even more substantial influence on low-inequality municipalities in dimensions other than schooling. Their influence in municipalities with above-median inequality is non-significant.

Conclusion

While far from perfect, political inclusion in Latin American political systems has improved dramatically over the last several decades. However, this significant increase in political equality, which should have undermined a critical force that sustained economic inequality and blocked development, has not produced a comparable change in inequality and development dynamics. In this paper, we used Colombia’s experience between 1997 and 2019 to investigate hypotheses that could explain why increased political equality has not decreased economic inequality. In this period, the country was dramatically democratized by the entry of new politicians who had never previously held office. Yet we showed that, on average, and consistent with the overall lack of a change in inequality, newcomers had little impact on adopting or implementing public policies that would have reduced inequality. Corruption also differed little between new and old politicians.

Digging deeper, however, we discovered that there was considerable heterogeneity by initial level of inequality. Indeed, breaking the sample into municipalities with above- and below-median disparities, there were significantly different public policies in relatively equal places, notably higher educational outcomes. Moreover, the same places saw improvements in local state institutions and performance and less corruption in several dimensions. The absence of change, on average, is entirely driven by municipalities with relatively high levels of inequality.

We proposed various theoretical mechanisms that could account for these findings. In particular, we distinguished between incentive and selection effects and found evidence for them. Concerning incentives, in relatively unequal municipalities local elites stand to lose a lot from policy change and can influence policy adoption. The combination of lack of policy change with greater levels of corruption suggests that such elites use their resources to dissuade elected newcomers mayors from choosing redistributive policies. This combination of outcomes is consistent with other mechanisms, however, in particular one where new mayors are tempted by high inequality to act in a predatory way and abandon any progressive policy agenda upon election. We also found that selection likely contributes to these mechanisms. Winning newcomers in relatively unequal municipalities were less likely to be from left-wing parties, thus reducing their propensity to propose redistributive public policy.

Whatever the mechanism that applies, these findings help explain the paper’s central question. Even if there has been a significant increase in democracy in Colombia and the entry of new politicians with different policy preferences, their impact on public policy and, thus, inequality has been muted. This is because of the initially high levels of inequality they have to deal with and the elite interests that this reflects. Though there is a positive change in municipalities with relatively low levels of inequality, this is almost entirely offset by the negative or zero change in highly unequal places. In short, we document the possibility of an inequality trap, whereby high inequality creates incentives that help sustain disparities over time. Our explanation complements other analyses with similar implications. In particular, also discussing persistent inequality in Latin America, Lupu (2024) offers an interesting complementary explanation working through the weakness of political parties: while Latin Americans support redistribution, these preferences do not emerge from legislatures partly because political parties are insufficiently institutionalized. In this case, the trap emerges because high inequality undermines party institutionalization.

We end by noting some limitations of our study that may motivate future work. First, despite being an interesting case, one should not automatically apply our conclusions for the Colombian case to all of Latin America. Nonetheless, our exploration of the underlying theoretical mechanisms is still of general interest for a region with persistent inequality combined with increased democratization and, in particular, the entry of previously excluded political voices. Second, it may take time for newcomers to have an observable impact. In our setting, however, evaluating longer-term impacts is not straightforward. The combination of no re-election and party-level incumbency disadvantage in Colombia (as in most of Latin America; Klašnja and Titiunik (2017)) complicates the interpretation of evaluating any effects over the longer term.

STUDY FUNDING

The authors thank the Latin America and Caribbean Inequality Review (LACIR) and its partner institutions for providing resources to complete this study.

CONFLICT OF INTEREST

The authors declare no conflict of interest.

AUTHORS’ CONTRIBUTIONS

Leopoldo Fergusson (Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Writing); James A. Robinson (Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Writing); Santiago Torres (Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Writing).

DATA AVAILABILITY

Data and code are available upon request. Some of the data are not disclosable, in which case we will provide the interested party with information on how to obtain it from the original source.

Footnotes

1

This literature has focused on several themes. One is whether or not inequality in Latin America was higher in the colonial period. Some studies have used probate data to argue this was not the case, e.g. Johnson and Frank (2006). Consistent with this, others have argued that Latin America became relatively unequal much later, possibly during the wave of globalization after 1850 (see Coatsworth (2008) or Williamson (2010)), or even in the twentieth century (Bleynat et al. 2021).

