He Who Counts Elects: Determinants of Fraud in the 1922 Colombian Presidential

Electiony

Isaías Chavesz Leopoldo Fergussonx James A. Robinson{ December 14, 2008

Abstract

This paper constructs measures of the extent of ballot stu¢ ng (fraudulent votes) and electoral coercion at the municipal level using data from Colombia’s 1922 Pres- idential elections. Our main …ndings are that the presence of the state reduced the extent of fraud, but that of the clergy, which was closely imbricated in partisan politics, increased it. We also show that landed elites to some extent substituted for the absence of the state and managed to reduce the extent of fraud where they were strong. At the same time, in places which were completely out of the sphere of the state, and thus partisan politics, both ballot stu¢ ng and coercion were relatively low. Thus the relationship between state presence and fraud is not monotonic.

Colombian political aphorism: In Spanish “el que escruta elige”(Bushnell, 1993, p. 125). yWe are particularly grateful to Eduardo Posada-Carbó for telling us about the data on the 1922 election in the National Archive in Bogotá. We also thank Daron Acemoglu, Maria Angélica Bautista, Camilo García, Maria Alejandra Palacio, Olga L. Romero and Daniel Ziblatt for their comments and help with the data. All translations from Spanish texts are our own. zHarvard University, IQSS, 1737 Cambridge Street, Cambridge MA02138. E-mail: [email protected] xMassachusetts Institute of Technology, Department of Economics, 50 memorial Drive, Cambridge MA, E-mail: [email protected]. {Harvard University, Department of Government and IQSS, 1737 Cambridge Street, Cambridge MA02138. E-mail: [email protected]

1 “Lo que han ganado las armas no se puede perder con papelitos”

Colombian Political Proverb1

1 Introduction

The preponderance of the literature on democracy in political science has focused on the origins and timing of the introduction of universal su¤rage (e.g., Rueschemeyer, Stephens and Stephens, 1992, Collier, 1999, Acemoglu and Robinson, 2006). While this approach is surely justi…ed in many cases, it also leaves aside many puzzles. For instance, Argentina had universal male su¤rage after the promulgation of the 1853 constitution, as did Mexico after its 1857 constitution, but neither country is typically coded as a democracy in the 19th century. In fact, the typical date for the introduction of democracy in Argentina is the passing of the Saenz Peña Law in 1914. This law did not change su¤rage however, its main aim was to eliminate electoral corruption and fraud, things which had previously negated the e¤ects of universal male su¤rage. This example, and others such as the introduction of the secret ballot in Chile in 1957 (Baland and Robinson, 2008), suggest that the consequences of variation in electoral fraud are possibly as large as that of the variation of the formal institutions of democracy. Despite the importance of electoral fraud and corruption it has been little studied, pos- sibly because it is so di¢ cult to measure and quantify. In consequence we have little idea about what causes variation in the extent or incidence of such fraud. Moreover, there are many ways to undermine the true outcome of elections ranging from vote buying through disenfranchising potential voters, to using coercion to keep voters away, coercion to force voters to vote in particular ways, or creating fraudulent ballots (‘ballot stu¢ ng’) in favor of one candidate. This lack of systematic evidence makes it di¢ cult to understand the determinants and implications of fraud. In this paper we use two unique data sources from the 1922 Colombian presidential election to examine electoral fraud. First, we are able to construct a (necessarily imperfect) measure of the extent of fraudulent voting or ‘ballot stu¢ ng’at the municipal election. We do this by combining data collected on the vote totals reported by mayors to the central government with estimates of the maximum potential franchise from the 1918 population census. This gives us at least a lower bound on the extent of ballot stu¢ ng. For 483 out of

1 “What you’ve won with arms cannot be taken away by a few slips of paper.”

2 the total 713 municipalities of Colombia for which we have data we …nd the reported vote totals to be larger than the maximum number of people who could possibly have voted. In such municipalities there was obvious ballot stu¢ ng and this was consequential. Indeed, according to this methodology the total number of stu¤ed ballots was 222,202 which was larger than the winning margin of 193,228 by which the Conservative candidate Pedro Nel Ospina defeated the Liberal loser Benjamín Herrera. Table 1 shows some of the basic data from this exercise by Colombian department. One can see here that there is a lot of variation. For instance in Antioquia, traditionally a bastion of the Conservative party, the total number of votes cast was 76,420 of which we calculate 11,648 were fraudulent. On the other hand, in the Liberal stronghold of Santander of the 55,492 votes ‘cast’almost 24,000 were fake, a far greater proportion. Second, we use various sources of information, particularly the proceedings of a conference held in the Colombian city of Ibagué after the election, to code a variable on incidents of electoral violence (Blanco, Solano and Rodríguez, 1922). This conference, held by the Liberal party in the wake of an election in which they were defrauded of the presidency, contained numerous accounts of both ballot stu¢ ng, fraud and coercion. We then study some of the potential sources of variation in ballot stu¢ ng and coercion across municipalities. Though we are not able to make strong claims about causality, we do uncover some very robust conditional correlations. First, for the whole of Colombia, we …nd that the presence of the state is negatively correlated with the extent of ballot stu¢ ng. In municipalities where there was a greater presence of the army or more government bureaucrats, ballot stu¢ ng was less. In addition, the presence of Catholic priests is also negatively correlated with ballot stu¢ ng. With respect to coercion, we …nd that the presence of priests is positively correlated with coercion. Second, focusing on the department of Cundinamarca for which we have much richer data, in addition to the above results, we …nd that ballot stu¢ ng is negatively correlated with land inequality and the extent to which local politics is monopolized by individuals (which we refer to as ‘political concentration’), but positively correlated with the extent to which large landowners and the political class were fused (a variable we call ‘overlap’). We also …nd that political concentration is signi…cantly negatively correlated with electoral coercion. We believe that there is a great deal of consistency between these …ndings and o¤er the following simple interpretations. First, they suggest that the Colombian state in 1922 was not partisan, but in fact rather ‘Weberian’and attempted to control electoral fraud when it

3 could. Unfortunately, however, its reach was limited. Second, priests played an important role in coercion of Liberal voters and this coercion was aimed primarily in stopping them voting. When Liberals were disenfranchised, the measured extent of ballot stu¢ ng in our data will be lower and this is why the number of priests in a municipality is positively correlated with the presence of coercion but negatively correlated with the extent of ballot stu¢ ng. This result is consistent with the conjecture of Posada-Carbó (1995, pp. 16-17) that outright violence was complementary to ballot stu¢ ng. Third, fraud was disliked not just by state o¢ cials, but also by traditional landed elites. Landed elites opposed the chaos and disorder that historical sources suggest accompanied and facilitated ballot stu¢ ng and stopped it when they could. This is why within Cundinamarca we …nd less ballot stu¢ ng where land inequality is high since these are precisely the core areas of the department with a strong consolidated elite. Fourth, not all landed elites were traditional. Some had entered this class through the use of political power and thus secured their property rights via politics. In these places, where overlap was high, ballot stu¢ ng was used as a method of sustaining the political in‡uence of such elites. Finally, areas where political concentration was high featured less ballot stu¢ ng and coercion because such concentration was a feature of areas with caudillos who were to a large extent outside the control and circuit of Conservative politicians. In these ‘independent republics’there was no major contribution to the incumbent (Conservative) cause one way or another. Though these …ndings about the political role of the Catholic Church may be surprising to some, they are in line with a rich historical literature on Colombia which has emphasized the connections between the Church and the Conservative party. For instance Deas observes “the church was the electoral arm of the Conservatives. Liberalism was a sin: Colombian sermons were intense and obstinate on this point” (Deas, 1993, p. 219). While Posada- Carbó has also noted the role of clergy in perpetrating fraud: in 1933, in Cocuy, Boyacá, “a group of 400 conservatives from a neighboring municipality, lead by priest Nepomuceno Goyeneche, invaded Cocuy with the intention of grabbing the ballot boxes”(Posada-Carbó, 1995, p. 10). Our results in Colombia contrast with and complement the small existing literature on electoral fraud. Most related is the research of Lehoucq and Molina (2002) who studied of electoral corruption in Costa Rica. These authors study over a thousand legal accusations of ballot rigging between 1901 and 1946, examining their intensity and spatial distribution.

4 They …nd that fraud accusations were more prevalent in the three poorest and least populated provinces of the country, and argue that this was caused by the socioeconomic backwardness of these regions where social di¤erentiation was more pronounced and it was harder to protect civil liberties than in the urban, literate centers, as well as by the incentives faced by politicians because of the di¤ering structures of electoral competition. Also closely related is the work of Ziblatt (2008) using data on complaints of electoral misconduct from pre-1914 Germany. His main hypothesis, for which he …nds strong support, is that electoral fraud was greater in areas with high land concentration since here strong elites controlled politics and thwarted the desire of the central state to implement fair elections. Finally, Baland and Robinson (2008) analyzed the e¤ects of changing electoral institutions on voting patterns and resource allocation organized by examining the introduction of the secret ballot in Chile in 1958. Their most important …nding for the current paper is that traditional landed elites coerced workers into voting for conservative parties prior to 1958.2 The …ndings of our paper are somewhat di¤erent from this literature. First, we try to investigate directly if the presence of the central state reduces fraud, which appears not to have been tested before. Second, unlike Ziblatt or Baland and Robinson, we …nd that higher land inequality is correlated with less electoral fraud, at least in the sense of ballot stu¢ ng. We believe that the di¤erence between late 19th century Germany or 1950s Chile lies in the relative strength of the state. For instance Ziblatt interprets his results as showing that landed elites blocked the spread of state power, and perpetuated fraud. In Colombia, however, the reach of the state was so weak that landed elites rather substituted for its presence. Since, unlike Prussian Junkers, they could not rely on basic things such as social order, it was not in the interests of landed elites to encourage the anarchy that went along with electoral fraud because it threatened their interests. Third, unlike Lehoucq and Molina, we do not …nd a lot of evidence that ‘modernization’reduced fraud. Though ballot stu¢ ng is negatively correlated with proxies for income we …nd that electoral coercion is positively correlated with measures of human capital. This …nding is in line with recent work on the empirical problems of simple modernization ideas (Acemoglu, Johnson, Robinson and Yared, 2007, 2008).3

2 A important but not very related paper on electoral fraud is Cox and Kousser’s(1981) analysis of election practices in New York State from 1879 to 1908 focuses on how political managers changed their strategies with the introduction of the secret ballot in 1890: politicos went from buying votes to paying voters to abstain from voting. There is also a rich case study literature on electoral fraud in the United States, see Bensel (2004). See also Posada-Carbó (2000) on Latin America and Lehoucq (2003) for a conceptual overview. 3 Unfortunately it is not possible to investigate in Colombia several of the issues which the literature raises. For instance, it is impossible to collect meaningful data on either turnout or political competition during

5 Our approach also has the advantage of actual data on the extent of fraud as opposed to complaints about fraud. Indeed, comparing our data to claims of fraud made at the time of the election shows that the complaints do not accurately re‡ect real fraud. It is quite likely that fraud accusations were used strategically by both parties and Posada-Carbó (2000, p. 636) notes “The denunciation of fraud (...) was also a tool commonly employed in …ghting an election”. Hence, not all accusations are faithful images of what occurred during elections. In addition to this political economy literature our paper makes a contribution to the historical literature on Colombian politics. Posada-Carbó (1997, p. 246, 248) notes “The Conservative Hegemony, particularly the last two decades of its rule, remains one of the relatively less studied periods in Colombia’spolitical history (....) To what extent and under what circumstances the conservative regime was able to manipulate the electoral process is a question that merits further consideration”. The empirical work in our paper throws new light on precisely this issue. Our paper also builds on the work of the political economy of Colombia by Acemoglu, Bautista, Querubín and Robinson (2008). We borrow heavily from their data construction for the case of Cundinamarca and several of our …ndings are very consistent with theirs. In particular we …nd that the presence of high land inequality tends to be associated with good outcomes and our interpretation of this is inspired by theirs.

2 A historical perspective: main antecedents and con- text4

Colombia is a particularly interesting place to study electoral fraud because of its long history of ‘democratic’elections. In the words of Deas (1993) “The periods of authoritarianism or militarism have been very scarce and very short in the one hundred and forty years of Colombia’s existence as an independent state. There have been numerous constitutional experiments, and this republic has had more elections, under more systems, central and federal, direct and indirect, hegemonic and proportional, and with more consequences than any American or European country that could attempt to dispute the title”(p. 207). This Leyenda Rosada of Colombian democracy must be quali…ed, however. Indeed, if elections

this period since elections were either very fraudulent, or were uncontested. Wilkinson (2004) for instance, …nds that electoral violence is more likely in close elections in India. Since we have no way to know if an election is close, we cannot investigate this claim with our data. Moreover, since we are examining data for a national election, it is not clear if these ideas are relevant in our setting. 4 This section draws mainly from Bushnell (1993), Mazzuca and Robinson (2009), Melo (1995) and Posada- Carbó (1997).

