The Observer Effect In International Politics (PDF)

THE OBSERVER EFFECT IN INTERNATIONAL POLITICS: Evidence from a Natural Experiment Susan D. Hyde World Politics, Volume 60, Number 1, October 2007, pp. 37-63 (Article) Published by Johns Hopkins University Press DOI: https://doi.org/10.1353/wp.0.0001 For additional informat...

THE OBSERVER EFFECT IN INTERNATIONAL POLITICS: Evidence from a Natural Experiment Susan D. Hyde World Politics, Volume 60, Number 1, October 2007, pp. 37-63 (Article) Published by Johns Hopkins University Press DOI: https://doi.org/10.1353/wp.0.0001 For additional information about this article https://muse.jhu.edu/article/236122 [133.9.85.196] Project MUSE (2024-10-17 07:11 GMT) Waseda University Library  THE OBSERVER EFFECT IN INTERNATIONAL POLITICS Evidence from a Natural Experiment By SUSAN D. HYDE* I NTERNATIONAL actors now play a prominent role in domestic elections and other democratic processes throughout the developing world. They pressure governments to hold democratic elections and they directly engage in the electoral process through provision of tech- nical assistance and funding or by sending teams of observers to moni- tor elections. International influences on democratization are receiving a small but growing amount of scholarly attention, and international dimensions are increasingly accepted as relevant variables in theories of political transition.1 Most recently, in a study of the international diffu- sion of democratization, Kristian Gleditsch and Michael Ward “firmly reject the idea that institutional change is driven entirely by domestic * I wish to acknowledge valuable comments on previous drafts from Eric Bjornlund, Carew Bould- ing, Gary Cox, Don Green, Thad Dunning, Clark Gibson, Kristian Gleditsch, Peter Gourevitch, Da- vid Lake, Mat McCubbins, Irfan Nooruddin, Elizabeth Saunders, Sue Stokes, and the participants in several seminars. I also thank Anders Eriksson for making information available for this project. Any [133.9.85.196] Project MUSE (2024-10-17 07:11 GMT) Waseda University Library remaining errors are my own. I am grateful for research support from Yale University, the University of California’s Institute for Global Conflict and Cooperation, and the Brookings Institution. 1 For examples of this literature, see Paul W. Drake “The International Causes of Democratiza- tion, 1974–1990,” in Paul Drake and Matthew McCubbins, eds., The Origins of Liberty: Political and Economic Liberalization in the Modern World (Princeton: Princeton University Press, 1998); Kristian S. Gleditsch, All International Politics Is Local: The Diffusion of Conflict, Integration, and Democratiza- tion (Ann Arbor: University of Michigan Press, 2002); Kristian S. Gleditsch and Michael D. Ward, “Diffusion and the International Context of Democratization,” International Organization 60 (Octo- ber 2006); Steven Levitsky and Lucan Way, “International Linkage and Democratization,” Journal of Democracy 16 ( July 2005); Jon C. Pevehouse, “Democracy from the Outside-In? International Orga- nizations and Democratization,” International Organization 56 (August 2002); idem, Democracy from Above? Regional Organizations and Democratization (Cambridge: Cambridge University Press, 2005); Philippe C. Schmitter, “The Influence of the International Context upon the Choice of National Institutions and Policies in Neo-Democracies,” in Laurence Whitehead, ed., The International Dimen- sions of Democratization: Europe and the Americas (New York: Oxford University Press, 1996); Laurence Whitehead, ed., The International Dimensions of Democratization: Europe and the Americas (New York: Oxford University Press, 1996). For work specifically addressing international influences on elections, see Vikram K. Chand, “Democratisation from the Outside In: NGO and International Efforts to Pro- mote Open Elections,” Third World Quarterly 18 (September 1997); Jorgen Elklit and Palle Svensson, “What Makes Elections Free and Fair?” Journal of Democracy 8 ( July 1997); Liisa Laakso, “The Politics of International Election Observation: The Case of Zimbabwe in 2000,” Journal of Modern African Studies 40 (September 2002). World Politics 60 (October 2007), 37–63 38 W O R L D P O LI T I C S processes and unaffected by regional and international events.”2 Al- though scholars have documented several macrolevel relationships be- tween international-level variables and movement toward democracy, there has been little attention paid to the microlevel effects of interna- tional involvement in the democratization process. How do interna- tional actors influence democratization in developing countries? In examining this question, I focus on one form of democracy pro- motion: international monitoring of elections. Until 1962 there had been no recorded cases of international election observation in sover- eign states. By 2004 upwards of 80 percent of elections held in noncon- solidated democracies were monitored, and any leader of a developing country wishing to hold a legitimate election was expected to invite in- ternational election observers.3 Although the record of election obser- vation demonstrates that observers grew willing to condemn fraudulent elections over the course of the 1990s, it remains unknown whether international monitors can actually bring about cleaner elections, as proponents of election monitoring assert.4 Any cross-national study attempting to examine the domestic effects of international observers would be plagued by endogeneity problems. At the aggregate level it would be difficult to distinguish between an election that was clean because of the presence of international ob- servers and an election that would have been clean regardless of their presence. One may use counterfactuals to make a persuasive argument, but demonstrating causality using cross-national evidence would be nearly impossible. This is similar to the problem faced by scholars at- tempting to assess the independent effect of political institutions, and it underscores the uncertainty of efforts to promote democracy abroad. As Adam Przeworski argues, the important empirical question under- lying democracy promotion is “whether one can stick any institutions into some particular conditions and expect that they would function in the same way as they have functioned elsewhere.”5 It is not clear from existing evidence that democracy-promotion efforts encourage democ- ratization independent of what would have occurred in the absence of international involvement. This article offers a solution to this problem of causal inference by examining the effects of democracy-promotion efforts at the microlevel. 2 Gleditsch and Ward (fn. 1). 3 Author’s calculations. Evidence of the trend is presented in Eric C. Bjornlund, Beyond Free and Fair: Monitoring Elections and Building Democracy (Washington, D.C.: Woodrow Wilson Center Press, 2004). 4 Thomas Carothers, “The Observers Observed,” Journal of Democracy 8 ( July 1997). 5 Przeworski, “Institutions Matter?” Government and Opposition 39 (Autumn 2004), 528.  O B S ERV ER EF F E C T I N I N T ER NAT I O NA L P O LI T I C S 39 In order to examine the causal effect of international observers (who represent a prominent element of those efforts), I exploit a natural ex- periment in which observers were assigned to polling stations using a method that approximates randomization. The natural experiment provides a direct test of whether international observers influence the behavior of domestic political actors on election day. If the presence of international observers causes a reduction in election-day fraud, the ef- fect of observers should be visible at the subnational level by comparing polling stations that were visited by observers with those that were not visited. More specifically, if international monitoring reduces election- day fraud directly, all else held equal, the cheating parties should gain less of their ill-gotten vote share in polling stations that were visited by international monitors. Using polling-station-level election results from the 2003 presidential elections in Armenia, the data show that in this case international observers caused a reduction in the vote share of the incumbent candidate who was widely expected to steal the election. Thus, by examining microlevel electoral data and a direct measure of international