2

One can, of course, think of purely economic mechanisms via which inequality can persist (Banerjee and Newman 1993; Galor and Zeira 1993). Nevertheless, we believe that the question we pose remains relevant because the advent of democracy ought to make policy more redistributive, which should relax the types of liquidity constraints on which these theories rest.

3

Appendix Figs C1 and C2 show that Latin America is also unique when examining other measures, in particular the increase in the effective number of parties for Congress and the rate of Executive turnovers.

4

Even in Central America, new elites emerged, for example, with the nineteenth-century take-off of the coffee economy and the twentieth-century industrialization of the economy (Dosal 1995; Paige 1998).

5

Undoubtedly, these mechanisms may apply elsewhere since the cross-national evidence does not suggest that, on average, democratization leads to falls in income inequality, see Acemoglu et al. (2015).

6

The first person to identify that democratization in Latin America was not creating a new transformational politics but rather a type of perverse form of democracy, what he called ‘delegative’ (rather than representative) was O’Donnell (1994). See Acemoglu and Robinson (2008b) for a model where democratization can have no effect on the economic equilibrium because elites can offset it.

7

One possible reason for the absence of change is that inequality in Latin America is sociological in ways in which political transition does not impact or does so only very slowly. Many non-political mechanisms may lead to the reproduction of inequality even in politically transformed societies. For example, Fernández et al. (2015) find that Latin American countries have the highest rates of assortative matching in the marriage market so that rich people are more likely to marry other rich people. The research of Bourdieu (1984) emphasized the profound and deeply rooted ways hierarchy reproduces itself, extending even to taste in music and art. His book will resonate with anyone familiar with Latin America. He also pointed out the critical role of schools in reproducing inequality Bourdieu and Passeron (1990), and recent work by Zimmerman (2019) has shown how the school system in Chile interacts with elite status to reproduce inequality powerfully. For Colombia, Fergusson et al. (2021a) discuss the cultural implications of studying in a private or public school and demonstrate in an experiment that the labor market penalizes public school graduates, especially in jobs requiring social interactions where their social networks, cultural capital and social prestige may be valuable. Unfortunately, the application of these ideas in Latin America is in its infancy (see Auyero (2012) for some ideas).

8

There is extensive literature on political clientelism in Latin America. See also Nichter (2018) and Calvo and Murillo (2019).

9

In the Colombian case, a salient example of this might be the Carrusel de la contratación organized by Samuel Moreno after he became mayor of Bogotá in 2010 (see Romero (2013)).

10

Several authors have examined direct financial incentives as a source of political selection in Latin America (e.g. Ferraz and Finan (2009) and Piqué (2019) for Brazil and Perú).

11

Article 7 reads “the state recognizes and protects the ethnic and cultural diversity of the Colombian nation”. Moreover, Article 13 states “ (...) the state will promote the conditions necessary for equality to be real and effective, and will adopt measures in favor of discriminated and marginalized groups”

12

Technically, we conduct a regression discontinuity analysis. The hypothesis is that by focusing on close elections, the fact that a municipality elected a new person rather than an old one is idiosyncratic and unrelated to other municipal features that might impact the adoption or implementation of policies.

13

This reform introduced three main changes: first, political parties would be recognized if they received at least 2% of the total national votes in a national election. Second, political parties must present a single list to legislative bodies (though they could be open or closed, attenuating the reform’s effect). Third, the seat allocation changed from the Hare quota to the D’Hondt method, introducing a minimum vote threshold for receiving a seat.

14

The only type of election we exclude, due to data limitations, is the ‘juntas de acción comunal’ or local action boards, which operate at the neighborhood or rural village level. While these are significant political entities, our definition of newcomer candidates remains pertinent as a measure of individuals who have either not participated in politics or have done so exclusively in this very local arena. Furthermore, for each election with available data since year |$t$|⁠, we incorporate a 10-year ‘buffer.’ For instance, since mayors were appointed rather than democratically elected before 1988, identifying ‘newcomers’ after this period could potentially introduce a bias. This is because many individuals who appear to be newcomers might have been appointed before. Therefore, we incorporate a minimum 10-year buffer before the election to define our measure (the election must occur in or later than |$t+10$|⁠), making |$2000$| the first election meeting this criterion.