6 have been traditional in Colombia, fraud has been an electoral tradition. As will be seen in this paper, during the early twentieth Century, electoral fraud was conspicuous. In the 1970 elections, former dictator Gustavo Rojas Pinilla lost to conserva- tive Misael Pastrana Borrero. Rojas had ruled the country from 1953 to 1957 in the only intermission of democracy in Colombia since 1910.5 His attempt to become President as a member of a third party, the ANAPO, was frustrated in the context of serious suspicions of fraud.6 In the 1982 presidential elections, conservative Belisario Betancur won o¢ ce, surprising liberal Alfonso López Michelsen by obtaining a large share of votes in several de- partments of the Costa, traditional liberal bastions. Later accusations indicated that Pablo Escobar Gaviria, the coca ma…oso par excellence, bought votes in the Costa in retaliation against López, who had refused an invitation for lunch to his renown Hacienda Nápoles.7 The period that we study was in the midst of a long period of domination by the Con- servative party which began in 1886, known as La Regeneración (‘The Regeneration’) or the Hegemonía Conservadora (‘Conservative Hegemony’). Following a devastating civil war between 1899 and 1902 the regime had been expanded to give the Liberal party represen- tation in Congress through the adoption of an electoral system known as the ‘incomplete vote.’ This system, introduced after 1905, created three member electoral districts where two legislative seats were given to the party with the most votes and the loser getting the third seat. Party elites thought of this as a form of proportional representation that was fraud-proof. To buy peace Conservative elites wished to share some power with Liberals, but they realized that proportional representation could not guarantee this because they could not stop local Conservatives committing fraud against the Liberals (Mazzuca and Robinson, 2009). Deas (1993, p. 213) notes “A conservative governor admitted in 1854 that though these [caciques] were ‘friends’he could have no control over them”and Reyes (1978) concurs, and argues that, in the early twentieth Century, "it was still hard for the Central government to confront a regional cacique" (p. 118). As Melo (1995) puts it, the reforms adopted during this period e¤ectively limited the absolute hegemony by a single party, yet “did not impede a signi…cant degree of electoral manipulation by the party controlling the Executive or with majority in Congress; the determination of electoral districts, the elaboration of electoral

5 Even this interrumption was sui generis by Latin American standards. Prominent leaders from the two traditional parties actually supported the coup mounted by Rojas on the government of authoritarian Laureano Gomez, and ultimately coalesced to remove Rojas from o¢ ce when he attempted to cling to power (see Payne, 1968, Chapter 7 and Robinson, 2005). 6 The episode is often cited, among other things, because the guerrilla group M-19 actually borrowed its name from this famous episode (M-19 is short for Movimiento 19 de abril de 1970 ). 7 See the recent documentary Los Archivos Secretos de Pablo Escobar.

7 lists and the control of voters favored the continuation of the well-established tradition of fraud in Colombian political habits.”8

2.1 The importance of the 1922 elections

The main reason for focusing on the 1922 election is that it was unique in this period for featuring an open contest between the Liberals and Conservatives. Prior to 1914 the election of the president was indirect. In 1914, the …rst direct elections of President since 1857 “took place with ‘an entire absence of party strife and feeling”’(Posada-Carbó, 1997, p. 261). For the 1918 elections, the Liberals, led by Benjamín Herrera, decided to “try again the old tactic of supporting a Conservative candidate, to promote divisions within the ruling party, which seemed impossible to beat in open confrontation” (Melo, 1995). Finally, the 1926 elections “could be described as the ‘private act of a few public employees”’(Posada-Carbó, 1997, p.260). This leaves the 1922 elections to examine. Fortunately, these elections were “heavily contested [and] attracted the largest proportion of voters”(Posada-Carbó, 1997, p. 260). At the same time, as we will see below, there were serious denounces of widespread fraud, prevalent enough for it to be observable. The 1922 elections also occurred at a moment of great economic importance, in a period of major changes in Colombian society.9 The economy experienced a period of unprece- dented prosperity. The 1920s were characterized by the rise of the co¤ee economy and an increase in other exports such as petroleum and bananas, rapid urbanization accompanied by the birth of the …rst ‘socialist’movements, and capital in‡ow from increased foreign invest- ment and borrowing. Also, with Law 56 of 1921 Colombia …nally approved the controversial Urrutia-Thompson treaty with the United States, and received $25 million dollars as com-

8 The problem is an old one. See Guerra (1922) quoting the Gaceta O…cial of 1860, when politicians in the opposition complaint: "Consejos Electorales elected by the majoritarian party in Congress and control of the Federal Executive Administration! Juntas de Círculo elected by the majoritarian forces in those Consejos! Jurados de Distrito elected by majorities in those Juntas! Jurados de Votación elected by majorities in those Jurados de Distrito! All the di¤erent electoral functions in the hands of a single party! Nothing less is this part of the Law! The party without control of the Legislative and Executive Federal powers has no other option than resignation, and no other destiny than inde…nite postponement" (p. 597). 9 The socioeconomic transformation of Colombia in the 1920’s has received considerable attention (see, for instance, the references to the several economic histories that have been written for the period in Posada- Carbó, 1997, p. 254). See also Melo (1995), Tirado Mejía (1995), Arrubla (1995) and Bushnell (1993, Chapter 7) for overviews of the period.

8 pensation from the loss of Panamá.10 These developments probably increased the value of controlling the executive in the 1920’s. The Liberals were clearly aware of the money from Panamá, as suggested by their book on the Convención de Ibagué (to which we will refer below). In contemptuous tone, they ask Conservative politicians (to whom they refer to as “big generals who know nothing but destruction”): “What clue will these soldiers without …nancial preparation have of the calm and skillful way in which the twenty …ve millions from Panamá should be handled?” (Solano and Paz, 1922, p. 25). Deas (1993) suggests that economic progress in the 1920s was indeed a strong shock to the political system: “Before the 1920s, when the good prices of co¤ee, petroleum and plantains, the twenty-million-dollar indemnity of Panamá, and big loans from abroad altered the equilibrium, the nexus between the central, departmental and municipal governments in times of general poverty is not too strong”(p. 28). Instead, after the twenties, “the local power truly had more resources at its disposal, congressional and departmental resources looked more formidable in Conservative hands than in the skeletal apparatus of the federal era (...) there was a large increase in national, departmental, and urban clientelism after the indemnity from Panamá and the new and substantial public loans of the twenties”(p. 220). The changes of the 1920s also strengthened the Liberal Party, as Liberal leaders con- sciously targeted urban masses and the incipient labor class. By 1922, as we will see below, Liberals were convinced that their party had good chances of gaining o¢ ce with fair elec- tions and fraud accusations were widely publicized in the liberal press.11 In sum, both for the characteristics of the electoral process and of the socioeconomic context, the 1922 presidential elections are particularly useful to study electoral fraud during the Hegemonía Conservadora. 10 The approval of the treaty was pushed forward by President Marco Fidel Suárez, who was convinced of the importance of restablishing good relations with the United States, a posture which came to be known as the “Polar Star”doctrine. Scandals by the end of his period precipitated his resignation to facilitate the approval of the treaty. See an account of this episode in Navarro (1935). 11 Just a bit over a decade earlier, Liberals were convinced that they had little chance of winning a popular election. Indeed, as Navarro (1935) points out, in the 1910 Constitutional Assembly Liberals had negotiated that the President would be elected by Congress. However, in the end direct elections for the President were established, because “Conservatives pulled back and proposed the reconsideration and elimination of the respective article”. Debating over this, “dr. Nicolás Esguerra defended election by Congress and dr. José Vicente Concha election by the people. That is: the conservative leader defended the liberal principle and the liberal leader the conservative”(p. 11). A few years later, Liberals were more con…dent of their popular support. For instance, Representative García Vásquez, debating over electoral reform in 1915, exclaimed “I believe, and I could also prove this, that the liberal party is majoritarian in the Nation”(ACR, 1915: 854)

9 2.2 Electoral Legislation during the Hegemonía Conservadora

Between the adoption of the incomplete vote in 1905 and 1922 there were continuing e¤orts to control electoral fraud. Debates in Congress then concentrated on the problem of creating “new legal measures aimed at constituting an electoral legislation that guaranteed the liberty and purity of su¤rage” (Montoya, 1935, p. 31). Mazzuca and Robinson (2009, p. 27) observe “The key perpetrators of electoral fraud were local politicians, known as caciques and gamonales, in relation to whom local politicians had little or only intermittent control (...) top leaders in the Conservative Party lacked the political resources needed to provide a formal solution to fraud that Liberals could …nd realistic ... The comparative advantage of the incomplete vote over proportional representation as a mechanism of power sharing in a context of moderate to high electoral corruption is what led Conservatives and Liberals”to resort to the incomplete vote. Of course, this solution was useless in presidential elections, where power can simply not be divided in thirds. This, together with the increased value of controlling the executive in the booming twenties, probably explains why there was so much concern regarding fraud during presidential elections, especially in 1922, when the Liberal Party plausibly started thinking that, with fair elections, they could actually win.12 This is also consistent with the increasing interest shown by Liberalism, after 1920, of introducing proportional representation plus measures against fraud, as analyzed in Mazzuca and Robinson (2009). Indeed, Montoya (1935), commenting on Article 12 of Law 96 of 1920, which rati…ed the principle of minority representation from 1910 (Article 45 of the Acto Legislativo number 3) states: “there was a growing disappointment among the Liberal party because the Law did not guarantee or protect the right proportionality in the representation, as it arbitrarily predetermined the parties’aspirations. One must speak frankly: in spite of its primitive usefulness, the system was an expired one”(p. 26).

12 It is also important to note that, under the 1886 Constitution, the President was fairly powerful. A telling anecdote on this respect occured in 1921. General Holguín, designated by Congress to replace Suarez and facilitate the approval of the Urrutia-Thompson treaty, invited liberals to participate in the cabinet. General Herrera refused but, according to Navarro (1935), when General Holguín was asked what he would have done if General Herrera le coje la caña (accepts his proposal), he answered: “Who cares if all ministers are liberals if I am the President and the Constitution is the 1886 Constitution?”(p. 37).

10 2.2.1 Laws and Main Reforms, 1888-191613

Following the basic framework provided by the 1886 Constitution, Law 7 of 1888 attempted to draft a comprehensive Electoral Code to organize electoral institutions. Though its scope was more limited than that of the Electoral Code to be adopted in 1916, Law 7 established the main electoral institutions and their functions. The basic structure included, in hierar- chical order, the Gran Consejo Electoral, the Consejos Electorales, the Juntas de Distrito Electoral, and the Jurados Electorales.14 This organization was modi…ed on several occasions during the Hegemonía (See Registraduría, 1991, Chapter 1), and some of these changes were probably aimed at controlling fraudulent practices by these organizations and to increase the in‡uence of the opposition in electoral institutions. In 1892, for instance, the Consejo Electoral substituted the President in electing two of the members of the Gran Consejo Elec- toral, thus reducing the Executive’s direct in‡uence over the ultimate electoral authority. In 1910, the election of the members of the Gran Consejo was changed again.15 Its nine members had to be elected by the House and the Senate, and the electoral formula was that of the incomplete vote, thus permitting the representation of minorities in the electoral organization. Legislative activity between the 1890s and 1916 reveals the ongoing concern of politicians to control fraud (see Montoya, 1935). The content of the numerous reform proposals shows that irregularities in the making of voting lists, vote buying, the strategic allocation of

13 This section draws mainly from Montoya (1938) and Registraduría Nacional del Estado Civil (1991). 14 The Gran Consejo Electoral declared the election of the President and Vice-President after counting votes from electores in each Department’s Asamblea Electoral. Its six members were elected, in equal shares, by the Senate, the House of Representatives, and the President. The Consejo Electoral in each Department was also composed of six members elected by the Senate, the House and the President, and it elected members of the Juntas de Distrito Electoral. These Juntas counted votes and declared the election of Representatives and Diputados, as well as of electores, based on the registries provided by the Jurados Electorales of the municipalities. The Jurado Electoral, one per each electoral district, was elected by the Junta de Distrito Electoral. It formed the lists of voters, elected the jurados de votación to be allocated at each voting table, and counted votes and declared the election of Consejeros Municipales. Note the resemblance between this structure and the one Conservatives complained about in the 1860’s (footnote 12). 15 Recall that in this year the Constitutional Assembly rati…ed the principle of minority representation. More precisely, it replaced the principle of representation of minorities with representation of parties, to avoid Conservatives running on di¤erent lists and efectively obtaining all seats in a given district, practice known as arrastre or listas corsarias. Montoya (1991) explains that despite this and subsequent legislation (Law 85 of 1916, Article 181) this type of violation of the spirit of the law persevered for years. Mazzuca and Robinson (2005, p. 28) note how this was banned again in Law 13 of 1920. Crucially, this reform reinstituted direct election of the President. Representatives, Diputados, and members of Local Councils were also elected directly, whereas indirect election was mantained only for Senators, through Departmental Assemblies.

11 voting tables, double-voting, and participation of the armed forces in elections were among the elements that, in the views of politicians, corrupted elections. Several of these elements re-appeared during the adoption of the 1916 Electoral Code which prevailed, with some reforms, in the 1922 presidential elections. The 1916 Electoral Code emerged of the numerous reforms e¤orts that preceded it. Apart from formally establishing in the law the principle of minority representation rati…ed in 1910 by the Constitutional Assembly, this code’s main objective was that of strengthening “controls over the electoral processes in order to reduce fraud and manipulation by local politicians”(Mazzuca and Robinson, 2009, p. 19). In 1915, two reform projects were being discussed in Congress. One, presented by Liberals to the House of Representatives was defended by Senators Luis de Grei¤ and Fabio Lozano. The second project, which was ultimately adopted with a number of modi…cations as Law 85 of December 31 of 1916, was a countero¤er by the government to the Liberal demands presented on August of 1915 by Bonifacio Vélez, member of the National State Council, to the Senate. It is worth revising some of the animated debates over each of these projects, which revolved mostly around fraud control.16 The Liberal project received attention in 1915. It was approved by the Senate in its initial debate and sent to the House of Representatives. However, Conservatives in the House of Representatives soon argued that the Liberal reform project was “retrogressive, absurd, and impracticable” (ACR, 1915: 853), that even among Liberals there was no consensus around. Controversy, not surprisingly, revolved around measures related with fraud. For instance, Conservative representative Jaramillo Isaza criticized the reform project’sproposal of reviving the Jueces de Escrutinio and assigning them some of the control functions that the Tribunales de lo Contencioso Administrativo had at the time, as well as the idea of creating Census Boards elected by Local Councils that would form the voting lists and, crucially, institute the cédulas a sort of electoral “title”for citizens that were eligible to vote. Although Liberal senator Luis De Grei¤ had defended the role of Local Councils in the expedition of cédulas, as their “independent character and the way they are organized are reasons to trust the impartiality and correctness with which they would happily ful…ll this service”(AS, 1915: 475), Jaramillo Isaza contended that it was inadvisable to taint Local Councils with this political role. In what would become a recurrent argument of Conservatives, he attributed the prevalence of fraud to backwardness in the electoral “culture”of the population, and not

16 The electoral system for minority representation was also a primal issue, with the incomplete vote system prevailing over the quotient rule in the end. This is analyzed in Mazzuca and Robinson (2009).