presence, this article is able to test for the effects of inter- national scrutiny in a way that country-level studies are unable to do because of endogeneity problems. Why might scholars of international relations and comparative poli- tics care about the effects of international election monitors? Elections are obviously central in theories of democracy and democratic gover- nance.6 As William Riker writes, “The essential democratic institu- tion is the ballot box and all that goes with it.”7 When elections are corrupted, voting as an “instrument of democracy” fails, the “institu- tionalized uncertainty” of democracy is lessened, and in extreme forms, elections no longer function as an effective method of representative government.8 Corrupt elections are not unique to the developing world or the third wave of democratization.9 However, as the formal institutions of de- mocracy have spread throughout the world, issues of election quality have 6 Robert A. Dahl, Polyarchy (New Haven: Yale University Press, 1971); Larry J. Diamond, “Is the Third Wave Over?” Journal of Democracy 7 ( July 1996); G. Bingham Powell, Jr., Elections as Instruments of Democracy: Majoritarian and Proportional Visions (New Haven: Yale University Press, 2000); Adam Przeworski, Democracy and the Market: Political and Economic Reforms in Eastern Europe and Latin America (New York: Cambridge University Press, 1991); Joseph A. Schumpeter, Capitalism, Socialism and Democracy (New York: Harper, 1942). 7 William H. Riker, Democracy in the United States (New York: Macmillan Company, 1965), 25. 8 Przeworski (fn. 6); Powell (fn. 6). 9 Gary W. Cox and J. Morgan Kousser, “Turnout and Rural Corruption: New York as a Test Case,” American Journal of Political Science 25 (November 1981); Fabrice E. Lehoucq and Iván Molina, Stuff- ing the Ballot Box: Fraud, Electoral Reform, and Democratization in Costa Rica (Cambridge: Cambridge University Press, 2002). 40 W O R L D P O LI T I C S become increasingly important for democracy activists and scholars. Fraudulent elections violate the principle of political equality argued to be fundamental to democratic governance,10 and they reduce rep- resentation and accountability, two of the most widely touted benefits of democracy. Thus, the efficacy of international efforts to bring about clean elections in countries that do not already experience them should be of interest to students of democratization. Recent theoretical and empirical research also points to international observers as potentially important actors in electoral revolutions. James Fearon argues that evaluations of elections by international observers may help facilitate popular enforcement of democracy by improving the probability of popular coordination in defense of democracy.11 By providing a possible solution to coordination problems, the reports of international observers may encourage the collective uprising of voters following a fraudulent election and may have the potential to lead to alterations in power even when elections are stolen. Additionally, findings about the consequences of international mon- itoring are relevant to debates about the effects of international institu- tions, particularly the call for research as to how international institu- tions matter,12 and about the consequences of international pressure within the broader research agendas on the second-image reversed and two-level games.13 Sovereign leaders of developing countries willingly invite international monitors to judge their elections. If election obser- vation is costless to the state leaders who invite them, this action is easy to explain as a form of cheap talk. However, if election observers cause (cheating) incumbent leaders to gain fewer votes, the puzzle of why leaders invite observers is more interesting and speaks to the impact of international pressure on the behavior of state leaders. Finally, policymakers also care about the effects of international ob- servers. There is growing skepticism among practitioners that democ- racy-promotion activities, including election monitoring, are having the effects intended by donors and democracy activists. As Carlos Santiso 10 Robert A. Dahl, On Democracy (New Haven: Yale University Press, 1998). 11 James D. Fearon, “Self-enforcing Democracy” (Manuscript, Institution of Governmental Stud- ies, University of California, Berkeley, April 2006). This argument is generally based on Przeworski (fn. 6) and Barry R. Weingast, “The Political Foundations of Democracy and the Rule of Law,” Ameri- can Political Science Review 91 ( June 1997). See also Joshua A. Tucker, “Enough! Electoral Fraud, Collective Action Problems, and Post-Communist Colored Revolutions,” Perspectives on Politics 5 (September 2007). 12 Beth A. Simmons and Lisa L. Martin, “Theories and Empirical Studies of International Institu- tions,” International Organization 52 (October 1998). 13 Peter G. Gourevitch, “The Second Image Reversed,” International Organization 32 (Autumn 1978); Robert D. Putnam, “Diplomacy and Domestic Politics: The Logic of Two-Level Games,” In- ternational Organization 42 (Summer 1988).  O B S ERV ER EF F E C T I N I N T ER NAT I O NA L P O LI T I C S 41 writes, “The initial enthusiasm within the international donor com- munity is giving way to increasing skepticism and even frustration with the pace and depth of democratic transitions.”14 However, part of the perceived problem of democracy assistance is likely due to the fact that the field is, as Thomas Carothers has described it, “understudied and poorly understood.”15 Macrolevel analyses of the relationship between democracy scores and levels of democracy aid cannot reveal which pro- grams work and which do not, nor whether efforts by international actors to encourage democratization have the intended effects. Evalua- tion of the effects of democracy-promotion activities, such as that pre- sented here, can aid in designing more effective programs, while also contributing to academic understandings of international influences on domestic politics. This article first presents a theory of how international observers influence election fraud and then provides background information on politics in Armenia, details surrounding the 2003 presidential election, and a description of the election-observation mission. It then describes the data, the use of natural experimental research design, and the test of whether international observers reduce the rate of election-day fraud. Finally, the article presents the results indicating that, in the case of the 2003 presidential elections in Armenia, international observers re- duced election-day fraud by about 6 percent in the polling stations they visited during the first round of the election and that the presence of international observers had statistically significant but nuanced effects [133.9.85.196] Project MUSE (2024-10-17 07:11 GMT) Waseda University Library in the second round. OBSERVERS AND ELECTION FRAUD Understanding how international observers influence election fraud requires paying attention to the incentives of those leaders who invite international observers and cheat in front of them. Today’s autocrats who choose to hold elections face a conflict between pressure to hold internationally certified elections and a desire to guarantee their hold on power. Although some formerly autocratic countries now allow democratic elections with broad participation on a level playing field, a number of leaders still try to have it both ways, either betting that they will not be caught cheating or attempting to conceal manipulation 14 Carlos Santiso, “International Cooperation for Democracy and Good Governance: Moving to- ward a Second Generation?” European Journal of Development Research 13 ( June 2001). 15 Thomas Carothers, Aiding Democracy Abroad: The Learning Curve (Washington, D.C.: Carnegie Endowment for International Peace, 1999), 8. 