15

Appendix Table A1 describes all variables and sources.

16

We exclude extraordinary elections not occurring on the scheduled date.

17

Using nighttime light intensity for a large range of periods requires harmonizing data from the Defense Meteorological Satellite Program (available from 1992 to 2013, with the Visible Infrared Imaging Radiometer Suite, available from 2014 on. We follow the recommendations in Li et al. (2020) to do this.

18

More specifically, we included all variables for which we had complete data for all municipalities between 2005 and 2015. Furthermore, we limit our analyses to a single sample that consists of all observations in which every index component is observed to avoid changes caused by sample composition.

19

We use the median value following convention and, more importantly, since it is best to optimize statistical power within each subsample.

20

While the results are strongly suggestive, we note some limitations of the available data. We cannot conduct a similar exercise for mayoral candidates, where we have no systematic data given the diversity of small movements and scarcity of politician-platform data at the local level. Also, these average patterns undoubtedly mask substantial heterogeneity, and some newcomers may align with old ideas and electoral practices.

21

We note that unobserved confounders may still affect our estimates.

A Appendix—variables and sources

 

Table A1

Variable definition and sources

Variable	Description	Source
Panel A. Dependent variables: public services
Aqueduct coverage	Total aqueduct coverage. Available for years: 2005, 2008–2016. Information retrieved from “Sistema Único de Información de Servicios Públicos -SUI”	CEDE’s municipal panel— Acevedo and Bornacelly (2014).
Sewerage coverage	Total sewerage coverage. Available for years: 2005, 2008–2016. Information retrieved from “Sistema Único de Información de Servicios Públicos -SUI”	CEDE’s municipal panel— Acevedo and Bornacelly (2014).
Internet penetration	Percentage of population that have full internet access, this is they have a permanent connection (365 days each year, 24 hours each day)	Terridata—https://terridata.dnp.gov.co/.
Electricity coverage	Percentage of households that have access to electricity,	Terridata— https://terridata.dnp.gov.co/.
Natural gas coverage	Effective natural gas coverage. Available for years: 2006–2016. Information retrieved from the Ministry of Mines and Energy.	CEDE’s municipal panel—Acevedo and Bornacelly (2014).
Panel B. Dependent variables: education
Basic education coverage	Percentage of 5-year-old students enrolled in an educational institution out of the total population of 5-year-olds.	Terridata—https://terridata.dnp.gov.co/.
Primary education coverage	Percentage of students between the ages of 6 and 10 who are enrolled in an educational institution out of the total population between the ages of 6 and 10.	Terridata— https://terridata.dnp.gov.co/.
Secondary education coverage	Percentage of students between the ages of 11 and 14 who are enrolled in an educational institution out of the total population between the ages of 11 and 14.	Terridata— https://terridata.dnp.gov.co/.
Teacher/student ratio in schools	The teacher/student ratio is calculated by dividing the number of teachers by the number of students enrolled in schools. These variables are available from 1996 to 2020 using the DANE C600 form.	CEDE’s municipal panel—Acevedo and Bornacelly (2014).
National standardized test results	Average score obtained by all students in a given municipality in all components of the national test-“Saber 11”.	CEDE’s municipal panel—Acevedo and Bornacelly (2014).
Panel C. Dependent variables: health
Infant mortality rate	Number of child deaths out of every 1000 registered newborns before the first year of life.	CEDE’s municipal panel—Acevedo and Bornacelly (2014).
Percentage of underweight newborns	Percentage of registered newborns diagnosed with underweight.	Terridata—https://terridata.dnp.gov.co/.
Coverage of the subsidized health regime	Coverage of the subsidized health regime. Retrieved from the Health Ministry.	CEDE’s municipal panel—Acevedo and Bornacelly (2014).
Teenage pregnacy rate	Number of child born per thousand women between the ages of 10 and 19.	Terridata—https://terridata.dnp.gov.co/.
Health facilities per capita	Health facilities per capita is calculated by dividing the number of health establishments by the total population.	CEDE’s municipal panel—Acevedo and Bornacelly (2014).
Panel D. Dependent variables: economic performance
Value added per capita	The value added per capita is calculated by dividing the value added by the total population. The value added is the additional economic that goods and services acquire during the production stage in a given municipality.	Terridata—https://terridata.dnp.gov.co/.
Mean nightlight intensity (urban)	Mean observed luminosity of urban areas in a given municipality.	National Oceanic and Atmospheric Administration (NOAA).
Mean nightlight intensity (rural)	Mean observed luminosity of rural areas in a given municipality.	National Oceanic and Atmospheric Administration (NOAA).
Percentage of the population occupied	The proportion of people who are occupied in relation to the total population.	Terridata—https://terridata.dnp.gov.co/.
Panel E. Dependent variables: security
Number of thefts per capita	The number of thefts per capita is calculated by dividing the total number of thefts by the total population. Information retrieved from the Ministry of Defense.	CEDE’s municipal panel— Acevedo and Bornacelly (2014).