12 to ‡aws in the legislation: “we must give civic education to the masses, the population must be instructed to vote freely and honestly; if we achieve this, and that the individuals that must enforce the law are guided always by a correct and truthful criteria, everything will be all right and fraud and scams will be gone”(ACR, 1915: 854).17 The Liberal project was abandoned in November 1915 but e¤orts to reform the electoral legislation continued in 1916. In August, both the Liberal and Velez’sprojects were studied by the electoral commission in the Senate. The debates that followed were animated, and the lengthy code was discussed in meticulous detail. Articles aimed at controlling the manip- ulation of electoral lists were especially important in the debate. Such is the case of article 19 of the project (ultimately appearing as article 18 in the Law) which was discussed over several sessions (see AS, 1915: 419, 449, 482). This article allowed any citizen to ask for the exclusion of individuals from the voting lists, as long as he presented two pertinent witnesses that backed his point. Citizens that were arbitrarily excluded from the list could ask for their inclusion in a similar way. Several modi…cations to the article were proposed, which in its …nal form demanded individuals to present, aside from the two witnesses, su¢ cient reasons (under oath) backing their demands. It also obliged electoral juries to publish the modi…- cations to the voting lists, so that anyone could present counterarguments to the changes made and prosecute others over perjury charges. Article 28 of the project, …nally appearing as article 99 in the Law, was also discussed in detail. This article forced electoral juries to provide, upon the citizen’spetition, a document certi…cating their inclusion in the electoral census. This certi…cate would be valid for voting even if the citizen’sname was excluded from the voting lists on election day. Some senators, like Dávila Florez, equated this certi…cate with the cédulas and argued that they would facilitate fraud (AS, 1915: 571-572), possibly fearing that electoral juries would confer the certi…cates arbitrarily or that they would be deceptively used by citizens. Other senators, such as senator Marcelino Arango, defended the article arguing that mistakes were often committed in the transcription of lists, and referred to an episode in which he had been mistakenly included in the list as “Marcelino Aranguren”. After some discussion, the issue was passed to a commission, and the …nal article, while approved, included a number of measures that sought to control corruption.18

17 As shown above, this same argument was put forward to criticize the implementation of the electoral quotient instead of the incomplete vote (See AS, 1915: 474-475 and Mazzuca and Robinson 2009, p. 20). Similar ideas were also present when discussing the implementation of mandatory voting, included in the liberal project, in the Senate (AS: 1915: 476). 18 More speci…cally, the electoral jury should be convinced of the identity of the voter. Also, the certi…cate and a copy would be given to the voter. Upon voting, the copy would be returned to the electoral jury, to facilitate later revision of its validity.

13 When the project was sent back to the House of Representatives for discussion, the elec- toral commission did not present a uni…ed report. While senators Julián Restrepo Hernandez and Arcadio Chari defended the project, albeit suggesting some modi…cations, senator Car- los Rosales o¤ered a separate report in which he did not endorse the modi…cations suggested by the majority and regretted that the Senate’scommission had "blended in a single project the separate ones sent by the State Council and by several members of the Liberal minority". Fearing that the Conservative government would set voting tables in Liberal municipalities in distant places, senator Rosales also criticized the fact that electoral tribunals were allowed to establish voting places in “veredas, without even de…ning what the term means, in itself vague and capricious, thus leaving the door open for all sorts of frauds.”Notwithstanding his disagreements, Rosales consented to open the debate of the proposed reform because there was little time left to improve the project substantially and its approval was in any case preferred to the prevailing legislation. His feeling that the project was still unsatisfactory in terms of fraud control is best illustrated by his …nal words in his report, urging future legislators to improve the electoral law in the direction “that public opinion has persistently asked for.” With only a few days left for the discussion and approval of the project before the termi- nation of the session, the House of Representatives continued the discussion. In spite of the limited time, the Conservative majority approved a number of modi…cations to the project, many of which were criticized by Liberals for weakening the controls against fraud. On December …rst, Liberal politicians proposed to vote for the approval of the entire project, omitting the changes introduced by the House. However, the majority rejected the proposi- tion and the discussion continued. The next day, the president sent a message to the House of Representative insisting on the importance of approving the project promptly (AS, 1916: 963). Some days later, with discussions over articles continuing, a group of Liberal senators insisted again on voting for the approval of the entire project as it was sent by the Senate. They argued that Conservatives were pretending to obstruct the project, “with new articles, modi…cations and reconsiderations of the ones already approved”(AS, 1916: 980). Finally, the reform project with the modi…cations from the House of Representatives was sent back to the Senate for the last debate and approval. Over the course of this …nal debate, the Conservative and Liberal positions echoed the ones heard in the House of Representatives. While Conservative politicians urged for the approval of the project as it was sent by the House, the Liberal minority made a …nal attempt to approve the code but eliminating some of the modi…cations that had been introduced to the original project. Liberal senator Lozano

14 defended this last position by criticizing some of the modi…cations that muted the controls against electoral corruption.19 He anticipated that

“the outrageous scandal of the prodigious multiplication of Conservative votes to drown the Liberal majorities in the most important centers of the country will continue. In election time we will still have what specialists call chocorazos in Magdalena; canastadas in Boyacá and Cundinamarca; milagros de Santa Isabel in Tolima”(AS, 1917: 1117).

The view that prevailed, however, was that of “not sacri…cing the good for the perfect,” and Law 85 of 1916 was approved in December 31, without meeting Liberal expectations regarding measures against fraud. In spite of its de…ciencies, Law 85 included several dispositions aimed at reducing electoral corruption. Apart from the regulation on the expedition of certi…cates and publication of voting lists, article 179 declared null the elections in which the number of voters exceeded the number of those inscribed in the electoral census. Fines were also established for Police and Army o¢ cials in‡uencing their subordinates in electoral matters, and prison was established for some electoral practices such as falsi…cation of electoral documents and violence against electoral authorities. Before the 1922 elections, two Laws were adopted that reformed some aspects of Law 85 of 1916: Law 70 of 1917 and Law 96 of 1920. In 1917, the project presented by Represen- tative Holguín y Caro was mainly concerned with nullity demands in presidential elections. The National State Council was given the responsibility of attending such demands, and pro- cedures were set, as well as a precise timeline, for their resolution. A very illustrative article in terms of the politicians’concern about ballot stu¢ ng was added over the course of the debate by Senator Arango and other senators (AS, 1917: 386, 392). The article disenfran- chised municipalities where the number of votes exceeded one third of the total population of the respective municipality. To this end, the municipalities’population would have to be computed with the latest civil census available and, in lack of it, with the latest national census available. Moreover, for municipalities too new to be included in either of the census,

19 Lozano, like Rosales, referred to the problem of the strategic distribution of voting tables to facilitate Conservative and obstruct Liberal voting. He also criticized, among others, a disposition that allowed electoral juries, at their own judgement, to refuse providing the electoral certi…cates mentioned above if they considered that this obstructed their “good service”as well as the introduction of a …ne that those without political rights could pay to regain them.

15 the article set a fairly arbitrary way of estimating the municipality’s population based on the number of members of its Local Council. Article 8 was also of interest. Probably mir- roring Liberal concerns of irregularities in distant municipalities, it forbid the establishment of voting tables in very small villages.20 In 1920, a group of Liberal politicians proposed a new modi…cation of the Electoral Code of 1916. Apart from insisting on the cédula, introducing precision on the principle of minority representation to avoid the creation of listas corsarias, and modifying some of the processes related with the elaboration of voting lists, the project proposed a lowered threshold for disenfranchising municipalities. Such threshold was set to 15% of the municipality’s population for elections of members of Local Councils and Departmental Assemblies (in which all males older than 21 years old could participate) and to 10% of the population for presidential and congressional elections (in which male citizens had to ful…ll the age requirement plus one of the following: being literate, owning property of $1,000 pesos or more, or earning a yearly income of over $300 pesos). The debate over each of the elements of the reform was extremely animated. Among Liberals, representative Enrique Olaya Herrera led the defense of several of the measures proposed, whereas the Conservative position was defended by representatives Hernando Holguín y Caro and Sotero Peñuela.21 Many of the arguments heard in earlier debates and reviewed above were heard once more,22 and both parties accused each other for their historical wrongdoings to the point that the Government Minister decided to participate in the debate, arguing that “the honorable House has a limited time, within which it is almost impossible to issue the law on electoral reform if we continue on the path of prolonging dis- quisitions that have little to do with the main issues discussed and some representatives just stubbornly stick to a series of Byzantine discussions around this issue” (ACR, 1920: 496). Law 96 of 1920, as …nally approved, did not introduce the cédula. The measures against fraud approved included the publication of the electoral census in a visible place and within time frames that facilitated reclamations from citizens, some clari…cations on the electoral certi…cate that allowed citizens to vote on election day if they had been excluded from the …nal list while …guring on the electoral census, and also clari…cations and timelines for nullity

20 Unfortunately, there are no records of the debates on these articles in the Anales of the Congress. 21 The spirit of the discussion may be illustrated with Peñuela’sclosing comment in one of his interventions: “When you in a family …nd an unruly young man, arrogant, vicious, if he is not Liberal, sooner or later he ends in that party. Doctor Tirado Macías once told us in the House that the women of certain life are all Liberal: the reason is clear”(ACR, 1920: 500). 22 See, in particular, Olaya Herrera’s intervention on september 11 (ACR, 1920: 248-254) and Holguín y Caro’sinterventions of september 9 (ACR, 1920: 241) and september 23 and 25 (ACR, 1920: 488, 491).

16 demands (demands to invalidate the results of a particular election because of fraud).23 It was against the backdrop of this institutional framework and ongoing debate on the electoral organization that the 1922 elections took place. We now review some key aspects of the 1922 presidential election and fraud episode.

2.3 The 1922 episode and the Convención de Ibagué

As mentioned above, the 1922 presidential elections occurred in the midst of rapid socioeco- nomic change that strengthened the Liberal Party, especially in urban centers. At the same time, the Conservatives entered the presidential race divided with José Vicente Concha and Pedro Nel Ospina leading the opposing factions. Initially, Liberals attempted a coalition with moderate Conservative, José Vicente Concha. The coalition e¤ort failed, however, and Liberals ultimately proclaimed General Benjamín Herrera the most notable General in the Liberal army during the War of a Thousand Days as their candidate. In a telegram sent by Nemesio Camacho, the President of the Liberal Convention of 1921 held in Bogotá, the events were summarized as follows: “Conservatives resolutely denied a coalition agreement in acceptable terms for Liberalism; they declared they will preserve the Candidacy of Concha without pact or platform. Facing this, the Convention unanimously acclaimed, this morn- ing, the candidacy of General Herrera (...) After reiterative and determined refusals, General Herrera accepted to o¤er his name. Delirious enthusiasm”(quoted in Paz and Solano, 1922, p. 35). Herrera’sinitial reluctance to become the o¢ cial candidate of Liberalism, according to his own explanation, stemmed from his sincere interest of reaching an agreement with Conservatives. In his message to the Convention proclaiming his candidacy, he stressed that, if he had refused at …rst, it was because of his “explicit, emphatic, and public desire of reaching an agreement around a name proclaimed by the progressive sectors of the country, that could represent, more than the ideas of a single Party, the consensus around eminently national aspirations”(quoted in Paz and Solano, 1922, p. 46).24 The contest between Ospina and Herrera was very competitive. Herrera was able to obtain victory in every major city. Ospina obtained high vote shares in the countryside. The elections were obscured, however, by fraud accusations, especially involving the o¢ cial

23 Regarding article 12 of the 1920 reform, which sought to eliminate the listas de arrastre, Navarro (1935) o¤ers an interesting anecdote. Apparently there was much heated debate over several sessions. Fearing rebellion, President Suárez o¤ered a banquet and the matter was solved easily: “what seemed impossible in thirty days of bitter discussion, was achieved in one hour of cordiality and sanity”(p. 27) 24 Herrera would refer to this once more in 1922, during the Convención de Ibagué (Paz and Solano, 1922, p. 118).

17 candidate. As Deas (1993) puts it, “In 1922 the Conservative divisions were exploited by an independent Liberal coalition, and the situation was saved by the use of force at the local level and a general reliance on fraud”(p. 218). As we have seen, fraud was far from a novelty in Colombia’selections. In 1879, the following description could be found in the Diario de Cundinamarca:

“elections in Colombia are ... terrible confrontations of press, agitation, intrigue, letters, bribes, weapons, vengeance incentives, politics, choler, menace”(Guerra, 1922, p. 608).