42 W O R L D P O LI T I C S of the election from international observers. Theoretically, observers should increase constraints on these leaders by increasing the costs of committing election fraud, thus making it marginally less likely that elections will be stolen outright. It is unlikely that election fraud can be eliminated. But by making outright theft of elections more costly to carry out and more likely to be criticized, international election observ- ers may improve the quality of electoral governance and contribute to democratization. How might observers influence election fraud? One possibility is that observers have no effect on fraud. There are leaders, such as Rob- ert Mugabe of Zimbabwe, who invite international observers and cheat openly. Other leaders, who behave as though they fear being caught by observers, engage in types of electoral manipulation that seem less likely to be caught by international observers.16 An unlikely possibility is that inviting observers eliminates election fraud. Were this the case, international observers would never witness fraud, but the record of election observation clearly demonstrates that this is not so.17 Observ- ers have encountered various types of electoral manipulation, including military intimidation of voters, ballot-box stuffing, improper attempts to influence voters inside the voting booth, vote-buying schemes, in- tentional inflation of the vote tallies, jailing of opposition voters, fail- ure to distribute ballots to opposition strongholds, and manipulation of voter-registration lists.18 The most intuitively plausible theory is that the effects of observers are not uniform across elections. In some cases, observers have no ef- fect, while in others they may increase the risks associated with election manipulation, thus decreasing the likelihood that domestic actors will attempt manipulation. Andreas Schedler outlines a number of possible methods of electoral manipulation, ranging from direct election-day electoral fraud such as ballot-box stuffing, to longer-term forms of elec- toral bias such as gerrymandering.19 In theory, international election observers or other monitors could have a deterrent effect on any form of election fraud, but testing for this type of effect is difficult. In order to gain empirical traction, this article limits the focus to the manipula- tion that occurs in and around the polling station on election day. Even 16 Mikhail Myagkov, Peter C. Ordeshook, and Dimitry Shakin, “Fraud and Fairytales: Russia and Ukraine’s Electoral Experience,” Post-Soviet Affairs 21(April–June 2005). 17 Bjornlund (fn. 3); Carothers (fn. 4); Carothers (fn. 15). 18 The statement is made based on the reading of numerous international observer reports citing firsthand evidence of election fraud. 19 Schedler, “The Nested Game of Democratization by Elections,” International Political Science Review 23 ( January 2002).  O B S ERV ER EF F E C T I N I N T ER NAT I O NA L P O LI T I C S 43 within this subcategory of electoral manipulation, election-day fraud takes many forms, all of which share the characteristic of increasing the vote share for the fraud-sponsoring candidate(s). It is my conten- tion that the presence of international election observers may directly influence the behavior of actors inside and in the immediate vicinity of polling stations. If officials at polling stations and other domestic actors feared being caught manipulating the election, they would be less likely to engage in manipulation in the physical presence of inter- national observers. I thus argue that if election-day fraud is occurring and if international observers reduce election fraud, then the presence of observers in a polling station should have a measurable effect on the average vote share of the fraud-sponsoring candidate(s). The following section provides background on the political climate in Armenia leading up to the 2003 election; this should aid in evaluat- ing the natural experimental research design. POST-SOVIET POLITICS IN ARMENIA Between independence in 1991 and the 2003 presidential election, Ar- menia held six elections. Political participation by voters and most can- didates generally complied with democratic standards, but the execu- tive office participated in elections in a manner that has been labeled “flagrantly undemocratic.”20 Following independence, the elected presi- dent Levon Ter-Petrossian and his supporters successfully consolidated power within the executive office while other nascent parties were still attempting to gain organizational strength.21 As a result, the president is generally the controlling force in Armenian politics, by virtue of his authority to dissolve parliament, appoint all judges, and declare martial law.22 Strong political parties did not develop as a challenge to execu- tive power, in part because of Ter-Petrossian’s overt efforts to prevent any such opposition party from organizing.23 As of 2003 there were more than one hundred registered political parties. Because success- ful election-day manipulation requires at least minimal organizational capacity, the political structure in Armenia points to the incumbent executive as the political actor with the preponderant ability to commit widespread fraud. 20 Ian Bremmer and Cory Welt, “Armenia’s New Autocrats,” Journal of Democracy 8 ( July 1997), 78. 21 Ibid. 22 Larry J. Diamond, Developing Democracy: Toward Consolidation (Baltimore: Johns Hopkins Uni- versity Press, 1999), 55. 23 Bremmer and Welt (fn. 20). 44 W O R L D P O LI T I C S The two most prominent political figures in the postindependence period are Ter-Pertrossian and Robert Kocharian. The former was president until 1998, when he resigned amid wide public dissatisfaction over his failure to increase the standard of living and his willingness to negotiate with Azerbaijan over the territorial conflict in Nagorno- Karabakh. The latter, who was elected to replace Ter-Pertrossian in 1998, was the incumbent candidate in the 2003 presidential election. The 2003 elections were viewed as a potential turning point for Ar- menian democracy. As an OSCE/ODIHR official report states: The election provided an important test of the progress of democratic practices in Armenia, since the previous presidential elections were characterized by seri- ous flaws and generally failed to meet international standards. Issues of con- cern at the two previous presidential elections... included inaccuracy of voter lists, shortcomings in the election administration, media bias, abuse of State resources, flawed voting by military personnel, the presence of unauthorized persons during polling and counting and discrepancies in the vote count.24 What variables correlate with political divisions in Armenian poli- tics? There are few demographic or other variables, such as ethnic di- visions, partisan registration, or support for particular issues, that can be used as predictors of political support. Unlike a number of other post-Soviet states, ethnic divisions play little role in internal Arme- nian politics. The parties do not have ethnic affiliations, as 96 percent of residents are ethnic Armenian, a homogeneity that simplifies the analysis of the patterns of vote share for the candidates. The only issue during the 2003 elections was the presidential race, with nine candidates on the ballot. The incumbent president Robert Kocharian was the front-runner; he faced a serious challenge from Ste- pan Demirchian, the son of the late speaker Karin Demirchian, who had been killed in a 1999 attack on parliament.25 The other notable challenger was Artashes Geghamian, the last Soviet-era mayor of the capital city of Yeveran. The ongoing conflict with Azerbaijan over the Nagorno-Karabakh region has been the single most important postindependence issue in Armenian politics. Kocharian, a native of Nagorno-Karabakh, is seen as a resolute supporter of independence for Nagorno-Karabakh. Ter- Petrossian’s willingness to negotiate with Azerbaijan over the territory 24 OSCE/ODIHR, Final Report on Presidential Elections in Armenia, 19 February and 5 March 2003 (Warsaw: Office for Democratic Institutions and Human Rights, 2003). 25 Armenian politics are characterized by violence, which overshadowed the 2003 elections. Most notably, in 1999, the parliament was attacked by gunmen, and eight prominent politicians were assas- sinated. The 2003 presidential elections were the first to be held after the attack.  