Number of kidnaps per capita	The number of kidnaps per capita is calculated by dividing the total number of kidnaps by the total population.	Terridata—https://terridata.dnp.gov.co/.
Number of murders per capita	The number of murders per capita is calculated by dividing the total number of murders by the total population.	Terridata—https://terridata.dnp.gov.co/.
Number of forcibly displaced per capita	The number of forcibly displaced per capita is calculated by dividing the total number of forcibly displaced people by the total population.	Terridata— https://terridata.dnp.gov.co/.
Panel F. Dependent variables: management performance
Expenditure efficacy index	DANE’s efficacy index. Efficacy is the extent to which a given action yields the desired outcomes. Efficacy involves focusing an entity’s efforts on the procedures and tasks that must be completed in order to accomplish the stated goals.	CEDE’s municipal panel—Acevedo and Bornacelly (2014).
Expenditure efficiency index	DANE’s efficiency index. Efficiency is the capacity of the municipality to increase output in the areas of education, health and drinking water while using the least amount of inputs possible.	CEDE’s municipal panel—Acevedo and Bornacelly (2014).
Fiscal performance index	DANE’s index of fiscal performance. This index, which measures annual fiscal performance globally, ranges from 0 to 100, with values close to 0 reflecting poor fiscal performance and values close to 100 indicating that the territorial entity as a whole achieved the following outcomes: (1) Its financial performance was balanced. (2) Enough resources to keep it operating. (3) Adherence to the Law 617/00 operating expenditure ceilings As a counterpoint to SGP resources, there is a significant level of own resources (tax solvency). (4) Significant investment. (5) Sufficient resources to support its debt service. (6) Creation of current savings, essential to ensuring its financial viability.	CEDE’s municipal panel—Acevedo and Bornacelly (2014).
Management practices index	DANE’s management practices index. This index is understood as the availability of human, physical and technological resources to support the different processes and procedures carried out within the organization. In this sense, administrative capacity is measured through the following five indicators: (1) Stability of management personnel. (2) Professionalization of the staff. (3) Availability of computers (Manager, Advisor, Professional and Technician). (4) Automation of processes. (5) Implementation of the Internal Control Standard Model -MECI.	CEDE’s municipal panel—Acevedo and Bornacelly (2014).
Legal transparency index	DANE’s legal trasparency index. This index evaluates compliance with the regulatory framework provided by Laws 715 of 2001, 1176 of 2007 and 1438 of 2011, and the regulations related to the execution of the resources of the General Participation System (SGP).	CEDE’s municipal panel—Acevedo and Bornacelly (2014).
Panel G. Dependent variables: others
Disciplinary sanctions	A disciplinary sanction is a reprimand issued to someone who breaks a rule, regulation, or instruction. This includes, among other things, disqualification from public office, suspensions, fines, written warnings and imprisonment.	SIRI-Producraduría
Panel H. Forcing variable
Newcomer win margin (normalized around 0)	Winning margin (in %) of the newcomer incumbent, normalized around 0. Values above 0 indicate that the newcomer won (below 0 = newcomer lost).	Colombia’s Electoral Data Archive-Torres et al (2024)
Panel I. Heterogeneity analysis covariates
GINI index	GINI index calculated in the municipality in 1993. Based on DANE’s 1993 National Census, the Encuesta de Calidad de Vida and Consejo Asesor Técnico del Sistema Estadístico Nacional.	CEDE’s municipal panel—Acevedo and Bornacelly (2014).
Panel J. Other predetermined covariates
Initial population	Number of inhabitants in the municipality in 1993, Based on DANE’s 1993 National Census	CEDE’s municipal panel—Acevedo and Bornacelly (2014).
Municipality’s height	Altitude of municipality seat above sea level, in meters.	CEDE’s municipal panel— Acevedo and Bornacelly (2014).
Literacy rate	Percentage of literate in the municipality in 1993. Based on DANE’s 1993 National Census.	CEDE’s municipal panel—Acevedo and Bornacelly (2014).
Distance to department capital, km	Straight line distance to the capital of the department in which the municipality is located.	CEDE’s municipal panel—Acevedo and Bornacelly (2014).
Coca crops 1994	Number of hectares containing coca crops in 1994 per municipality. Reported by Colombia’s National Police	CEDE’s municipal panel—Acevedo and Bornacelly (2014).
Panel K. Candidate-level covariates
Years in public office	Number of years in a democratically appointed public office.	Torres (2024)
Age	Age (in years) at the moment of election.	Torres (2024)
Male	Dummy variable indicating the sex at birth of the candidate.	Torres (2024)
Education	Highest education attained by the candidate.	SISBEN IV
Income	Categorical value indicating the income level of the candidate: 1. Not Poor. 2. Vulnerable. 3. Moderate Poverty. 4. Extreme Poverty.	SISBEN IV
Runs via a traditional party	Dummy variable indicating the candidate runs via a traditional party using the party classification of Cabra-Ruíz (2023).	Cabra-Ruíz (2023)
Party ideology	Categorical variable indicating the ideology of the party via which the candidate runs using the party classification of Cabra-Ruíz (2023).	Cabra-Ruíz (2023)