That is, legal and illegal expressions of partisan competition coexisted. This time, how- ever, fraud accusations were widespread enough for the Liberal representative Luis de Grei¤ to demand, upon completion of vote counting by the Gran Consejo Electoral, that the fol- lowing be added to the record: “the Liberal representatives in the Gran Consejo accept the result that appears in the corresponding partial su¤rage acts veri…ed by the electoral Coun- cils as a ful…lled fact, despite our intimate conviction that such verdict is not the genuine expression of popular will, but the result of the most scandalous fraud, tolerated by author- ities and facilitated, in many cases, by government agents” (quoted in Blanco et al, 1922, p. 403). The proposition also included a list of the frauds denounced by Liberals, and the announcement of the publication of a book demonstrating their claims. The Conservative majority rejected the proposition, and proclaimed Ospina as President without any mention of the fraud denunciations. Following the elections, Herrera decided to call for an extraordinary Liberal convention in the city of Ibagué, to decide, among other things, on the posture that the party would take regarding the new government. According to Pedro Juan Navarro, after the 1922 elections and with the Convención de Ibagué “the nation’shorizon was tragically obscured with the possibility of a Civil War”(Navarro, 1935, p. 46). The threat gradually disappeared, however, and General Herrera’smotto at the time “The Nation before the parties”became famous. Still, as Navarro (1935) narrates in detail, Herrera …rmly denied Ospina’sproposal for Liberal participation in government, arguing that Liberals were convinced that they had lost because of fraud, while at the same time reassuring Ospina that they would not go to war. During the convention, Liberals emphasized on numerous occasions their intention to remain in peaceful terms with the new government. The o¢ cial Liberal summary of the convention acknowledges the “irreproachable behavior”of the authorities at Ibagué, and the general atmosphere of security, due in large part to the “chivalrous attitude of tolimense

18 conservatism”(Paz and Solano, 1922, p. 91). The convención de Ibagué left a very complete record of Liberal complaints, both in the o¢ cial summary of the convention and in a book commissioned by the convention to demonstrate Conservative abuses.25 Many of these complaints mirrored denunciations pub- lished during the campaign by the Liberal press. The irregularities denounced include the alteration of the electoral registry,26 the political activity of the clergy, and the involvement of local authorities and armed forces in irregularities and repression, including homicide of Liberals that led a Santandereano politician to summarize: “We sum up the situation of the province as follows: Police threatens public security”(Paz and Solano, 1922, p. 63). It is worth reproducing the following passage from Los Partidos Políticos en Colombia, were Liberals summarized their view on the tools that Conservatism used to remain in power:

But Conservatism takes shelter in a castle of illegal strengths even if the il- lustrated and educated majority of public opinion asks for the contrary as in the recent electoral debate. These strengths are: the army, whose nationalization has been obstructed, subordinating the armed forces to the whims of political direc- tories and committees, that use them as the decisive weapon, since the legislative Conservative majorities have refused to suspend the su¤rage right for citizens in active service (...) The electoral law, interpreted and executed by and ad-hoc power of eminently political origin, autonomous only in appearance, yet docile mirror in reality of the executive will. It has been impossible to introduce, into this law, the reforms that Liberalism has requested over and over, except when those reforms are innocuous and do not e¤ectively threaten the Conservative

25 The objective of the later book, as stated by the authors, was to “demostrate that Liberalism was not beaten in the past elections in a positive and real sense, but through fraudulent maneuvers (...) The Liberal party demostrates that it lacked the constitutional guarantees in the last electoral debate; that its adversary relied, not only on o¢ cial in‡uence, but on the strength of its authority; that Liberals were denied their rights; that in very important municipalities they disappeared from voting lists; that electoral juries, even in capitals of the Departments, operated in inaccessible locations; that minor municipalities register more votes than their entire population, even including women and children; that undescribable homicides were committed and the authors remain unpunished”(Blanco et al., 1922, p. 1). 26 For instance: “Here, all sorts of obstructions are being placed to Liberal voters, and the [electoral] census record has been distorted, once sealed and signed, to in‡ate it in the last minute with nine hundred additional names, and in spite of protests, it appears that this scandal will not be recti…ed” (quoted in Paz and Solano, 1922, p. 54, from a telegram of the Liberal Comitee in Barranquilla, Atlántico); “In this city inscription activity observed, only Liberals, who are the majority and reach one thousand names. However, in the de…nite lists six thousand Conservatives appeared also, …lling the allowed legal space. In the remaining municipalities, except Buenavista, Sutamarchán, worse conditions [and on the role of the church] Dominicans stoke multitudinary hatred" (quoted on Paz and Solano, 1922, p. 65, from Liberal Commitee in Chiquinquirá).

19 hegemony, that is, the perpetuation of that party in the Government. Liberalism accepted in good faith, to improve it gradually and in agreement with all parties, the incomplete vote system; now it has maliciously been positioned in the eternal role of minority and it is expected that it surrenders to such condition even where it is an overwhelming majority (...) Then we have public schooling, stolen its high mission and tainted with the brightest partisan color, it lacks today, other end than "conservatizing" new generations, directing them on a path of acute religious proselytism [and] if we add the combative and at times implacable atti- tude of priests it is clear that we …nd ourselves, as a nation, witnessing maybe a unique problem in the world (Blanco et al, 1922, p. 15, 17).

Even considering some degree of exaggeration on the Liberal discourse, it is clear that Conservatives used diverse fraudulent methods during the elections. Regarding ballot stu¤- ing, Liberals accusations claimed that the “fraudulent multiplication” of votes was largest in Cundinamarca and the Santanderes, were there was Liberal majority and hence “it was necessary, to deceive it, to rely on the greatest frauds ever registered. The multiplication of votes caused vertigo”(Blanco et al, 1922, p. 27). Regarding other departments, they claim that in Valle, Antioquia and Caldas, fraud consisted mostly of inscribing Conservatives in the voting lists even when they did not meet the legal requirements, and obstructing Liberal inscriptions. Apparently, fraud was less widespread there, “where, if there were irregularities, at least the scandalous ‘chocorazos’[ballot stu¢ ng] of other departments were not observed” (p. 399). In Atlántico and Magdalena, the substitution of voting lists with fake ones is re- garded as the most common fraud, and …nally in Nariño and Boyacá, where Conservatism was the norm amongst “illiterate farmers”, Conservatism “multiplied votes appallingly, and hence the two illiterate Departments lead the number of voters”(p. 27). What about Liberal wrongdoings? Unfortunately, possibly because they were victorious but also partly because Liberal fraud was less widespread, there is no comparable Conser- vative e¤ort to denounce the Liberal abuses systematically. Liberals themselves recognized possible abuses, but claimed they were minor: “It might be that one or two irresponsible Liberals made a small fraud in just a few Liberal municipalities to counterbalance”(Blanco et al, 1922, p. 25). A response to the Liberal discourse can be found also in Guerra (1922). In similar vein as Peñuela (1922, p.4), Guerra (1922) accuses Liberals for predicating princi- ples that they violated when in power.27 He dismisses the accusations of fraud by Liberals in

27 “Del dicho al hecho hay mucho trecho” (roughly, from word to fact there is a long tract) he exclaims,

20 1922 by saying, “We have seen how the word fraud was used and abused carelessly, echoed in the Convención de Ibagué, attributing to it and never to the numbers, the Liberal party’s defeat, even in merely Conservative towns, during the electoral debate preceding the reunion of such organization”(p. 617, see also p. 380). In any case, there is no systematic accusation of fraud by Liberals, nor any systematic denial of Liberal accusations. For this reason, the anecdotal evidence on electoral fraud that is readily available for the 1922 election is heavily biased towards Liberal accusations. It is likely that a more balanced picture of electoral fraud will be obtained with our approach to quantify it, which will be explained in the next section.

3 The Data

3.1 Ballot Stu¢ ng

In this paper, we will approximate the extent electoral fraud by constructing a measure of ballot stu¢ ng. Our measure relies on the comparison of the total number of votes cast in each municipality with a reasonable estimate of the size of the franchise from information of the 1918 National Census. This measure is imprecise since we do not have accurate data on the real level of turnout. The arbitrary exclusion of voters from the electoral registries, which historical evidence suggests was common, is especially problematic, as several of the munic- ipalities that reveal no ballot stu¢ ng in our database might have experimented electoral fraud nonetheless. In spite of this caveat, that should be kept in mind, ballot stu¢ ng was perhaps one of the most prevalent forms of fraud (indeed, it was so common that there were names for the practice of it, as shown in the quote from Lozano above). Finally, our measure of ballot stu¢ ng is not likely to be in‡uenced by the strategic e¤ects of other measures of electoral fraud based on testimonies of party followers.28 to criticize the use of extraordinary powers, and also of “scandalous electoral frauds” (p. 189) by Liberals during the period of the Olimpo Radical. He also has an entire Chapter entitled El Derecho Electoral (?) with a collection of quotations on electoral fraud starting in the early nineteenth Century. 28 The idea of constructing a measure of ballot stu¢ ng based on reasonable estimates of the real size of the franchise is not new. As soon as the 1922 elections were over and the o¢ cial count was relased, Liberals soon noticed that General Benjamín Herrera obtained around one thousand more votes than the ones obtained by Marco Fidel Suárez, the winning conservative candidate, in the 1918 elections (see, e.g., Navarro, 1935, p. 37). In Los Partidos Políticos en Colombia, where Liberal complaints were summarized, Liberals used the available statistics to draw some calculations in the spirit of the ones we construct in this section showing results for each department and the country as a whole, and attributing the “multiplication of votes”to the conservative party (see Rodríguez, 1922).

21 To estimate ballot stu¢ ng we proceed as follows. As explained above, under the 1916 Electoral Code, su¤rage rights were restricted to adult males (over 21 years of age), and for presidential elections male citizens had to ful…ll the age requirement plus one of the following: being literate, owning property of $1,000 pesos or more, or earning a yearly income of over $300 pesos. The income and wealth requirements implied by these thresholds are fairly restrictive. For example, nominal GDP per-capita in Colombia in 1922 was about $84 (GRECO, 2002) so that to qualify to vote using the income criterion an illiterate person would have had to earn almost 3 times average income. Given that around 50% of adult males were literate in 1950, it is plausible that very few illiterates could have earned such high incomes. Using data on land ownership for the department of Cundinamarca in 1890 (see below) and adjusting for prices suggests that if one owned $1,000 worth of land one would be in the top 21% of landowners. Hence, it seems very unlikely that an illiterate male would have been able to qualify to vote on the basis of wealth holdings either. In consequence we assume that everyone who could qualify to vote on the basis of land ownership and income were also literate. This assumption implies that landowners and earners of income over $300 are subsets of the literate males, and that the number of adult literate males is a reasonable estimate of the franchise.29 We therefore use the 1918 National Census to compute the number of males over 19 years of age in every municipality ( the census does not report males over 21), and multiply this number by the literacy rate of men in each municipality. Since the presidential election was held in 1922 and the Census was made in 1918, we may be underestimating the franchise. Hence, assuming a rate of population growth consistent with the information from the 1918 and 1928 National Censuses, we also adjust our estimate of the adult literate male population to allow for population growth. This constitutes our measure of the size of the franchise in each municipality. It is clear that this measure of the franchise is an overestimate since it assumes a 100% voter turnout and since only people older than 21 could vote. This will therefore tend to create relatively conservative measures of electoral fraud. We combine our estimate of the franchise with the total number of votes cast in each municipality according to the o¢ cial electoral registries sent by local authorities to the

29 These assumptions are probably implausible in a number of cases. For instance, there are municipalities with extremely low literacy rates, which deliver extremely low estimates of literate adults (2 men, in the most extreme case in San Andrés de Sotavento, Bolívar). This implies that stu¤ed ballots as percent of estimated franchise in San Andrés de Sotavento was 22,000%! One could expect it to be likely that where so few men were literate there are some non-literate individuals with the income or wealth requirements to get voting rights. There is no easy solution to this problem other than to check that our results are robust to dropping such municipalities, which they are (details from authors upon request).

22 Gran Consejo Electoral. Totals for the whole country indicate that around 60% of the municipalities exhibited ballot stu¢ ng. Although it is present throughout the whole country, places where accusations were less widespread such as Antioquia do seem to exhibit less incidence of ballot stu¢ ng and a moderate intensity as well. The ratio of stu¤ed ballots to the estimated franchise is very large, reaching over 200% on average because of the very extreme cases in departments like Guajira with very low literacy rates. The estimated number of stu¤ed ballots for the whole country according to this …rst measure is around 225,000, which is somewhat larger than the o¢ cial margin of victory of Ospina over Herrera (431,901 votes against 238,673). Ballot stu¢ ng is very prevalent in the eastern Andean region (Boyacá, Cundinamarca and the Santanderes) as well as in the Coast, with somewhat less prevalence in places like Antioquia.

3.2 Measuring Coercion

To measure coercion we coded the information from the proceedings of the Convención de Ibagué. In the book there are many accusations of coercion which we sorted into di¤erent types of coercion using dummy variables to capture whether or not a particular type of violence was present in a municipality. These are

1. Violence=1 if the municipality had reports of actual violence breaking out: brawls, gun-shots hit their target confrontations with injured or casualties.

2. Intimidation/Harassment=1 for reports of incarcerating Liberals, subjecting them to random searches and detainment, coercive measures to prevent Liberal propagandizing or activism.

3. Arms distribution/paramilitary activity=1 for reports of organized armed Conserva- tives who are not police or army or distribution of arms for these bodies. Activities of intimidation by these bodies.

4. Coercion=1 indicator for the union of violence, intimidation, arms etc. 1 if any of the above happened.

In the empirical work we investigate only Coercion. See the data appendix for more details on the construction of this data.

23 3.3 Explanatory Variables

One of the most important hypotheses we wish to investigate in our paper is that the presence of the state reduced the extent of ballot stu¢ ng and coercion. As proxies for the presence of the state in di¤erent dimensions we use data from the 1918 population census on the number of public employees and the number of agents of the armed forces in each municipality. We use these variables on their own and also add them to construct a simple index of the presence of the state.30 The historical literature also emphasizes the important political role of clergy during elections, and as we have already seen, this was also much remarked upon by Liberals at the time. From the 1918 Census we have the number of priests in each municipality which we express as a proportion of the total population. Unfortunately we do not have a good control for the level of economic development at the municipality level. Though the Census does report data on literacy and schooling we obviously cannot use this as a control variable for ballot stu¢ ng since they are mechanically related to our measure of ballot stu¢ ng given that we use the literate male population to construct the number of stu¤ed ballots. We do use these as control variables when coercion is the dependent variable however. However, the Census does include the proportion of people in a municipality who were vaccinated. Since this is very likely related to income per-capita we use this variable as an imperfect control for income when ballot stu¢ ng is the dependent variable. The literature also suggests that there may be large di¤erences between core and periph- eral areas of municipalities and it is desirable to control for this directly. To do so we include the earliest foundation date of the municipality. We expect newer and more peripheral mu- nicipalities to exhibit more fraud, as the presence of the state is likely to be weaker in such municipalities.31

30 Ideally, we would like to examine the impact of the police and the army separately. Unfortunately, the 1918 Census does not distinguish between the two. In future research we expect to use information from the 1912 National Census, which does discriminate between police and army o¢ cials. Pinzón (1994) notes that “the o¢ cial bias of the police during the campaign was clear. On January 13 of 1922, a political demonstration of liberals in Bogotá was violently crashed by armed conservatives with police support (...) For a long time there had been little discussion over the electoral participation of the army, but precisely some of the accusations of corruption in these elections referred to that”(p. 78, 79). 31 We also used distance (in kms) to the departmental capital which gave similar results.