O B S ERV ER EF F E C T I N I N T ER NAT I O NA L P O LI T I C S 45 in 1998 was partly responsible for his resignation from the presidency and Kocharian’s succession to his post through the 1998 special elections. Kocharian, who did not have his own political party, officially ran as an independent. He had been supported by a shifting coalition, which in 2003 included the ruling Republican Party of Armenia and the Arme- nian Revolutionary Federation (also known as “Dashnak” or the Social- ist Party). He also enjoyed the strong support of the military. Although his resolute unwillingness to negotiate on the Nagorno-Karabakh con- flict was his most defining characteristic, he also campaigned in 2003 on the promise of economic stability, as did all of the candidates. Thus, the relative homogeneity of Armenian politics, the lack of other issues on the ballot, the fact that the incumbent ran without a political party, the presence of a dominant executive, and the central issue of Nagorno- Karabakh provide the background to the 2003 presidential election and the context of the natural experiment. The first round of the 2003 presidential elections in Armenia took place on February 19 followed by a runoff on March 5. The Armenian constitution requires a second-round runoff if no presidential candi- date garners more than 50 percent of the vote in the first round in the single-district national election. The official first-round vote share for Kocharian was 49.48 percent, thus triggering a runoff election. Several months prior to the election, the Armenian Ministry of Foreign Affairs invited the OSCE/ODIHR to sponsor an international election-observation mission. The delegation included members of the [133.9.85.196] Project MUSE (2024-10-17 07:11 GMT) Waseda University Library Parliamentary Assembly Council of Europe. In the first round of the election, the OSCE deployed 233 observers from 35 OSCE participating states. The second round was observed by 193 short-term observers from 21 participating states. THE NATURAL EXPERIMENT RESEARCH DESIGN IN ARMENIA The use of experimental methods has experienced renewed popular- ity in political science and economics. The distinguishing characteristic of experimental methods versus observational research is that the cen- tral independent variable, or “treatment” variable, is randomly assigned. In field experiments such as those conducted recently by Gerber and Green, Miguel and Kremer, Olken, and Wantchekon, the researcher su- pervises the random assignment of the treatment variable.26 In natural 26 Alan S. Gerber and Donald P. Green, “The Effects of Canvassing, Phone Calls, and Direct Mail on Voter Turnout: A Field Experiment,” American Political Science Review 94 (September 2000); Edward Miguel and Michael Kremer, “Worms: Identifying Impacts on Education and Health in the 46 W O R L D P O LI T I C S experiments, the researcher does not manage the assignment of the treatment variable, but natural experiments can occur when the vari- able is assigned “as-if ” the assignment were random.27 The burden in natural experiments rests on the researcher to provide evidence that the treatment can, in fact, be treated “as-if ” it had been randomly assigned. Existing natural experiments vary in the degree to which the treatment approaches true randomization.28 There are few published natural or field experiments within compara- tive politics and fewer within international relations. Exceptions within comparative politics include the work of Guan and Green; Humphreys, Masters, and Sandbu; and Wantchekon.29 Despite the relative scarcity of experimental work, the advantages are well established: properly conducted experimental work is one of the few means by which causal inference can be tested.30 Stated more forcefully, experiments have an “unrivaled capacity to demonstrate cause and effect.”31 I present a natural experiment in which international observers were assigned to polling stations on election day using a method that I did not supervise but that comes very close to random assignment. This natural experimental setup allows a test of whether international ob- servers reduced the rate of election-day fraud during the 2003 presi- dential elections in Armenia. If election-day fraud occurs in any elec- tion, it should have the observable implication of increasing the vote share of the fraud-sponsoring candidate. In the case of Armenia, the incumbent sponsored the majority of election-day fraud. Therefore, if international observers have no effect on election-day fraud, then Presence of Treatment Externalities,” Econometrica 72 ( January 2004); Benjamin Olken, “Monitoring Corruption: Evidence from a Field Experiment in Indonesia,” Journal of Political Economy 115 (April 2007); Leonard Wantchekon, “Clientelism and Voting Behavior: Evidence from a Field Experiment in Benin,” World Politics 55 (April 2003). 27 For examples of natural experiments, see John Snow, On the Mode of Communication of Cholera (London: Churchill, 1849); and Sebastian Galiani and Ernesto Schargrodsky, “Property Rights for the Poor: Effects of Land Titling” (Manuscript, March 16, 2006). 28 Thad Dunning, “Improving Causal Inference: Strengths and Limitations of Natural Experi- ments” (Paper presented at the annual meeting of the American Political Science Association, Wash- ington, D.C., August 31–September 4, 2005). 29 Mei Guan and Donald P. Green, “Noncoercive Mobilization in State-Controlled Elections: An Experimental Study in Beijing,” Comparative Political Studies 39 (December 2006); Macartan Humphreys, William A. Masters, and Martin E. Sandbu, “The Role of Leaders in Democratic De- liberations: Results from a Field Experiment in São Tomé e Príncipe,” World Politics 58 ( July 2006); Wantchekon (fn. 26). 30 Alan S. Gerber, Donald P. Green, and Edward H. Kaplan, “The Illusion of Learning from Observational Research,” Working Paper (New Haven: Institution for Social and Policy Studies, Yale University, 2002). 31 James N. Druckman, Donald P. Green, James H. Kuklinski, and Arthur Lupia, “The Growth and Development of Experimental Research in Political Science,” American Political Science Review 100 (November 2006), 627.  O B S ERV ER EF F E C T I N I N T ER NAT I O NA L P O LI T I C S 47 the incumbent should perform equally well in both groups of polling stations: those that were monitored and those that were not. If inter- national observers reduce election fraud, the incumbent’s average vote share should be lower in monitored polling stations than in unmoni- tored polling stations. There are three unique features of the 2003 Armenian elections that allow a test of whether the presence of international observers re- duced election-day fraud. First, widespread and centrally orchestrated fraud occurred on election day. As the Economist described it, the 2003 election was “one of the dirtiest even Armenians can remember.”32 As stated, fraud (and therefore fraud deterrence) can occur at many points in the electoral process. However, it would be difficult to test for an election-day deterrent effect if no fraud occurred. Eyewitness reports from international observers, domestic observers, and journalists documented many varieties of election-day fraud. The OSCE/ODIHR observed “significant irregularities” in more than 10 per- cent of the polling stations they visited, the most blatant of which were ballot-box stuffing, “carousel” voting, direct vote buying, individuals vot- ing more than once, the intimidation of witnesses for political parties, the presence of government officials inside polling stations who attempted to intimidate officials and voters, and one isolated incident of the removal of more than fifty passports from a polling station by a policeman. During the counting process there were numerous attempts to change the vote totals by the polling-station officials, and observers recorded additional evidence of blatant ballot-box stuffing. In some cases the international observers were prevented from observing the counting process, which was interpreted as an attempt to conceal illicit behavior. The second characteristic of the 2003 presidential elections is that the Armenian Central Election Commission made disaggregated election results available. The process of recording and making public polling-station-level election results requires a certain level of admin- istrative competence and transparency that is not always present, even in developed democracies. In countries that experience significant amounts of electoral fraud, these data are often intentionally “lost” or kept private. The Armenian election data, disaggregated to the level of the polling sta- tion, were made public by the election commission on their Web site.33 The third and most important favorable feature of the 2003 Arme- nian elections is that the international observers were assigned in a way 32 Economist, “Democracy, It’s Wonderful,” February 22, 2003. 