Table B1

Economic development indices components

Development indices
Component	Source	Component	Source	Component	Source
Panel A: public services coverage index		Panel C: health index		Panel E: suecurity index
|$\diamond $| Aqueduct coverage	CEDE	|$\diamond $| Infant mortality rate	CEDE	|$\diamond $| Number of thefts per capita	CEDE
|$\diamond $| Sewerage coverage	CEDE	|$\diamond $| No. of underweight newborns	Terridata	|$\diamond $| Number of kidnaps per capita	Terridata
|$\diamond $| Internet penetration	Terridata	|$\diamond $| Coverage of the subsidized regime	CEDE	|$\diamond $| Number of murders per capita	Terridata
|$\diamond $| Electricity coverage	Terridata	|$\diamond $| Teenage pregnancy rate	Terridata	|$\diamond $| Number of forcibly displaced per capita	Terridata
|$\diamond $| Natural gas coverage	CEDE	|$\diamond $| Health facilities per capita	CEDE
Panel B: education index		Panel D: economic performance index		Panel F: management practices index
|$\diamond $| Basic education coverage	Terridata	|$\diamond $| Value added per capita	Terridata	|$\diamond $| Expenditure efficacy index	CEDE
|$\diamond $| Primary education coverage	Terridata	|$\diamond $| Mean nightlight intensity (urban)	NOAA	|$\diamond $| Expenditure efficiency index	CEDE
|$\diamond $| Secondary education coverage	Terridata	|$\diamond $| Mean nightlight intensity (rural)	NOAA	|$\diamond $| Fiscal performance index	CEDE
|$\diamond $| Teacher/student ratio in schools	CEDE	|$\diamond $| Percentage of the population occupied	Terridata	|$\diamond $| Management practices index	CEDE
|$\diamond $| National standardized test results	CEDE			|$\diamond $| Legal transparency index	CEDE