24 3.4 Descriptive Statistics from Colombia

Table 2 reports the descriptive statistics for the whole of Colombia. The …rst row reports the ratio of stu¤ed ballots to adult males for the 546 municipalities for which we have complete data (we lack occupational data for Boyaca, Bolivar and Magdalena). Table 3 shows the distribution of municipalities across our base sample. This sample is even smaller for coercion and the * marks the departments for which we do not have coercion data. The …rst row shows that the mean number of stu¤ed ballots was 19% of the total adult male population. Looking across this row it is interesting to note that the proportion of ballots stu¤ed seems to vary little between municipalities with and without recorded coercion. With respect to the presence of the state the proportion of the population which were clergy was 50% greater in areas with low stu¤ed ballots (municipalities less than the median) while the presence of members of the army and of the bureaucracy also seems to be higher in places with relatively low levels of ballot stu¢ ng (0.55 instead of 0.47). Though as noted, we cannot say anything about the relationship between human capital and ballot stu¢ ng, the table also suggests with respect to coercion that schooling is greater in places with coercion compared to those without. Looking at the vaccination rate, this is higher in municipalities with low ballot stu¢ ng but, interestingly, higher in places where coercion is present than in those where coercion is absent, though the di¤erent is quite small. Table 4 shows the correlation matrix for our di¤erent variables. There is some quite small negative correlation between both the presence of clergy and government bureaucrats and the extent of ballot stu¢ ng. However, ballot stu¢ ng and coercion are almost uncorrelated.

3.5 Data on Cundinamarca

In addition to the data on stu¤ed ballots and coercion and the covariates from the 1918 Census we have a variety of other historical data for Cundinamarca from Acemoglu et al. (2008). These authors collected cadastral (land census) data collected by the state of Cun- dinamarca in 1879 and 1890. We use a very standard measure of land inequality from their paper - the land gini coe¢ cient, which measures land inequality among landowners.32 For

32 Despite its widespread use, the land gini su¤ers from an obvious problem. An area in which all land is held by two very large landowners will have a low value of the land gini, because land is equally distributed among landowners. But if we looked at the population as a whole, there would be tremendous amount of land inequality. To alleviate this problem, Acemoglu et al. (2008) constructed an alternative measure, overall land gini, which again computes equation (1), but uses the total number of families and assigns zero

25 each municipality at each date, we construct the gini coe¢ cient using the standard formula

1 nt nt g = y y (1) mt n2y i;t j;t t t i=1 j=1 j j X X

where i = 1; :::; nt denotes the total number of land owners at time t, yi;t is the value of land nt owned by individual i at time t, and and y = 1 y is the average value of land at time t nt i;t i=1 t. Throughout most of our analysis, we average theX gini coe¢ cients across the two dates for each municipality to arrive to our measure of (average) land gini. The average gini over this entire period was 0.65 (see Table 5 below). To measure political concentration Acemoglu et al. (2008) collected data on politician (mayor) names from the Registro del Estado and Gaceta de Cundinamarca, o¢ cial news- papers which published the names of principal and substitute mayors appointed in each municipality between 1875 and 1895. They used these data to construct a measure of the concentration of political power. Their measure of political concentration for municipality m at time t is computed as:

Number of Di¤erent Individuals in Powermt pmt = : Number of mayor appointmentsmt The negative sign in front is introduced so that higher values of the index correspond to higher political concentration (thus making the interpretation of the coe¢ cients easier). Consequently, our political concentration index takes a value of -1 when there is very low political concentration, and values close to 0 for high levels of concentration. We computed this index for the whole period 1875-1895. The mean of this variable is -0.56. In addition to basic measures of economic inequality and political concentration, Ace- moglu et al. (2008) constructed a measure of the overlap between political o¢ ceholding and landed wealth. To do this they classi…ed the individuals according to whether they were politicians, rich, those individuals whose land plots are in the top 25% most valuable plots,

or both. De…ning Rmt to be the set of adult males living in municipality m at time t who are

rich and Pmt be those with political power (mayors). Let #Rmt be the number of individuals

in the set Rmt, and de…ne #Pmt similarly. Overlap in municipality m at time t is computed as land holdings to the families who do not appear in the catastro. We found this variable to be completely ucorrelated with ballot stu¢ ng and other outcomes and it did not in‡uence any of the results we report here so we dropped it from the analysis.

26 #(Rmt Pmt) omt = \ : #(Rmt Pmt) [ Their measure of overlap, which we use in this paper, is the average of this index for the two dates 1879 and 1890. In Table 5 we report some descriptive statistics just for Cundinamarca. For the variables we discussed above there are a few di¤erences. For instance, ballot stu¢ ng is now much higher in municipalities without coercion (0.34) compared to municipalities with coercion (0.19). In terms of our new explanatory variables the land Gini is higher in places with low ballot stu¢ ng (0.67 compared to 0.63) while political concentration is also higher in municipalities with less than median levels of ballot stu¢ ng. This table also suggests that schooling and literacy are higher in places with less ballot stu¢ ng and more coercion. Finally Table 6 examines pairwise correlations for Cundinamarca. Now we see a distinct negative correlation between coercion and ballot stu¢ ng. We also see a strong negative correlation between both land inequality and political concentration and ballot stu¢ ng. The presence of the clergy is negatively correlated with ballot stu¢ ng and positively correlated with coercion.

4 Econometric Analysis

Having presented the main features of our measures of ballot stu¢ ng and coercion, this section analyzes some of the correlations between those measures and other variables at the municipality level, in an attempt to describe which characteristics were associated with a higher incidence and intensity of ballot stu¢ ng and coercion.

4.1 Ballot stu¢ ng and Coercion in Colombia

Table 7 present the results of simple ordinary least squares regressions for the whole of Colombia’smunicipalities. The basic model we estimate is

ym = Gm0 + d + Xm0  + "m (2)

In (2) y is our explanatory variable of interest. In panel A this is log 1 + sm where s m pm m represents the number of stu¤ed ballots in municipality m and pm the population of the municipality. In panel B the dependent variable is cm which is a dummy such that cm = 1 if

27 municipality m experienced a coercion incidence. In (2) Gm0 is a vector of variables capturing

the presence of clergy and of the state in municipality m, d is a department …xed e¤ect,

where the subscript d indexes department, and Xm0 is a vector of covariates, such as the vaccination rate or the foundation date of the municipality which also includes a constant.

The error term "m captures all omitted in‡uences, including any deviations from linearity.

Equation (2) will consistently estimate the parameters of interest if Cov(gm;"m) = 0 for all gm Gm. Nevertheless, we emphasize that these covariance restrictions are unlikely 2 to hold in practice, since the presence of the state and political outcomes, such as ballot stu¢ ng and coercion are all jointly determined and this is why we do not emphasize causal interpretations of out …ndings. We …rst turn to panel A and regressions where the dependent variable is log 1 + sm . pm The …rst three columns introduce sequentially the proportion of the population which are clergy, armed forces or bureaucracy. All regressions have departmental …xed e¤ects. The …rst column shows that the presence of clergy is signi…cantly negatively correlated with ballot stu¢ ng. The estimated coe¢ cient is 23:54 with a standard error of 5:85 and so is highly signi…cant. The second column shows that though the estimated coe¢ cient on armed forces is negative it is not statistically signi…cant. In the third column the proportion of the population which is bureaucrats is estimated to have a coe¢ cient of 4:38 (s.e.=2:11) which is again signi…cant. Column 4 adds all of these three explanatory variables at the same time. Though this reduces the coe¢ cient on bureaucracy and makes it insigni…cant, the correlation with the clergy is almost unchanged as is the level of signi…cance. Since there is a concern about the presence of multicollinearity, in column 5 we amalgamate the data for army and bureaucracy by simply adding them to form an index of state presence. This has a negative coe¢ cient though is not signi…cant, and nothing happens to he coe¢ cient on clergy or its standard error. In column 6 we drop the departmental capitals because the case study literature suggests that the political dynamics of these places may be distinct. Nevertheless, this has little impact on the results. In column 7 we add the foundation date of the municipality as a simple control for how central or peripheral it is. This seems to be positively correlated with ballot stu¢ ng suggesting, as we would have conjectured, that more recently founded municipalities have more ballot stu¢ ng, but it does not qualitatively change the other results. Finally, in column 8 we add the proportion of the population who is vaccinated as a simple

28 control for the prosperity of the municipality. The coe¢ cient on this variable is negative, as we would have anticipated, but it is not signi…cant. Moreover, it has little in‡uence on the results of interest suggesting that it is unlikely that our …ndings are being driven by the simple fact that ballot stu¢ ng is lower in more prosperous municipalities. Panel B estimates very similar regressions except that now the dependent variable is the dummy variable for coercion. One can think of these regressions then as a linear probability model. The big fact that jumps out of this panel of the table is that the presence of the clergy is positively correlated with coercion. The e¤ect is very robust across the di¤erent speci…cations and no other variable appears to be signi…cantly correlated with coercion. It is interesting to examine some of these results graphically. Figure 1, for instance, plots the data for the ratio of clergy to municipal population against the extent of ballot stu¢ ng. There is evidence of a quite distinct negative relationship. Figure 2 replaces the clergy by the number of government bureaucrats and again shows evidence of a negative relationship.

4.2 Preliminary Interpretation

The results from the regressions for Colombia as a whole suggest a couple of things. First, the presence of priests is positively correlated with coercion but negatively correlated with ballot stu¢ ng. The …rst fact is certainly consistent with the claims of Liberals at the time and a great deal of historical literature. What is important for interpreting the second correlation, however, is the form of coercion in which priests were involved. Speci…cally, this was targeted at keeping Liberals away from the polls. Hence where there were priests, Liberals were stopped from participation, thus turnout was lower, and thus the extent of ballot stu¢ ng according to our measure we lower. One could imagine that using coercion to keep Liberals away from the polling booths would have given Conservatives the freedom to engage in even more egregious ballot stu¢ ng. Our results suggest however though while this may have increased the margin by which Conservatives won in a particular municipality, other things equal, it actually reduced the extent of ballot stu¢ ng relative to the potential franchise. Second, the regressions also provide some weak support for the notion that the presence of state bureaucrats is negatively correlated with ballot stu¢ ng, though this does not appear very robust.

29 4.3 Ballot stu¢ ng in Cundinamarca

We now focus just on the department of Cundinamarca. Table 8 re-estimates the models of Table 7 using only our data for Cundinamarca. This shows some interesting patterns. First, as for the whole of Colombia, the presence of clergy is negatively correlated with ballot stu¢ ng and positively correlated with coercion. The coe¢ cients are nearly always statistically signi…cant and their magnitude is not very dissimilar than that for the entire country. Table 8 Panel A however also shows much more robust e¤ects of state presence on ballot stu¢ ng. The presence of the bureaucracy is always signi…cantly negatively correlated with ballot stu¢ ng and the estimated impact of the army, though not typically signi…cant at standard con…dence levels, is always negative. Finally in Panel A the vaccination rate is also signi…cantly negatively correlated with ballot stu¢ ng and while this reduces the impact and signi…cance of the presence of clergy, it has little impact on the correlation between bureaucracy and ballot stu¢ ng. We would interpret these results as suggesting that our …ndings are not being driven by unobservables such as income per-capita. Turning to panel B we see once more a very robust relationship between clergy and coercion. Coercion and the clergy go together. This part of the table also suggests two other relatively robust …ndings. First, the presence of the army reduces coercion. Second, coercion is greater in places with higher human capital. For instance in column 9 the schooling rate has an estimated coe¢ cient of 0:323 (s.e.=0.096) suggesting that greater schooling is associated with greater coercion. As with our …ndings with respect to vaccination, we believe this …ndings shows that our results cannot be driven by some simple implicit modernization thesis suggesting that coercion and ballot stu¢ ng are lower in relatively prosperous places. In Table 9 we introduce our new explanatory variables which we only have available for Cundinamarca. Again in panel A our measure of ballot stu¢ ng is the dependent variable. The …ndings here are very robust across speci…cations. In column 1 we see that the land Gini is negatively correlated with ballot stu¢ ng with a coe¢ cient of 0:623 (s.e.=0.201). In the second column we see that although the correlation between ballot stu¢ ng and political concentration is negative, on its own it is not signi…cant. Nevertheless, looking across the rest of the columns of the table one sees that there is a robustly estimated negative correlation between political concentration and ballot stu¢ ng. For instance in column 6 when land gini, political concentration and overlap are all introduced, the estimated coe¢ cient on political concentration is -0.82 (s.e.=0.20) and thus highly signi…cant.

30 In column 3 we control just for overlap, which is positively correlated with ballot stu¢ ng. The estimated e¤ect is 1.34 (s.e.=0.65) and once the covariates are added together this e¤ect becomes slightly larger and estimated just as precisely. Panel A therefore suggests that while land inequality and political concentration are negatively correlated with ballot stu¢ ng, the extent of overlap is positively correlated with stu¢ ng. In panel B we estimate the same model except that the coercion dummy is now the explanatory variable and we can now also use our education data (in columns 9 and 10). There is one very robust …nding from these regressions which is that there is a signi…cant negative correlation between political concentration and coercion. Interestingly, column 9 of this panel also suggests that there is a positive signi…cant correlation between schooling and coercion, something that was suggested by our analysis of the raw data and simple bivariate correlations above. Table 10 then combines the covariates of Tables 8 and 9. The results mirror closely those of the previous two tables. Figures 3 and 4 re-examine the relationships in Figures 1 and 2 with data just from Cundinamarca where we now use the names of the municipalities since this is reasonably clear. Both show evidence of the same negative correlation. One might be concerned in Figure 3 that this was being driven by Bogotá, but recall that our regression results are robust to dropping departmental capitals. Figure 5 depicts the scatterplot between the land gini and ballot stu¢ ng. Here there is a quite distinct negative relationship. Municipalities on the Sabana de Bogotá, the plain surrounding the city where the …rst Spanish conquistadors settled in the early 16th century, such as Mosquera or Sopó, have high land inequality and relatively low levels of ballot stu¢ ng. Figure 6 examines political concentration and ballot stu¢ ng. It is much less clear from this picture that there is any signi…cant correlation between the variables. Finally in Figure 7 we look at the relationship between overlap and ballot stu¢ ng. There is clearly some suggestion of a positive correlation.