33 Government-reported election results were made available online at http://www.elections.am by the Central Election Commission of Armenia. 48 W O R L D P O LI T I C S that approximates random assignment. I learned from conversations with staff and participants in the OSCE observation mission to Armenia that the method used to assign observers to polling stations was func- tionally equivalent to random assignment. This permits the use of nat- ural experimental design. Although the OSCE/ODIHR mission did not assign observers using a random numbers table or its equivalent, the method would have been highly unlikely to produce a list of assigned polling stations that were systematically different from the polling sta- tions that observers were not assigned to visit. Each team’s assigned list was selected arbitrarily from a complete list of polling stations. Those making the lists did not possess information about polling-station at- tributes that would have allowed them to choose polling stations ac- cording to criteria that could have predicted voting patterns. Since the validity of this natural experiment rests upon this point, I will take some time to support it. In this particular election the delega- tion leaders gave each team of short-term observers a preassigned list of polling stations to visit during election day. These lists were made with two objectives in mind: (1) to distribute the observers throughout the entire country (including rural and urban areas) and (2) to give each observer team a list of polling stations that did not overlap with that of other teams. Observers were encouraged to go only to those polling stations on their list and to travel between polling stations in a way that minimized travel time and still ensured coverage of their assigned polling stations. It is important for the validity of the natural experi- ment that the travel routes would not have been predictable by external observers, including government officials. In the Armenian case, individuals who made these lists had little knowledge of polling-station characteristics other than their general geographic location. If the assignment of observers had been done with consideration to other variables that might be correlated with perfor- mance of the incumbent candidate, then the assignment could not be considered near random. In addition, the preceding discussion of Ar- menian politics indicated few observable characteristics of the popu- lation (such as socioeconomic status or ethnicity) that might be cor- related with the incumbent’s popularity. In this case, the staff did not have access to disaggregated data on the demographic characteristics of the Armenian voting population. OSCE/ODIHR staff members have assured me that they had no desire to (and did not) choose polling sta- tions on any basis other than the two criteria cited above. In addition, even if this were not true, it is highly unlikely that the mission’s office had the capability of choosing polling stations that were more or less  O B S ERV ER EF F E C T I N I N T ER NAT I O NA L P O LI T I C S 49 likely to favor the incumbent or the opposition candidates or that were more likely to experience election-day fraud.34 The fact that Armenian politics is not predictable along partisan or demographic lines under- scores that this type of selection bias in the assignment of international observers would have required enormous effort, access to data that do not exist, and foresight about the trajectory of Armenian politics that would be unusual for foreigners to possess. Additionally, assigning specific polling stations to each team cut out much of the agency on the part of individual observer teams that, in the absence of a directive, could choose to visit polling sites based on their own selection criteria within a defined geographic area. When observers are given leeway in choosing polling stations, the two most common alternative selection criteria (based on information from ob- servation missions outside of Armenia) are to choose polling stations that are either “convenient” or “interesting.” Each of these decision criteria may create significant observation bias. This has been pointed out as a problem by several critics of election observation.35 Observer teams that select “interesting” polling stations typically go to areas in which problems are expected, so that teams using this decision rule may disproportionately observe and report irregularities. This is a common strategy among the more ambitious and enthusiastic international ob- servers but was discouraged in this particular case. Observer teams that go to “convenient” areas are criticized for being electoral tourists. This was more common in early election observa- [133.9.85.196] Project MUSE (2024-10-17 07:11 GMT) Waseda University Library tion, although it seems to have continued as a favorite point to criticize in other observer groups. Other “convenient” selection methods may be observing near the observers’ hotel in the most comfortable urban areas or going to polling stations that are near tourist destinations. Clearly, these selection criteria are nonrandom and could lead to bias in both the observers’ reported observations and the natural experiment proposed here, particularly because a clever politician could recognize the tendency of observers to stick to certain areas and therefore con- centrate any electoral manipulation in places where observers would be unlikely to go. For these reasons it was particularly important for this natural experiment that this type of observer agency was explicitly discouraged. 34 Even if this information were inaccurately communicated to me, if observers were more likely to visit stations they believe to be problematic, as suggested by one reviewer, then this would dampen an observed effect of observers on fraud. For the reasons cited, however, this is an unlikely scenario. 35 Carothers (fn. 4); Gisela Geisler, “Fair? What Has Fairness Got to Do with It? Vagaries of Election Observations and Democratic Standards,” Journal of Modern African Studies 31 (December 1993). 50 W O R L D P O LI T I C S In sum, the assignment of international observers to polling stations for both rounds of the 2003 presidential elections in Armenia is very close to random. The selection was made arbitrarily from a list of all polling stations with only geographic logistics in mind, and the as- signment was completed with no knowledge of variables that might be correlated with the incumbent’s likely vote share. Teams were in- structed to visit only the polling stations assigned to them, and, due to the relatively small geographic area and limited number of polling stations assigned to each team, they had a high probability of reaching their assigned polling stations. There are a few other points about election-observation methodol- ogy that are not unique to Armenia but that are important components of the validity of this natural experimental design. First, international observers do not preannounce which polling stations they will observe on election day. Keeping deployment plans confidential is standard practice for reputable international observer groups and is intended to enhance the safety of the observers by making it difficult for attackers to anticipate where their potential targets will be. It also makes it diffi- cult for the candidates to anticipate the arrival of observers and thereby restrict their cheating to polling stations where international observers are not expected. Second, international observers were mobile, moving from polling station to polling station throughout election day. During the course of one day, an observer team could visit ten to twenty polling stations based on the length of the election day and the distance between poll- ing stations. Critics of election observation are fond of pointing out that it would be very difficult to catch any irregularities in twenty minutes of observation. However, if there are ongoing problems or “red flags” indicating that there might be problems, observer teams are instructed to stay for as long as they think is useful, which in some cases can be as long as several hours; however, they are not permitted to interfere in the electoral process. The questionnaires filled out by international observers include a number of observations related to the structure of the polling station, the available staff and materials, and the order of voting procedures that are immediately obvious. If the seal on the bal- lot box has been broken, international observers are most likely to see this and other evidence of fraud as soon as they enter a polling station. In addition, the partisan witnesses in each polling station remain in the same polling station throughout the day and are often able to report irregularities to the international observers.  