Table C1

Balance test: effect of electing a newcomer mayor on municipal characteristics

Dependent variable	Mean	Standard Deviation	Newcomer Victory	Std. Error.	Obs	Bandwidth
	(1)	(2)	(3)	(4)	(5)	(6)
Panel A. Election year
Year elected	2013.43	1.95	-0.34	0.51	314	0.18
Panel B. Geographic characteristics
Average precipitation	93.85	19.15	-8.48	6.15	210	0.11
Distance to department capital, km	70.10	49.48	-13.04	12.38	273	0.15
Distance to main city, km	101.74	68.57	14.95	18.75	306	0.17
Andean region dummy	0.64	0.48	-0.05	0.14	258	0.14
Pacific region dummy	0.07	0.25	-0.07	0.08	315	0.18
Eastern region dummy	0.07	0.25	-0.01	0.07	266	0.14
Caribbean region dummy	0.22	0.42	0.03	0.11	329	0.19
Panel C. Socioeconomic characteristics
Rurality index, 1994	0.56	0.23	0.03	0.06	333	0.19
Initial population (thousands), 1994	28.951	40.78	5.05	13.27	254	0.14
Coca hectares, 1994	0.02	0.13	-0.00	0.01	208	0.11
No. municipal-level employees, 1995	70.55	139.15	-8.70	27.63	301	0.17
No. national-level employees, 1995	740.84	1375.77	-65.45	366.06	314	0.17
No. public agencies, 1995	25.71	27.14	-3.32	6.71	282	0.16
Gini index, 2005	0.45	0.03	-0.01	0.01	288	0.16
Unmet basic needs index, 1995	45.02	17.54	6.64	5.17	299	0.17
Panel D. Fiscal capacity
Total income (per capita)	1.42	0.87	0.07	0.15	253	0.14
Non tax income (per capita)	0.09	0.08	0.01	0.02	290	0.15
Capital income (per capita)	1.01	0.67	0.12	0.11	258	0.14
Panel E. Electoral variables
Number of candidates in election	4.44	1.81	-0.24	0.47	274	0.15
Number of parties in election	4.44	1.80	-0.23	0.47	274	0.15

Notes: Columns 1 and 2 report the basic descriptive statistics of each variable. Column 3 reports RDD point estimates of the effect of a newcomer victory in mayor elections on each variable, using Calonico et al. (2014)’s optimal bandwidths (reported in column 6), bias correction and standard errors clustered at the municipal level (column 4), with linear local polynomials and triangular kernels. Column 5 reports the number of observations included in each estimation.

B Appendix—constructing the indices

We construct indices following Kling et al. (2007) for |$G$| categories, each comprised by |$K_{g}$|⁠, |$g \in \{1,\cdots ,G\}$| indicators. For each unit |$i$|⁠, let |$Y_{i,k,g}$| be the vector containing the |$k$|th outcome of category |$g$| and let |$\bar{Y}_{k,g}$| and |$s_{k,g}$| be the sample mean and standard deviation of the outcome observed in the control group respectively. The (signed) standardized measure for unit |$i$| and outcome |$k$| of category |$g$| can be calculated as |$Y_{i,k,g} ^{*}= \pm (Y_{i,k,g}-\bar{Y}_{k,g})/s_{k,g}$|⁠. The sign is given to preserve interpretation: for instance, if the index measures health system performance, indicators such as the coverage rate and the number of hospitals should be assigned a positive sign, while infant mortality or the prevalence of airborne diseases have a negative sign. The summary index for unit |$i$| corresponding to category |$g$| is given by the average of the standardized measures, namely:

$$ \begin{align*} &\tilde{Y}_{i,g}= \dfrac{1}{K_g} \sum_{k=1}^{K_g} Y_{i,k,g}^*.\end{align*} $$

C Appendix—additional figures and tables

 