4.4 Interpretation

Building on Acemoglu et al. (2008) and our earlier observations there seems to be some immediate interpretations of these …ndings. Though the Colombian state may have been weak in 1922, where it was present it served to reduce the extent of fraud. Bureaucrats reduced ballot stu¢ ng and the army seems to have reduced coercion. As Montoya (1935, p.

31 42) argues

“It is undeniable that for long time and under di¤erent political regimes, the Colombian government used the armed forces as an instrument for fraud, and that members of the army were docile and at times eager agents of such condemnable system; but it is not less evident, to the honor and joy of our Nation, that those practices have disappeared”

The role of the church was very di¤erent however. Even if during the colonial period the church is often seen as synonymous with the state, this was clearly not true in Colombia in 1922. Our evidence is very consistent with historical work and contemporary sources which stressed the close relationship between the clergy and the Conservative party. It was the Conservatives who mostly persecuted the fraud of 1922, and they were helped by clergy organizing anti-Liberal violence. The most interesting …ndings of this section however relate to land inequality, political concentration and overlap. Land inequality and political concentration are negatively corre- lated with ballot stu¢ ng and this is very robust. Similarly, overlap is positively correlated with ballot stu¢ ng. Finally we found political concentration to be signi…cantly negatively correlated with coercion. These …ndings can be …tted together. In Cundinamarca landed elites were not competing with the state, as they may have been in 19th century Germany (Ziblatt) or Chile in the 1950s (Baland and Robinson, 2009). Instead, they were substituting for it and in doing so reduced the extent of fraud. This explains why high land inequality is negatively correlated with ballot stu¢ ng. Not all elites were traditional however and some had used their political o¢ ces and connections to acquire wealth. In these places, munici- palities of high overlap, elites needed to supply fraudulent ballots to national politicians to guarantee their power and newly found wealth. Finally, our results of political concentration are easy to understand. As Acemoglu et al. (2008) point out, places where political con- centration was high were peripheral areas where power was often in the hands of a powerful caudillo. These places, known as ‘independent republics’were likely little involved in na- tional politics so as a result one sees little coercion of ballot stu¢ ng. There is an interesting non-monotonicity here. In Cundinamarca in places at the core of the department where the state functioned or landed elites could substitute for it, there was relatively little coercion and ballot stu¢ ng. Further away, as the grip of the state and elites weakened, ballot stu¢ ng and coercion increased. Yet in the most distant parts of the department, where caudillos ruled almost without concern for the state, electoral fraud fell o¤ because these areas were

32 disconnected from the partisan rivalries which tore the nation apart.

5 Conclusions

In this paper we have investigated electoral corruption in Colombia’s1922 presidential elec- tions by o¤ering a measure of the presence and extent of ballot stu¢ ng at the municipality level and the incidence of coercion. Our …ndings, summarized above, are quite distinct from existing studies. For one, our focus is very much on the role of the state. Though politics in Colombia was highly partisan, our results suggest that by 1922 some element of a ‘Weberian state’had emerged in Colombia and that state o¢ cials were focused on reducing, not imple- menting fraud. The church, on the other hand, played a signi…cant role in coercing Liberal voters and keeping them away from the poles. For another, we found that land inequality is negatively correlated with ballot stu¢ ng. Contrary to other studies which have found evi- dence suggesting that landed elites are implicated in fraud, the data from Colombia suggests the opposite. We argue that this is a consequence of state weakness. Though the state might have fought against fraud, its ability to do so was highly limited. Consequently, landed elites in Colombia found it in their interests to substitute for the state, not undermine it, and as such fought against the chaos and illegality that went along with electoral fraud. Nevertheless, one should keep in mind the problems with the data sources which we have used to draw these conclusions. Though we have real data on recorded vote totals from the 1922 election we had to estimate the franchise because we have to objective information on turnout. We believe that this means that our estimates are a lower bound on the extent of ballot stu¢ ng but this obviously introduces potential biases into our estimates that are hard to evaluate because we do not know what in‡uenced the extent of turnout across municipalities or the extent to which it might have been correlated with our explanatory variables. This measure of fraud is complementary to existing measures which only rely on accusations of fraud. Indeed, we used accusations ourselves to construct a measure of electoral coercion. Nevertheless, there seems to be a lot of potential in using the type of information we used in this paper for examining electoral corruption. For example, in the critical presidential run-o¤ election in Sierra Leone in November 2007 477 polling stations (out of 5,679) had turn-outs of over 100% (i.e. more votes cast than the registered electorate) which is impossible without fraud or illegal action (see Robinson, 2008).33 Exploring the

33 By district the number of polling stations disenfranchised was: Kailahun: 90, Kenema: 65, Kono: 9, Bombali: 17, Kambia: 8, Koinadugu: 12, Port Loko: 2, Tonkolili: 6, Bo: 123, Bonthe: 2, Moyamba: 24,

33 determinants of variation in such ballot stu¢ ng is obviously a fascinating research agenda.

6 References

Acemoglu, Daron, María Angélica Bautista, Pablo Querubín and James A. Robinson (2008) “Economic and Political Inequality in Development: The Case of Cun- dinamarca, Colombia,” in Elhanan Helpman ed. Institutions and Economic Performance, Cambridge: Harvard University Press. Acemoglu, Daron, Simon Johnson, James A. Robinson and Pierre Yared (2007) “Revisiting the Modernization Hypothesis,”NBER Working Paper # Acemoglu, Daron, Simon Johnson, James A. Robinson and Pierre Yared (2008) “Income and Democracy,”American Economic Review, Acemoglu, Daron and James A. Robinson (2006) Economic Origins of Dictator- ship and Democracy, New York: Cambridge University Press. Anales de la Camara de Representantes (ACR), Various numbers. Anales del Senado (AS), Various numbers. Arrubla, Mario (1995) “Sintesis de Historia Política Contemporánea,”in Melo, Jorge Orlando (ed), Colombia Hoy: Perspectivas Hacia el Siglo XXI, Bogotá: Tercer Mundo. Blanco, Pedro, Armando Solano, and Jorge Rodríguez et al. (1922) Los Par- tidos Políticos en Colombia, Bogotá: Aguila Negra. Baland, Jean Marie and James A. Robinson (2008) “Land and Power: Theory and Evidence from Chile,”forthcoming American Economic Review. Bensel, Richard F. (2004) The American ballot box in the mid-nineteenth century, New York: Cambridge University Press. Bernard, Olivier and Fabio Zambrano (1993) Ciudad y Territorio: El Proceso de Poblamiento en Colombia, Bogotá: Academia de Historia de Bogotá. Bushnell, David (1993) The Making of Modern Colombia: A Nation in Spite of Itself. Berkeley: University of California Press. Collier, Ruth Berins (1999) Paths Towards Democracy: The Working Class and Elites n Westrn Europe and South America, New York: Cambridge University Press. Cox, Gary W. and J. Morgan Kousser (1981) “Turnout and Rural Corruption: New York as a Test Case,”American Journal of Political Science, 25, 646-663.

Pujehun: 113, Western Rural: 1 and Western Urban: 5.

34 Deas, Malcolm (1993) “Algunas notas sobre el caciquismo en Colombia,”In Del poder y la gramática y otros ensayos sobre historia, política y literatura colombianas, Bogotá: Ter- cer Mundo Editores. Deas, Malcolm (1996) “The Role of the Church, the Army, and the Police in Colom- bian Elections, c. 1850-1930,” in Posada-Carbó (ed), Elections Before Democracy: the History of Elections in Europe and Latin America, London: Institute of Latin American Studies/Macmilan, 163-180. GRECO (2002) El Crecimiento Económico Colombiano en el Siglo XX, Bogotá: Fond de Cultura Económica. Guerra, José Joaquín (1922) Viceversas Liberales. Documentos Relativos a la Histo- ria del Liberalismo Colombiano, Bogotá: Editorial La Cruzada. Holguín Arboleda, Julio (1959) Mucho en Serio y Algo en Broma, Bogotá: Editorial Pío X. Lehoucq, Fabrice E. and Iván Molina (2002) Stu¢ ng the Ballot Box. Fraud, Elec- toral Reform, and Democratization in Costa Rica, New York: Cambridge University Press. Lehoucq, Fabrice (2003) “Electoral Fraud: Causes, Types, and Consequences,” An- nual Review of Political Science, 6, 233-256. Mazzuca, Sebastián and James A. Robinson (2009) “Political Con‡ict and Power- sharing in the Origins of Modern Colombia,”forthcoming in the Hispanic American Histor- ical Review. Medina, Medó…lo (1991) “Obisbos, Curas, y Elecciones. 1919-1930,”Anuario Colom- biano de Historia Social y de la Cultura, 18-19, 185-204. Medina, Medó…lo (1994) “La Historiografía Política del Siglo XX en Colombia,” In Bernardo Tovar (ed), La Historia al Final del Milenio: Ensayos de Historiografía Colombiana y Latinoamericana,Vol II, Bogotá : Editorial Universidad Nacional, 433-536. Melo, Jorge Orlando (1995) “La República Conservadora,”in Melo, Jorge Orlando (ed), Colombia Hoy: Perspectivas Hacia el Siglo XXI, Bogotá: Tercer Mundo. Montoya, Hernán (1938) La Cédula y el Sufragio, Bogotá: República de Colombia, Ministerio de Gobierno. Navarro, Pedro Juan (1935) El Parlamento en Pijama. Este libro también podría llamarse Aguafuerte del Partido Conservador, Bogotá: Mundo al Día. Paz, Felipe and Armando Solano (1922) La Convención de Ibagué, Bogotá: Edito- rial de Cromos. Peñuela, Sotero (1922) El Liberalismo: lo que ha sido y lo que es, Bogotá: Casa

35 Editorial de La Cruzada. Pinzón, Patricia (1994) El ejército y las elecciones: ensayo histórico, Bogotá: Cerec. Posada-Carbó, Eduardo (1995) “Civilizar las urnas 1830-1930,” Boletín cultural y bibliográ…co,39:3-25 Posada-Carbó, Eduardo (1997) “Limits of Power: Elections Under the Conservative Hegemony in Colombia, 1886-1930,”The Hispanic American Economic Review, 77 (2): 245- 279. Posada-Carbó, Eduardo (2000) “Electoral Juggling: A Comparative History of the Corruption of Su¤rage in Latin America, 1830-1930,” Journal of Latin American Studies, 32, 611-644. Registraduría Nacional del Estado Civil (1991) Historia Electoral Colombiana, Bogotá: Imprenta Nacional de Colombia. Reyes, Alejandro (1978) Latifundio y Poder Político. La Hacienda Ganadera en Sucre. Bogotá: Editorial CINEP. Robinson, James A. (2008) “Governance and Political Economy Constraints to World Bank Country Assistance Strategy Priorities in Sierra Leone,”Unpublished. Rodríguez, Jorge (1922) “Prestidigitación Electoral,” In Blanco, Pedro et al., Los Partidos Políticos en Colombia, Bogotá: Aguila Negra. Rueschemeyer, Dietrich, Evelyn H. Stephens and John D. Stephens (1992) Capitalist Development and Democracy, Chicago: University of Chicago Press. Tirado Mejía, Alvaro (1995) “Colombia: Siglo y Medio de Bipartidismo,” in Melo, Jorge Orlando (ed), Colombia Hoy: Perspectivas Hacia el Siglo XXI, Bogotá: Tercer Mundo. Wilkinson, Steven I. (2004) Votes and violence : electoral competition and ethnic riots in India, New York: Cambridge University Press. Ziblatt, Daniel (2008) “Shaping Democratic Practice and the Causes of Electoral Fraud: Theory and Evidence from pre-1914 Germany,” Unpublished, Department of Gov- ernment, Harvard University.

36 Table 1

Department Total votes Pedro Nel Ospina Benjamin Herrera Ospina-Herrera Stuffed Ballots Total Population in 1918 Antioquia 76,420 47,987 28,403 19,584 11,658 821,027 Arauca 425 146 278 -132 10 6,070 Atlantico 9,905 4,840 5,064 -224 1,488 117,915 Bolivar 49,548 33,650 15,888 17,762 29,207 416,561 Boyaca 83,764 61,977 21,747 40,230 39,688 590,587 Caldas 50,186 28,610 21,566 7,044 7,104 419,697 Caqueta 363 270 93 177 124 2,957 Casanare 68 10 58 -48 1 1,382 Cauca 23,024 13,644 9,367 4,277 8,416 238,071 Choco 7,214 3,467 3,746 -279 2,649 61,371 Cundinamarca 117,471 76,634 40,723 35,911 48,517 800,439 Guajira 1,460 1,063 397 666 1,407 2,908 Huila 13,864 8,830 4,997 3,833 2,513 181,202 Magdalena 18,577 11,657 6,918 4,739 7,465 186,254 Meta 2,477 1,255 1,221 34 808 10,695 Narino 29,843 23,880 5,959 17,921 4,237 327,367 Norte de Santander 35,705 26,894 8,804 18,090 17,449 222,552 Putumayo 850 660 190 470 414 5,009 San Andres y Providencia 566 465 100 365 0 5,953 Santander 55,492 37,784 17,699 20,085 23,926 439,161 Tolima 39,083 19,019 20,057 -1,038 18,485 320,084 Valle 35,547 17,284 18,249 -965 4,440 266,371 TOTAL 651,852 420,026 231,524 188,502 230,007 5,443,633 Table 2

Descriptive statistics, Colombia All municipalities By level of stuffed ballots By coercion

No coercion Co e r ci o n PercentileLow (N=273) High (N=273) (N=320) (N=68) Variable N mean sd 25th 50th 75th mean sd mean sd mean sd mean sd

Stuffed Ballots/ Adult Males 546 0.19 0.27 0.00 0.08 0.24 0.01 0.02 0.36 0.30 0.21 0.29 0.24 0.28 Log of 1+Stuffed Ballots/ Adult Males 546 0.15 0.20 0.00 0.07 0.22 0.01 0.02 0.29 0.20 0.17 0.21 0.20 0.20 Coercion Dummy 388 0.18 0.38 0 0 0 0.13 0.34 0.21 0.41 0 0 1 0