O B S ERV ER EF F E C T I N I N T ER NAT I O NA L P O LI T I C S 51 DATA AND RESULTS The central measurable effect of observers on election-day fraud is to de- crease the vote share for the incumbent. Holding all else constant, if international observers did in fact reduce fraud at the polling stations they visited, then the incumbent should perform worse in observed polling sta- tions. Random assignment (or “as-if ” random assignment) of the treat- ment of international observers is equivalent to holding all else constant. Because international observers can be considered randomly assigned to polling stations and because there were two rounds of the presidential election, the natural experimental design involves two rounds of “treat- ment” and a separate observation of vote share for each round. Data on vote share for Kocharian are reported.36 The international observ- ers went to different polling stations in each round of the election but had some overlap between rounds. This divides the sample of polling- station-level election results into four experimental groups based on the treatment of international observation during the course of election day: one group of polling stations was never monitored (N=755), one group was monitored only in the first round (N=385), one was moni- tored only in the second round (N=260), and one group was monitored in both rounds (N=363). Groups of polling stations received all possible combinations of the international observer treatment, including no treatment in either round of the election. Therefore the natural experiment also allows a test of whether first-round observation has any lasting effect in the second round. Approximately 43 percent of polling stations were not observed in either round of the election, and about 21 percent were observed in both rounds.37 THE EFFECT OF MONITORS ON VOTE SHARE The dependent variable is the vote share for the incumbent presidential candidate, Kocharian. Depending on the comparison being made, three different forms of Kocharian’s vote share are used. For tests that compare groups separated by a difference in first-round treatment, the depen- dent variable is the mean vote percentage for Kocharian in the first round. For tests between groups separated by second-round treatment, 36 The same tests for the other candidates are available upon request. They do not change the conclusions drawn from these results. 37 Outside of the Yerevan region (where polling stations were equally likely to be visited in both rounds) polling stations that were visited in round 1 were twice as likely to be visited again in round 2. 52 W O R L D P O LI T I C S the dependent variable is the mean vote for Kocharian in the second round. When appropriate, I also make comparisons using the average vote for Kocharian across both rounds as the dependent variable. A series of two-sample difference of means tests examines whether in- ternational observers reduce fraud and, if so, to what degree. A difference of means test, or t-test, compares two groups of observations and tests the hypothesis that the mean of the two groups is identical. The test assumes equal variances between the two groups, computes the result in terms of a specific confidence interval (in this case 95 percent). The results in Table 1 present the difference of means tests and show that inter- national observers influenced election-day manipulation in the polling stations that they observed.38 To illustrate, the first comparison shown in Table 1 is the average vote share received in round 1 by Kocharian in unobserved polling stations compared with the average vote share for Kocharian in all observed polling stations in round 1. The parenthetical material in columns 1 and 2 refers to whether the percentage is in terms of Kocharian’s round 1 vote share, his round 2 vote share, or an average across the group in both rounds. The last column presents the differ- ence in Kocharian’s vote share between the two groups and indicates whether the difference is statistically different from zero. The results presented in Table 1 show clear evidence that during the 2003 presidential elections in Armenia, the presence of international observers reduced the vote share for the incumbent politician by about 5.9 percent in the first round and by more than 2 percent in the second round. Both results are statistically significant. This allows a rejection of the null hypothesis that there is no difference between observed and unobserved polling stations. Figure 1 illustrates this difference with a kernel density plot of Kocharian’s round 1 vote share in monitored and unmonitored polling stations. Note the unusual distribution of vote share in unmonitored polling stations. In addition to testing whether international observers deterred elec- tion-day fraud (and the degree to which they did so), the difference of means tests also show whether the presence of international observers had a “transitory” or “persistent” effect. A transitory effect is similar to the effect of a painkiller on a headache. The treatment affects only one headache and has no effect on subsequent headaches. A persistent treatment has a lasting effect, such as an injection to relieve frequently occurring migraines. In the case presented here in which there were two rounds of the election, it appears that a visit by monitors in the first 38 Additional tests compare vote share for all candidates and show that the only candidate to per- form better in unmonitored polling stations was Sargsian, who received only 0.04 percent more of the vote in unmonitored polling stations.  TABLE 1 DIFFERENCE OF MEANS TESTS COMPARING “TREATMENT” AND “CONTROL” GROUPS a Average Incumbent Vote Average Incumbent Vote Share among Polling Stations Share among Polling Stations That Were... vs. That Were... Difference 1. Not observed in R1 vs. Observed in R1 5.9% 54.2% 48.3% t(1762)=5.92 (R1 vote share) (R1 vote share) P>|t|=0.00 2. Not observed in R2 vs. Observed in R2 2.0% 69.3% 67.3% t(1761)=2.47 (R2 vote share) (R2 vote share) P>|t|=0.014 3. Never observed vs. Observed in both R1 and R2 4.5% 70.7% 66.2% t(1116)=4.48 (R2 vote share) (R2 vote share) P>|t|=0.00 4. Never observed vs. Observed in both R1 and R2 5.8% 62.8% 57% t(1116)=5.36 (Average of R1 and (Average of R1 and P>|t|=0.00 R2 vote share) R2 vote share) 5. Never observed vs. Observed in one or both rounds 4.6% 62.7% 58.1% t(1761)=5.65 (Average of R1 and (Average of R1 and P>|t|=0.00 R2 vote share) R2 vote share) 6. Never observed vs. Observed only in R1 4.4% 70.7% 66.3% t(1138)=4.40 [133.9.85.196] Project MUSE (2024-10-17 07:11 GMT) Waseda University Library (R2 vote share) (R2 vote share) P>|t|=0.00 7. Never observed vs. Observed only in R2 2.0% 70.7% 68.7% t(1013)=1.73 (R2 vote share) (R2 vote share) P>|t|=0.084 8. Observed only in R2 vs. Observed in both R1 and R2 2.5% 68.7% 66.2% t(621)=1.93 (R2 vote share) (R2 vote share) P>|t|=0.054 9. Observed in both R1 and R2 vs. Observed only in R1.11% 66.3% 66.2% t(746)=0.094 (R2 vote share) (R2 vote share) P>|t|=0.93 10. Observed only in R1 vs. Observed only in R2 2.4% 68.7% 66.3% t(643)=1.83 (R2 vote share) (R2 vote share) P>|t|=0.067 a Reported results reflect two-sample t-tests with equal variances. 