Table C2

Balance test: suample split

	Above-median GINI, 1993							Below-median GINI, 1993
Dependent variable	Mean	Standard deviation	Newcomer victory	Std. Error.	Obs	Bandwidth		Mean	Standard Deviation	Newcomer Victory	Std. Error.	Obs	Bandwidth
	(1)	(2)	(3)	(4)	(5)	(6)		(7)	(8)	(9)	(10)	(11)	(12)
Panel A. Election year
Year elected	2013.6	1.9	-1.3*	0.7	149	0.2		2013.2	2.0	1.0	0.9	121	0.1
Panel B. Geographic characteristics
Average precipitation	94.6	16.6	-0.2	6.4	150	0.2		93.0	21.7	-11.9	9.5	112	0.1
Distance to department capital, km	77.3	48.2	-12.0	18.2	150	0.2		61.9	49.8	-8.9	15.3	118	0.1
Distance to main city, km	111.3	59.8	22.2	23.6	129	0.1		90.8	76.1	22.6	36.1	115	0.1
Andean region dummy	0.8	0.4	-0.1	0.1	163	0.2		0.5	0.5	0.0	0.2	142	0.2
Pacific region dummy	0.0	0.2	0.0	0.1	160	0.2		0.1	0.3	-0.1	0.1	121	0.1
Eastern region dummy	0.1	0.3	-0.1	0.1	163	0.2		0.0	0.2	-0.0	0.0	82	0.1
Caribbean region dummy	0.1	0.2	0.1	0.1	156	0.2		0.4	0.5	0.1	0.2	123	0.1
Panel C. Socioeconomic characteristics
Rurality index, 1994	0.6	0.2	0.1	0.1	109	0.1		0.5	0.2	0.0	0.1	115	0.1
Initial population (thousands), 1994	18.5	21.6	1.4	6.4	116	0.1		40.9	52.7	6.0	27.4	112	0.1
Coca hectares, 1994	0.0	0.1	0.0	0.0	143	0.1		0.0	0.1	-0.0	0.0	80	0.1
No. municipal-level employees, 1995	43.0	89.1	3.6	9.6	91	0.1		102.2	175.2	-57.5	56.5	82	0.1
No. national-level employees, 1995	515.9	1071.6	4.3	129.8	94	0.1		998.8	1621.6	-352.9	769.7	108	0.1
No. public agencies, 1995	20.3	15.8	-0.7	3.7	124	0.1		2013.2	2.0	-13.6	14.2	89	0.1
Unmet basic needs index, 1995	44.5	15.6	2.1	5.0	149	0.2		45.6	19.6	12.9	10.2	131	0.1
Panel D. Fiscal capacity
Total income (per capita)	1.6	0.9	0.1	0.2	143	0.1		1.2	0.9	0.0	0.2	132	0.1
Non tax income (per capita)	0.1	0.1	0.0	0.0	126	0.1		0.1	0.0	-0.0	0.0	121	0.1
Capital income (per capita)	1.1	0.7	0.1	0.1	138	0.1		0.9	0.6	0.2	0.2	91	0.1
Panel E. Electoral variables
Number of candidates in election	4.3	1.8	-0.9	0.6	154	0.2		4.6	1.8	0.2	0.7	117	0.1
Number of parties in election	4.3	1.8	-0.9	0.6	160	0.2		4.6	1.8	0.2	0.7	117	0.1

Notes: Balance test after sample splitting. Columns 1, 2, 7 and 8 report the basic descriptive statistics of each variable. Columns 3 and 9 exhibit RDD point estimates of the effect of a newcomer victory in mayor elections on each variable, using Calonico et al. (2014)’s optimal bandwidths (reported in column 6), bias correction, and standard errors clustered at the municipal level (column 4 and 10), with linear local polynomials and triangular kernels. Columns 5 and 11 report the number of observations included in each estimation.

 