Share of population in each category, 1918 census (in percent) Clergy 546 0.05 0.09 0.01 0.03 0.06 0.06 0.10 0.04 0.08 0.04 0.06 0.08 0.16 Armed forces 546 0.06 0.35 0.00 0.00 0.00 0.07 0.39 0.05 0.30 0.05 0.39 0.14 0.39 Bureaucracy 546 0.51 0.36 0.31 0.43 0.62 0.55 0.41 0.47 0.29 0.49 0.35 0.55 0.37 Literate 546 28.8 11.9 20.1 27.1 36.2 31.9 12.4 25.7 10.4 28.9 12.0 28.8 12.3 Goes to school* 546 25.4 17.7 14.2 22.2 32.0 27.4 19.7 23.4 15.2 24.9 13.9 27.7 31.9 Vaccined 546 30.8 19.8 15.7 27.2 42.4 33.6 21.2 28.0 17.8 27.7 16.4 29.2 17.9

Date of foundation 530 1740 117 1627 1772 1840 1738 115 1742 120 1747 110 1709 121.8 Table 3

Sample Distribution by Department: municipalities with electoral data

Base sample: municipalities with occupation data Departament Freq. Percent

Antioquia 93 17.03 Arauca 3 0.55 Atlantico 18 3.3 Caldas 33 6.04 Caq u et a 1 0 . 1 8 Cauca* 32 5.86 Ch o co 9 1 . 65 Cundinamarca 107 19.6 Guaj i r a 1 0.18 Huila* 28 5.13 Meta 4 0.73 Narino* 46 8.42 Norte de Santander 30 5.49 Put umayo 4 0.73 San t an d e r 7 0 1 2 . 8 2 Tolima 38 6.96 Valle* 29 5.31 Total 546 100 Municipalities without occupation data Bolivar 49 Boyaca 111 Casan ar e 1 Guaj i r a 1 Magdalena 28 San Andres y Providencia 2 Total 192 Table 4

Matrix of pairwise correlations, Colombia

Log of 1+Stuffed Co e r ci o n Armed Dat e of Vaccination Ballots/Adult Males Dummy Clergy forces Bureaucracy foundation rate Literacy rate

Log of 1+Stuffed Ballots/ Adult Males 1

Co e r ci o n Du m m y 0.0574 1 (0.260)

Cl er gy -0.1037 0.1779 1 (0.015) (0.000)

Armed forces 0.0075 0.0945 0.3492 1 (0.861) (0.063) (0.000)

Bureaucracy -0.1138 0.0608 0.2646 0.1442 1 (0.008) (0.232) (0.000) (0.001)

Date of foundation 0.0317 -0.1288 -0.1651 -0.0831 -0.048 1 (0.467) (0.011) (0.000) (0.056) (0.270)

Vaccination rate -0.155 0.0339 0.2071 0.0928 0.0997 -0.1385 1 (0.000) (0.506) (0.000) (0.030) (0.020) (0.001)

Literacy rate -0.2377 -0.0018 0.3101 0.1399 0.297 -0.0159 0.4502 1 (0.000) (0.971) (0.000) (0.001) (0.000) (0.715) (0.000)

Schooling rate -0.1316 0.0593 0.2632 0.1076 0.1874 -0.1176 0.3563 0.5244 (0.002) (0.244) (0.000) (0.012) (0.000) (0.007) (0.000) (0.000) Table 5

Descriptive Statistics, Cundinamarca All municipalities By level of stuffed ballots By coercion No coercion Percentile Low (N=54) High (N=53) (N=81) Coercion (N=20) variable Nmeansdp25 p50 p75 mean sd mean sd mean sd mean sd

Elect oral Out comes Stuffed Ballts/ Adult Males 107 0.30 0.31 0.04 0.16 0.58 0.05 0.05 0.56 0.25 0.34 0.32 0.19 0.24

Log of (1+Stuffed Ballots/Adult Males) 107 0.24 0.23 0.03 0.15 0.46 0.04 0.05 0.43 0.16 0.26 0.23 0.16 0.19 Co e r ci o n Du m m y 101 0.20 0.40 0 0 0 0.27 0.45 0.13 0.34 0 0 1 0

Economic and Political Inequality Land Gini (average 1879, 1890) 97 0.65 0.10 0.59 0.66 0.72 0.67 0.09 0.63 0.10 0.64 0.11 0.68 0.07 Political Concentration Index (1875- 1895) 105 -0.55 0.10 -0.62 -0.56 -0.48 -0.53 0.10 -0.57 0.09 -0.54 0.09 -0.60 0.08 Average Overlap (Top 25%) 98 0.07 0.04 0.04 0.07 0.09 0.06 0.04 0.08 0.03 0.07 0.04 0.06 0.03

Share of population in each category, 1918 census (in percent) Cl e r gy 107 0.06 0.12 0.01 0.03 0.06 0.08 0.16 0.04 0.04 0.04 0.04 0.11 0.20 Armed Forces 107 0.10 0.59 0.00 0.00 0.00 0.16 0.81 0.04 0.19 0.09 0.65 0.14 0.36 Bureaucracy 107 0.43 0.38 0.23 0.36 0.58 0.49 0.48 0.37 0.23 0.37 0.22 0.52 0.45 Literate 107 26.0 10.1 19.1 23.6 32.9 28.3 11.7 23.6 7.6 24.3 8.9 29.3 12.1 Goes to school* 107 24.6 26.4 12.3 19.7 28.4 27.9 33.8 21.3 15.5 20.7 12.7 37.6 53.1 Vaccined 107 34.8 17.4 20.5 34.1 47.1 37.7 17.1 31.8 17.4 33.6 16.8 39.8 19.9

Date of foundation 107 1672 110 1592 1627 1777 1668 111 1676 109 1685 108 1652 113 Table 6 Matrix of pairwise correlation coefficients, Cundinamarca

Log of Political (1+Stuffed Concentration Average Ballots/Adult Co er ci o n Land Index (1875- Overlap (Top Armed Dat e of Vaccinat ion Li t er acy Males) dummy gini 1895) 25%) Clergy forces Bureaucracy foundation rate rate

Log of (1+Stuffed Ballots/Adult Males) 1

y -0.185 1 Co e r ci o n d u m m (0.064)

Land gi ni -0.278 0.132 1 (0.006) (0.212)

Political Concentration Index (1875-1895) -0.102 -0.241 -0.198 1 (0.299) (0.016) (0.053)

Average Overlap (Top 25%) 0.224 -0.072 -0.048 0.438 1 (0.027) (0.493) (0.641) (0.000) y -0.186 0.30 5 0.124 -0.117 -0.148 1 Cl er g (0.055) (0.002) (0.228) (0.236) (0.146)

Armed forces -0.104 0.032 0.096 -0.154 -0.093 0.188 1 (0.287) (0.753) (0.351) (0.118) (0.365) (0.053)

Bureaucracy -0.215 0.20 9 0.265 -0.101 -0.150 0.337 0.121 (0.026) (0.036) (0.009) (0.303) (0.141) (0.000) (0.215) 1

Date of foundation 0.140 -0.120 -0.173 0.159 -0.044 -0.216 0.048 -0.085 1 (0.152) (0.234) (0.090) (0.105) (0.667) (0.026) (0.626) (0.383)

Vaccination rate -0.247 0.142 0.394 -0.139 -0.206 0.348 0.18 9 0.159 -0.183 1 (0.011) (0.158) (0.000) (0.157) (0.042) (0.000) (0.052) (0.102) (0.059)

Literacy rate -0.277 0.206 0.357 -0.047 0.073 0.404 0.22 6 0.469 -0.299 0.47 7 1 (0.004) (0.039) (0.000) (0.636) (0.477) (0.000) (0.019) (0.000) (0.002) (0.000)

Schooling rate -0.1695 0.250.005 -0.054 -0.156 0.153 0.071 0.078 -0.147 0.11 7 0.322 (0.082) (0.010) (0.965) (0.583) (0.126) (0.116) (0.466) (0.423) (0.132) (0.230) (0.001) Figure 1

Colombia. Stuffed ballots and clergy 1 .8 .6 .4 .2 0

0 .002 .004 .006 .008 Clergy/Total population Figure 2

Colombia. Stuffed ballots and bureaucracy 1 .8 .6 .4 .2 0

0 .01 .02 .03 Bureaucracy/Total population Figure 3

Cundinamarca. Stuffed ballots and clergy .8

VillapinzonGama

UsmeViotaSan Juan de Rioseco Quebradanegra Fomeque .6 Puli GutierrezGuascaPaimeTenjo MachetaYacopiEl Penon LaUneGirardot Pena NimaimaLa Calera Chia ChipaqueVergaraSan Francisco LaUbaque VegaArbelaezSasaima BeltranChoachiFosca

.4 Alban Soacha SupataCoguaNocaima TibiritaSuescaPasca FusagasugaJuninUtica Sopo AnapoimaSesquileLa Palma MantaChocontaCajica .2 Tocaima La Mesa VilletaGuatavita UbalaGachalaQuetameBojaca Mosquera Caparrapi QuipileEl GachetaColegio TibacuyBituimaElLenguazaqueViani CarmenAnolaima de Carupa ChaguaniSutatausaSusaNarinoSanZipaconNemoconGachancipa AntonioTocancipaFacatativa Nilo Guacheta Guataqui Zipaquira Medina Suba 0 RicaurteTenaTausaJerusalenGuayabalEngativaPachoCaquezaGuaduasSanCucunubaFunzaPandiFuqueneFontibonSimijaca BosaCayetanoTabioMadrid deCota Siquima UsaquenUbate Bogota

0 .002 .004 .006 .008 Clergy/Total population Figure 4

Cundinamarca. Stuffed ballots and bureaucracy .8

VillapinzonGama ViotaUsmeSan Juan de Rioseco FomequeQuebradanegra .6 Puli Tenjo GuascaPaimeGutierrez ElMacheta PenonYacopi LaUne Pena Girardot ChipaqueNimaimaSanChia FranciscoLa Calera UbaqueVergaraArbelaezLa VegaSasaima FoscaChoachiBeltran

.4 Alban Soacha Cogua SupataNocaima Pasca Suesca JuninFusagasugaUtica Tibirita Sopo AnapoimaLa PalmaSesquile Manta ChocontaCajica .2 Tocaima La Mesa Guatavita UbalaBojacaQuetameVilleta Mosquera CaparrapiGachala El ColegioQuipileGacheta ElLenguazaqueAnolaima CarmenTibacuyViani de CarupaBituima Susa FacatativaGachancipaZipaconNemocon Narino ChaguaniGuachetaSutatausaSanNilo AntonioGuataquiTocancipaZipaquira Suba Medina 0 GuayabalCaquezaPachoPandiSanRicaurteBosa CucunubaTenadeCota CayetanoTabioGuaduas FunzaSiquimaSimijacaTausaMadridFuqueneUbateJerusalenUsaquenFontibon Bogota Engativa -.2 0 .01 .02 .03 Bureaucracy/Total population Figure 5

Cundinamarca. Stuffed ballots and land gini .8

Villapinzon

San Juan de RiosecoViotaUsme Quebradanegra Fomeque .6 Paime TenjoPuli Guasca El Penon Macheta La Pena Une Girardot Nimaima La Calera San FranciscoChia Vergara Chipaque Ubaque SasaimaLa Vega Fosca Choachi Beltran .4 Soacha NocaimaCogua TibiritaPasca Suesca Junin Utica Fusagasuga Sopo La Palma AnapoimaSesquile Manta Choconta Cajica .2 Tocaima La Mesa GuatavitaVilleta Mosquera Ubala Quetame Gachala Bojaca Caparrapi Gacheta ElQuipile Colegio Viani Bituima LenguazaqueEl Carmen de Carupa Sutatausa ChaguaniZipacon NarinoSusa GachancipaSanNemocon AntonioTocancipaFacatativa Nilo Guataqui Zipaquira Guacheta Suba

0 SanTausaCucunubaCaqueza CayetanoGuayabalPandi de TabioSiquimaUsaquenJerusalenEngativaPachoMadridUbateGuaduasSimijacaBosaCotaFuqueneFunzaFontibonTenaRicaurte .4 .5 .6 .7 .8 .9 Average and gini 1879, 1890 Figure 6

Cundinamarca. Stuffed ballots and political concentration .8

Villapinzon

San Juan deViotaUsme Rioseco Quebradanegra Fomeque .6 Gutierrez TenjoPuli Paime Guasca MachetaYacopi El Penon Girardot UneLa Pena San FranciscoChia La CaleraNimaima VergaraChipaque La VegaSasaima Ubaque Arbelaez Choachi FoscaBeltran .4 Soacha SupataNocaimaCogua Suesca PascaTibirita JuninFusagasugaUtica Sopo Sesquile AnapoimaLa Palma ChocontaCajica Manta .2 La Mesa Tocaima Villeta Guatavita Mosquera BojacaQuetameGachala Ubala Caparrapi Gacheta El Colegio Quipile Tibacuy ElLenguazaque Carmen de CarupaAnolaimaBituima Viani Facatativa Zipacon SanNemoconSusa AntonioNarinoTocancipa SutatausaGachancipaChaguani Zipaquira Nilo Guacheta Guataqui Medina Suba 0 FuqueneUsaquenUbateGuaduasMadridPachoCaquezaTabioGuayabalBogotaSimijacaJerusalenFontibon deFunza SiquimaSanTausaPandi CayetanoCucunubaRicaurteTenaCota Engativa Bosa -.8 -.7 -.6 -.5 -.4 -.3 Political concentration Index, 1875-1895 Figure 7