54 W O R L D P O LI T I C S Density 0.2.4.6.8 1 Kocharian Vote Share Incumbent vote in observed polling stations Incumbent vote in unobserved polling stations FIGURE 1 KERNEL DENSITY PLOT OF ROUND 1 VOTE SHARE FOR KOCHARIAN round has a persistent effect on second-round behavior, indicating that international observers had both transitory and persistent effects. TRANSITORY EFFECTS The first two tests presented in Table 1 reveal the transitory effect of observers on electoral fraud committed by supporters of the incumbent. The first test compares the incumbent’s performance in the first round between unmonitored and monitored polling stations. The incumbent did about 6 percent better in polling stations without international observers, suggesting that the presence of international observers in round 1 reduced election-day fraud by about 6 percent in visited poll- ing stations. The level of statistical significance is high. In the second round test 2 shows that the deterrent effect of observers continued. The incumbent performed about 2 percent better in polling stations where observers were not present, suggesting a reduction in the size of the effect. However, test 2 does not control for any persistent effects that first-round observation may have had on second-round behavior. Tests 3 and 4 compare the two groups of polling stations that re- ceived the same treatment in both rounds (never monitored versus monitored in both rounds). Test 3 uses the round 2 incumbent vote  O B S ERV ER EF F E C T I N I N T ER NAT I O NA L P O LI T I C S 55 share to compare polling stations that were never monitored and polling stations that were monitored in both rounds. In this comparison, the in- cumbent received 4.5 percent more of the round 2 vote in polling stations that were never observed relative to those observed in both rounds. Test 4 makes the same comparison using the two-round average vote share and shows a similar result. Across both rounds, there is a 5.8 per- cent difference in average vote share for the incumbent. Together, tests 3 and 4 provide more nuanced empirical support for the finding that international observers had a strong deterrent effect on the incidence of election-day fraud. Test 5 adds one additional piece of support for the hypothesis that observers reduced election-day fraud. Test 5 uses the average vote share for both rounds and shows that over the course of the election, the in- cumbent received 4.6 percent more of the vote in polling stations that were not monitored in either round as compared with the average vote share in polling stations visited in one or both rounds. PERSISTENT EFFECTS The second set of tests examines the possibility that the presence of in- ternational observers in the first round had a persistent effect in the sec- ond round. If the effect of international observers had lasting effects on fraudulent behavior, the polling stations that were monitored in the first round should be less likely to experience fraud in the second round. Tests 2 and 3 demonstrate a persistent effect. If first-round observa- tion had no effect on second-round fraud, then the difference in the incumbent vote share should be about the same in the two tests. When polling stations that were monitored in round 1 but not in round 2 are included in the control group and the round 2 vote share is used as the dependent variable, as in test 2, the observed deterrent effect shrinks by more than half, from 4.5 percent to 1.98 percent. In order to test whether the treatment of first-round monitoring had a persistent deterrent effect on fraud in the second round of the elec- tion, test 6 compares the round 2 vote share between two groups that were not monitored in the second round. One group was monitored only in the first round; the other group was not monitored in either round. The result of this test provides additional evidence that observ- ers had a persistent deterrent effect on electoral fraud. Monitoring in the first round is associated with a decrease in the incumbent’s second round vote share by 4.4 percent. This implies that polling station offi- cials who were visited by international observers in the first round were less likely to commit fraud in the second round. 56 W O R L D P O LI T I C S Comparing the round 2 vote share between polling stations that were monitored in round 2 and those that were never monitored, test 7 shows that the incumbent performs about 2 percent worse in monitored polling stations, but the difference is not quite statistically significant. Similarly, test 8 compares the round 2 vote between polling stations that were monitored in both rounds with those that were monitored only in round 2. The difference of means test shows a small difference between the two groups that nearly achieves statistical significance. The rela- tively smaller size of the round 2 effect is not explained by the results but could stem from the fact that Kocharian earned close to 50 percent of the round 1 vote and around 70 percent of the round 2 vote. These results suggest that if first-round monitoring took place then second- round monitoring had only a marginal additional deterrent effect. Within first-round monitored polling stations test 9, like test 8, provides additional support that if international observers were present in round 1, round 2 monitoring does not have a statistically significant effect. Test 10 indicates that there is little difference in overall fraud reduction between the two groups that were monitored only once, regardless of whether the monitoring took place in the first round or the second round. CHECKING AS-IF RANDOMIZATION Ideally, in any experimental research design the assignment of the treat- ment could be examined in relation to a background covariate in order to test for balance between the treatment and control groups. In this case, the ideal covariate would be an independent measure of the can- didates’ likely vote share, such as public opinion polling. These data are not available for Armenia at the polling-station level for the first round, but as Table 2 shows, observer distribution does not appear to follow a clear pattern that would predict Kocharian vote share. Coverage varies by region from a low of 28 percent of polling stations monitored in Ara- gatsotn to 69.59 percent of polling stations monitored in the capitol of Yerevan. The last column of Table 2 suggests that much of this difference is due to voter density, as there is relative balance in voters per monitored polling station within each region. Additionally, the OSCE mission ob- served extensively outside of urban areas where there are fewer voters and travel is more time consuming. To illustrate, an urban polling sta- tion is defined as one that is in the region of Yerevan, is a regional capi- tal, or is one of the seven biggest cities (population > 40,000). All other polling stations are nonurban, which includes rural and periurban poll- ing stations. Using these criteria, 45 percent of all polling stations are nonurban. International observers visited 38 percent nonurban polling  O B S ERV ER EF F E C T I N I N T ER NAT I O NA L P O LI T I C S 57 TABLE 2 ROUND 1 OBSERVER COVERAGE BY REGION Average Voters Voters/Total Total Polling Percent per Polling Monitored Region Stations Monitored Station Polling Stations Aragatsotn 133 27.82 701 2520 Ararat 137 53.28 1355 2543 Armavir 153 38.56 1319 3420 Gegharkunik 148 32.43 1140 3515 Kotayk 132 43.18 1429 3309 Lori 226 33.63 1059 3148 Shirak 273 25.64 907 3537 Syunik 54 37.04 859 2319 Tavush 80 28.75 1152 4007 Vayots Dzor 40 37.50 1017 2712 Yerevan 388 69.59 1751 2516 stations in the first round of the election and 35 percent nonurban in the second round. Given that there are more voters in each urban polling station, observers covered nonurban areas extensively. As a further examination of the “as-if ” randomization, round 2 treat- ment of observers is compared with vote share and turnout in round 1. These round 1 outcomes should be equal between polling stations that were monitored in the second round and those that were not. The re- sults presented in Table 3 support the assertion that the method used [133.9.85.196] Project MUSE (2024-10-17 07:11 GMT) Waseda University Library by the OSCE/ODIHR to assign observers approximates randomization. In an OLS regression, the presence of observers in the second round is unrelated to voter turnout or to vote share for Demirchian.39 ALTERNATIVE EXPLANATIONS AND POLLING-STATION “FINDABILITY” Even though the assignment of international observers approximated randomization and observers were instructed to visit only the polling stations assigned to them, and in theory random assignment controls for possible covariates, it is still possible to test whether other variables that could influence the incumbent’s vote share bias the results. The three most likely alternative explanations for which data are available are observer difficulty in reaching small polling stations, the nature of 39 Demirchian vote share is used to make the comparison clearer. If there were systematic biases in the polling stations monitored in round 2, they should show as a statistically significant difference between these two groups. Comparisons using round 1 Kocharian vote share are available from the author, but for reasons that are likely related to polling-station accessibility, round 1–monitored poll- ing stations are more likely to be monitored in round 2, thus making it unlikely that observed round 1 Kocharian vote share would be statistically independent of round 2 monitoring. 