Table C3

Incidence of observed selection on outcomes by level of inequality

	(1)	(2)	(3)	(4)	(5)	(6)
	Development	Public services	Education	Health	Economy	Security
	Index	Index	Index	Index	Index	Index
Panel A: entire sample
Original	0.049	-0.083	0.468	-0.225	0.127	0.051
	(0.054)	(0.169)	(0.182)	(0.157)	(0.120)	(0.106)
	[−0.04, 0.14 ]	[−0.36, 0.20]	[ 0.17, 0.77]	[−0.48, 0.03]	[−0.07, 0.32]	[−0.12, 0.23]
Correction on observables	0.067	-0.129	0.572	-0.218	0.210	0.033
	(0.053)	(0.171)	(0.188)	(0.149)	(0.119)	(0.106)
	[−0.02, 0.15 ]	[−0.41, 0.15]	[ 0.26, 0.88]	[−0.46, 0.03]	[ 0.01, 0.41]	[−0.14, 0.21]
Observations	422	422	422	422	422	422
Bandwidth size (Original)	0.128	0.120	0.115	0.114	0.129	0.102
Bandwidth size (Correction)	0.123	0.108	0.103	0.112	0.125	0.096
Observations in Bandwidth (Original)	199	192	184	184	201	160
Observations in Bandwidth (Correction)	196	172	161	180	198	154
Panel B: above-median GINI
Original	0.045	0.055	0.162	-0.205	0.069	0.144
	(0.057)	(0.171)	(0.163)	(0.145)	(0.146)	(0.137)
	[−0.05, 0.14 ]	[−0.23, 0.34]	[−0.11, 0.43]	[−0.44, 0.03]	[−0.17, 0.31]	[−0.08, 0.37]
Correction on observables	0.052	0.025	0.092	-0.224	0.134	0.059
	(0.058)	(0.156)	(0.157)	(0.144)	(0.145)	(0.130)
	[−0.04, 0.15 ]	[−0.23, 0.28]	[−0.17, 0.35]	[−0.46, 0.01]	[−0.11, 0.37]	[−0.15, 0.27]
Observations	228	228	228	228	228	228
Bandwidth size (Original)	0.117	0.111	0.161	0.161	0.162	0.111
Bandwidth size (Correction)	0.108	0.108	0.183	0.165	0.171	0.127
Observations in Bandwidth (Original)	97	91	130	129	130	92
Observations in Bandwidth (Correction)	90	90	139	130	134	106
Panel C: Below-median GINI
Original	0.121	-0.076	0.795	-0.155	0.266	-0.082
	(0.080)	(0.231)	(0.342)	(0.239)	(0.204)	(0.123)
	[−0.01, 0.25 ]	[−0.46, 0.30]	[ 0.23, 1.36]	[−0.55, 0.24]	[−0.07, 0.60]	[−0.28, 0.12]
Correction on observables	0.149	-0.214	0.710	-0.182	0.371	0.040
	(0.078)	(0.216)	(0.341)	(0.234)	(0.184)	(0.147)
	[ 0.02, 0.28 ]	[−0.57, 0.14]	[ 0.15, 1.27]	[−0.57, 0.20]	[ 0.07, 0.67]	[−0.20, 0.28]
Observations	194	194	194	194	194	194
Bandwidth size (Original)	0.131	0.110	0.102	0.132	0.120	0.115
Bandwidth size (Correction)	0.121	0.098	0.112	0.125	0.128	0.094
Observations in Bandwidth (Original)	94	85	75	97	92	91
Observations in Bandwidth (Correction)	93	73	88	93	93	72

Notes: RD estimate correction on individual-level covariates using the method developed in Torres (2023). Each panel displays a different sample. Within each panel, we present the original estimate and the corrected estimate. We report MSE-optimal point estimates, standard errors clustered at the municipal level and inflated to account for recentering (in parenthesis) and 90% bias-recentered confidence intervals (in squared brackets).

Figure C1

Share of electoral turnovers in Latin American and the Caribbean elections: 1946–2018. Notes: An electoral turnover is the defeat of the candidate or party representing the incumbency in an election. Source: Marx et al. (2022). Sample: Presidential and parliamentary elections in 42 LAC countries, 1946–2018.

Figure C2

Evolution of political competition in lower chambers in Latin America and the Caribbean. Notes: Evolution of the effective number of parties competing in lower chambers for various regions. The region composition varies with data availability. The effective number of parties was introduced by Laakso and Taagepera (1979), representing a fragmentation measure of a political system between its political parties. For instance, an effective number of parties of 2.2 indicates more than two but less than three major parties competing in elections. Source: The Constituency-Level Data Archive (CLEA)- Kollman (2019).

Figure C3

Close elections newcomer winners—attribute unbalance. Notes: RD plots for observable candidate-level attributes on newcomers’ relative vote margin. We report the MSE-optimal point estimate using a local linear polynomial and a triangular kernel following Calonico et al. (2014b). We also report standard errors clustered at the municipal level and inflated to account for confidence interval recentering

Figure C4

No-manipulation tests. Notes: (A) presents manipulation test proposed by Cattaneo et al. (2020). Bias-corrected RD estimate of the density discontinuity reported with robust standard errors. (B) presents a manipulation test proposed by McCrary (2008). Discontinuity statistics are reported together with estimated standard errors

Figure C5

No manipulation tests—sample split. Notes: (A) and (B) present the manipulation test proposed by Cattaneo et al. (2020). (C) and (D) present a manipulation test proposed by McCrary (2008)