Cundinamarca Stuffed ballots and landowner-politician overlap .8 Villapinzon

San Juan deViota Rioseco Usme Quebradanegra Fomeque .6 Paime Tenjo PuliGuasca MachetaEl Penon Girardot La Pena Une San Francisco ChiaNimaima ChipaqueLa Calera La Vega UbaqueVergaraSasaima ChoachiFosca Beltran .4 Soacha Nocaima Cogua Tibirita Suesca Pasca Utica Junin Fusagasuga Sopo AnapoimaSesquile La Palma Choconta CajicaManta .2 La Mesa Tocaima Guatavita MosqueraQuetameVilleta Ubala Bojaca Caparrapi Gachala QuipileEl ColegioGacheta El Carmen de Carupa Viani AnolaimaLenguazaqueBituima SutatausaFacatativaSan AntonioZipaconGachancipaSusaTocancipaNemoconNarino Chaguani ZipaquiraGuacheta GuataquiNilo Suba 0 Fuquene CaquezaFunzaGuayabalUsaquenSimijacaMadridJerusalenPacho UbatedeCucunubaRicaurte SiquimaSanGuaduasFontibon CayetanoTabioPandiEngativaTausaBosaCota Tena 0 .05 .1 .15 .2 Average overlap, top 25% Table 7

Ballot Stuffing, coercion, and state presence in Colombia (1)(2)(3)(4)(5)(6)(7)(8)(9)(10) Panel A. Dependent Variable: Log of 1+Stuffed Ballots/ Adult Males

Cl e r gy -23.54* * * -21.22* * -20.57* * -22.14* * -16.74* -17.65* (5.849) (8.797) (8.425) (9.147) (8.715) (9.244) Armed forces -1.896 0.549 1.057 0.583 0.714 (3.517) (4.204) (4.822) (4.265) (4.088) Bureaucracy -4.383* * -2.95 -1.165 -3.348 -2.897 (2.110) (2.387) (2.667) (2.226) (2.412) Bureaucracy+Armed Forces -1.253 (2.095) Foundation Date 0.182* * (0.074) Vaccination rate -0.084 5 (0.054)

Observations 546 546 546 546 546 530 530 546 R-squared 0.17 0.16 0.16 0.17 0.17 0.18 0.19 0.18 Panel B. Dependent Variable: Coercion Dummy (=1 if any form of coercion)

Cl ey r g 79.17* * * 71.82* * * 71.45* * * 53.09* * 66.77* * * 70.66* * * 66.80* * * 71.41* * * (20.840) (21.490) (21.980) (26.870) (21.850) (21.720) (22.810) (22.030) Armed forces 6.969 1.17 -3.073 1.322 1.118 1.06 1.14 5 (7.886) (4.844) (2.411) (4.693) (4.858) (4.732) (4.824) Bureaucracy 10.66* 5.666 4.802 5.439 5.599 5.568 5.62 (5.872) (5.025) (5.762) (5.019) (5.021) (5.027) (4.997) Bureaucracy+Armed Forces 3.104 (3.844) Foundation Date -0.28 (0.189) Vaccination rate 0.0267 (0.114) Schooling rate 0.102 (0.151) Literacy rate 0.013 (0.180)

Observations 388 388 388 388 388 376 388 388 388 388 R-squared 0.10 0.07 0.08 0.10 0.10 0.07 0.11 0.10 0.10 0.10 Table 8 a (1)Ballot Stuffing, (2) coercion, (3) and state (4) presence (5) in Cundinamarc (6) (7) (8) (9) (10)

Panel A. Dependent Variable: Log of 1+Stuffed Ballots/ Adult Males Clergy -35.80* * * -22.69* * -30.76* * * -26.59* * * -17.97 -11.11 (8.138) (10.760) (11.660) (9.766) (10.890) (13.440) Armed Forces -3.982* * * -2.354 -2.704 -2.74 -1.421 (1.370) (1.925) (1.702) (2.019) (1.961) Bureaucracy -12.90* * * -10.06* * * -10.78* * * -9.934* * * -9.612* * (3.593) (3.416) (3.598) (3.490) (3.692) Bureaucracy+Armed Forces -4.555* * * (1.529) Foundation Date 0.224 (0.203) Vaccination rate -0.251* (0.151)

Observat ion s 107 107 107 107 106 107 107 107 R-squared 0.04 0.01 0.05 0.07 0.06 0.05 0.08 0.10 Panel B. Dependent Variable: Coercion Dummy (=1 if any form of coercion)

Clergy 124.4* * * 121.6* * * 145.0* * 131.5* * * 111.9* * 118.1* * * 103.8* * 110.1* * (30.400) (43.150) (67.600) (32.420) (46.130) (41.250) (45.770) (45.570) Armed Forces 2.095 -3.065 -2.794 -2.693 -3.390* -3.662* * -3.888* * (6.040) (1.862) (1.922) (1.945) (2.011) (1.665) (1.936) Bureaucracy 29.64* 4.042 4.129 6.238 2.766 6.538 1.689 (15.840) (18.520) (18.380) (18.670) (19.540) (18.220) (18.850) Bureaucracy+Armed Forces -2.132 (3.011) Foundation Date -0.255 (0.372) Vaccination rate 0.103 (0.246) Schooling rate 0.323* * * (0.096) Literacy rate 0.361 (0.474)

Observat ion s 101 101 101 101 100 101 101 101 101 101 R-squared 0.09 0.00 0.04 0.10 0.06 0.09 0.10 0.10 0.14 0.10 Table 9

Ballot Stuffing, economic and political concentration in Cundinamarca (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Panel A. Dependent variable: Log of 1+Stuffed Ballots/ Adult Males

Land gini -0.623*** -0.704*** -0.736*** -0.721*** -0.638*** (0.201) (0.199) (0.201) (0.204) (0.233) Political concentration index -0.236 -0.438* * -0.662* * * -0.819* * * -0.843* * * -0.836* * * (0.214) (0.209) (0.212) (0.201) (0.208) (0.207) Landowner-politician overlap (Top 25%) 1.343* * 2.077* * * 2.174* * * 2.216* * * 2.060* * * (0.651) (0.704) (0.630) (0.665) (0.652) Foundation Date 0.102 (0.196) Vaccination rate -0.152 (0.157)

Observations 97105989798979797 R-squared 0.08 0.01 0.05 0.11 0.11 0.22 0.22 0.23 Panel B. Dependent variable: Coercion Dummy (=1 if any form of coercion)

Land gini 0.543 0.275 0.268 0.275 0.233 0.289 0.151 (0.336) (0.318) (0.324) (0.335) (0.365) (0.317) (0.360) Political concentration index -1.031* * -1.177* * -1.372* * * -1.283* * * -1.311* * -1.278* * * -1.251* * -1.232* * (0.405) (0.461) (0.475) (0.486) (0.535) (0.484) (0.481) (0.472) Landowner-politician overlap (Top 25%) -0.791 0.624 0.599 0.641 0.63 0.933 0.453 (1.033) (1.035) (1.056) (1.120) (1.083) (1.038) (1.143) Foundation Date 0.0837 (0.476) Vaccination rate 0.0496 (0.284) Schooling rate 0.320* * * (0.076) Literacy rate 0.386 (0.511)

Observations 91 99 92 91 92 91 91 91 91 91 R-squared 0.02 0.06 0.01 0.08 0.08 0.08 0.08 0.08 0.13 0.09 Table 10 Ballot stuffing and coercion in Cundinamarca: state presence, and economic and political concentration (1) (2) (3) (4) (5) (6) (7) (8) (9) Panel A. Dependent variable: Log of 1+ Stuffed Ballots/ Adult Males

Land gini -0.714* * * -0.725* * * -0.676* * * -0.652* * * -0.669* * * -0.640* * * -0.573* * (0.204) (0.202) (0.211) (0.214) (0.208) (0.217) (0.247) Political concentration index -0.828* * * -0.845* * * -0.804* * * -0.838* * * -0.850* * * -0.861* * * -0.846* * * (0.196) (0.205) (0.199) (0.198) (0.198) (0.208) (0.203) Landowner-politician overlap (Top 25%) 2.107* * * 2.162* * * 2.078* * * 2.018* * * 2.045* * * 2.061* * * 1.935* * * (0.638) (0.633) (0.638) (0.648) (0.648) (0.688) (0.665) y -20.9 -18.1 5 -18.93 -16.49 -12.74 Cl er g (13.270) (13.360) (13.420) (13.730) (14.690) Armed Forces -2.775* * * -2.665* * -2.898* * -2.352* * (0.998) (1.096) (1.185) (1.138) Bureaucracy -6.054* -5.431 -5.382 -5.768 (3.470) (3.423) (3.379) (3.605) Armed Forces+Bureaucracy -3.384* * * (1.188) Foundation Date 0.0941 (0.206) Vaccination rate -0.127 (0.167)

Observations 97 97 97 97 97 97 97 R-squared 0.224 0.222 0.226 0.237 0.235 0.239 0.244

Panel B. Dependent variable: Coercion Dummy (=1 if any for of coercion) Land gini 0.09 0.289 0.172 0.109 0.134 0.134 0.129 0.114 0.0676 (0.302) (0.329) (0.331) (0.341) (0.313) (0.354) (0.373) (0.336) (0.375) Political concentration index -1.143* * -1.334* * * -1.257* * -1.196* * -1.192* * -1.261* * -1.197* * -1.169* * -1.183* * (0.472) (0.496) (0.482)5 (0.484)0.656 (0.489)0.588 (0.555)0.725 0.639(0.486) 1.043(0.477) (0.479)0.551 Landowner-politician overlap (Top 25%)(1.047) 0.678 (1.060) 0.577 (1.137) 0.78 (1.186) (1.072) (1.232) (1.198) (1.158) (1.316) yCl er g 122.2* 124.9* 132.4* * 129.0* 126.4* 114.0* 117.8* (64.560) (64.530) (63.330) (67.320) (64.080) (65.770) (66.760) Armed Forces -5.098* * -5.720* * * -6.159* * -5.639* * -5.986* * * -6.159* * * (2.135) (2.007) (2.362) (2.162) (1.898) (2.316) Bureaucracy 12.78 0.139 -1.499 0.328 4.517 -1.224 (24.460) (23.830) (24.710) (24.350) (23.240) (24.790) Armed Forces+Bureaucracy -5.079 (3.387) Foundation Date 0.184 (0.496) Vaccination rate -0.0349 (0.284) Schooling rate 0.318* * * (0.092) Literacy rate 0.21 (0.552)

Observations 91 91 91 91 91 91 91 91 91 5 R-squared 0.117 0.089 0.089 0.124 0.123 0.126 0.124 0.168 0.12

Appendix Table 1. Variable Sources and Description

Variable Description Source

Stuffed Ballots

Stuffed Ballots max{0, Total votes-Estimated Franchise}

Total votes Votes cast in each municipality in the 1922 Presidential elections. Telegrams sent by Consejos Electorales to the Gran Consejo Electoral. (National Archive, Asuntos Electorales)

Estimated franchise Estimated literate male population over 19 in 1918 multiplied by 1.09. This corresponds to an adjustment for a yearly growth rate of 2% in population.

Estimated literate male Male literacy rate*Male population over 19 years of age 1918 National Census population over 19 in 1918

Male literacy rate Literate male population to total male population ratio. Population with 1918 National Census unspecified literacy distributed proportionally between the two groups. Available at the municipality level except from the department of Guajira, where the departmental rate, of 231/10591 was applied to all municipalities.

Male population over Male population over 19 years of age with population with unspecified 1918 National Census 19 years of age age distributed proportionally between the two groups

Coercion

Coercion Dummy Indicator equals to 1 if there is indication of (i) violence, (ii) intimidation or harassment, or (iii) arms distribution, and zero otherwise. Details of events included in (i)-(iii) below.

violence If the town had reports of actual violence breaking out. For example, Telegrams sent by Consejos Electorales to the brawls, gun-shots hit their target, confrontations with injured, or Gran Consejo Electoral (National Archive, casualties. Basically, people-get-hurt kind of things Asuntos Electorales), and liberal complaints intimidation or If there are reports of incarcerating Liberals, subjecting them to random summarized in Los Partidos Políticos en harassment searches and detainment, or other coercive measures to prevent Liberal Colombia. propagandizing or activism, such as "pedreadas" (throwing stones) to Liberal houses or establishments. arms distribution If there are reports of organized armed Conservatives who are not police or army; distribution of arms for these bodies. Activities of intimidation by these bodies.

Appendix Table 1 continued. Variable Sources and Description

Other variables

Land gini in XIXth Century For Cundinamarca only. Average land gini from 1879 and 1890, calculated Acemoglu, Bautista, Querubin, and Robinson only for rural plots. (2008)

Political Concentration For Cundinamarca only. Political concentration for municipality m at time t Acemoglu, Bautista, Querubin, and Robinson Index (1875-1895) is contructed as the (negative) of the ratio of the number of different (2008) from the Diario Oficial de Cundinamarca. individuals in power in such place and time to the number of mayor appointments. Very low political concentration (-1) indicates as many mayors as appointment opportunities, and high levels of concentration (close to 0) indicate a low ratio of people in power to number of appointments. The index is for the period 1875-1975

Average Overlap (Top For Cundinamarca only. Overlap between landowners and politicians. Acemoglu, Bautista, Querubin, and Robinson 25%) Calculated as the ratio between the number of individuals who are both rich (2008) from the Diario Oficial de Cundinamarca and politicians to the number of individuals who are either rich or politicians. and the 1879 and 1890 cadastral records of Politicians include all mayors from 1875 to 1895, and rich people include Cundinamarca. those individuals whose land plots are in the top 25% most valuable plots.

Armed forces proportion Ratio of armed forces agents to total popualtion 1918 National Census

Clerical proportion Ratio of religious ministers to total population 1918 National Census

Bureaucracy proportion Ratio of public employees to total population 1918 National Census

Literacy rate Literate population to total population ratio. Population with unspecified 1918 National Census literacy distributed proportionally between the two groups.

Schooling rate Ratio of population attending school to population from 5 to 14 years of age 1918 National Census

Vaccination rate Ratio of vaccined population to total population 1918 National Census

Foundation date Earliest known foundation date of the municipality Bernand and Zambrano (1993)

Departmental dummies Dummies for each department. Excluded departments are those with just a few municipalities (Arauca, Caquetá, Casanare, Choco, Guajira, Meta, Putumayo, San Andrés y Providencia)