58 W O R L D P O LI T I C S TABLE 3 ROUND 2 MONITORING VS. BACKGROUND COVARIATES DV: Voter Turnout DV: Demirchian D.V. Demirchian Round 1 Vote Share Round 1 Vote Share Round 1 Coefficient Coefficient Coefficient (S.E.) (S.E.) (Robust S.E.) a Round 2 Monitored 0.018 0.002 0.002 (0.010) (0.007) (0.021) Constant 0.637** 0.276** 0.276** (0.006) (0.004) (0.018) N 1763 1763 1763 F (1, 1761) = 2.96 (1,1761) =0.10 F(1, 10)=0.01 Prob > F 0.0853 0.7471 0.9179 * significant at 5%; ** significant at 1% a Robust standard errors are clustered by region. the urban-rural split, and whether a given polling station is located in one of the regions that border Nagorno-Karabakh. First, note that the variables recorded in the analysis are actual visits by international observers rather than assignment of the polling station to the treatment group. If observers had systematic difficulty in reach- ing polling stations they were asked to visit due to, for example, factors such as muddy roads, the observed difference between monitored and unmonitored polling stations could be the result of systematic differ- ences between polling stations that are easy to find and those that are not.40 This would lead to a biased test if and only if polling stations that were difficult for observers to find were systematically related to support for a particular candidate. Table 4 presents three alternative specifications of an OLS model examining the effects of monitoring and proxies for polling-station accessibility on Kocharian vote share in round 1. Two measures of polling-station size are used: registered voters and total votes cast, as both could be influenced by fraudulent practices and as the total number of votes cast could also be a reason- able proxy for the difficulty in reaching polling stations (that is, both voters and observers have a difficult time reaching the polling station). The effect of monitoring on Kocharian vote share remains statistically significant in models 1–3, although the size of the monitoring effect varies somewhat with the specification and is smaller than the effect presented in Table 1. This suggests that polling-station size is related to monitoring (most likely because smaller stations are harder to find), 40 I am grateful to an anonymous reviewer for suggesting this be included as an alternative explana- tion.  O B S ERV ER EF F E C T I N I N T ER NAT I O NA L P O LI T I C S 59 TABLE 4 ALTERNATIVE SPECIFICATIONS: THE EFFECTS OF MONITORING ON ROUND 1 VOTE SHARE Dependent Variable= Model 1 Model 2 Model 3 Round 1 Kocharian OLSCoefficient OLSCoefficient OLSCoefficient Vote Share (S.E.) (S.E.) (S.E.) Monitored voting –0.031** –0.019* –0.039** (0.010) (0. 009) (0.010) Total registered voters 0.0003** –0.0001** (0.00002) (7.43e-06) Total votes cast –0.0003** –0.00008** (0.00003) (0.00001) Constant 0.633** 0.648** 0.592** (0.001) (0.010) (0.010) N 1764 1764 1764 F (3, 1760) = 132.50 (2, 1761) = 108.88 (2, 1761) = 37.52 Prob > F 0.0000 0.0000 0.0000 * significant at 5%; ** significant at 1% but the effect of monitoring is robust to the inclusion of measures of polling-station size. The second alternative explanation is that if Kocharian performed disproportionately well in rural areas and if the selection of polling sta- tions included a disproportionate number of urban polling stations, then it is possible that Kocharian’s support in rural areas could be driv- ing the difference between monitored and unmonitored polling sta- tions. Table 5 presents the results of difference of means tests between experimental groups using round 1 vote share. The other dependent variables (round 2 vote share and average vote share between rounds) are consistent with the results presented below. Table 5 shows that the reported differences in average Kocharian vote share between monitored and unmonitored polling stations are not driven by urban-rural bias or by voters in the Nagorno-Karabakh region. The first two rows of Table 5 show that the deterrent effect of observers is present in urban areas. If Kocharian’s high performance in rural areas were driving the main results, then the difference in Kocha- rian’s performance between monitored and unmonitored polling sta- tions should disappear when rural areas are excluded. The final alternative explanation is that because Kocharian is a native of Nagorno-Karabakh and is overwhelmingly popular there, the results could be driven by disproportionately high support near Nagorno- Karabakh if observers avoided the area. Rows 3 and 4 of Table 5 show 60 W O R L D P O LI T I C S TABLE 5 DIFFERENCE OF MEANS TESTS OF ROUND 1 KOCHARIAN VOTE SHARE Monitored (N) Not Monitored (N) Difference a Urban 46.5% 54.3% 7.8% (463) (516) t=6.46 P>|t| =0.00 Nonurban 51.2% 54.0% 2.8% (285) (500) t=1.67 P>|t| =0.095 Adjacent to Nagorno-Karabakh 74.4% 71.0% –3.4% (35) (59) t=–0.09 P>|t| =0.37 Not adjacent to Nagorno-Karabakh 47.0% 53.2% 6.1% (713) (957) t=6.14 P>|t| =0.00 a Reported results reflect two-sample t-tests with equal variances. If the null hypothesis can be rejected, that the difference between groups is not significantly different from zero, the difference is reported in bold. that the observed deterrent effect of observers is not a result of unusually high support in the regions most affected by the conflict in Nagorno- Karabakh. If observation were biased away from the Kocharian strong- hold near Nagorno-Karabakh, the nationwide results could be biased. Table 5 shows that the results hold in the regions outside of the region of Kocharian’s highest support. In summary, the first-round deterrent effect is consistently present in comparisons between monitored and unmonitored groups, and three alternative explanations do not account for the difference in the incum- bent’s vote share between monitored and unmonitored polling stations. DISCUSSION OF FRAUD REDUCTION This natural experimental design has provided a unique test of whether international observers can deter election-day fraud, providing causal evidence of how international actors can influence domestic politics. The evidence presented above shows that the presence of international observers depressed the incumbent’s round 1 average vote share by 6 percent in polling stations that were observed. Since the incumbent was only 0.52 percent away from avoiding a round 2 runoff, international observers may have been responsible for triggering a second round of competition.41 It is also possible that amidst the already charged at- 41 In the first round, observers visited 42 percent of polling stations. The 5.8 percent average reduction in Kocharian vote share, reflected nationally, was 2.44 percent (aggregate observer effect = 42% * 5.8%).  O B S ERV ER EF F E C T I N I N T ER NAT I O NA L P O LI T I C S 61 mosphere surrounding the election and Kocharian’s desire to play the election game (if not abide by all the rules), Kocharian engineered a last-minute reduction in his vote share in order to look more like a democrat. This possibility does not detract from the demonstrated de- terrent effect of international observers on election-day fraud; it simply moderates the possible claim that international observers caused the second-round runoff election. The evidence does not suggest that international observers deterred fraud to the point that they could not observe it. As already detailed, observer teams witnessed a significant number of electoral irregulari- ties, including evidence of ballot-box stuffing, political violence, and intimidation of polling officials. The official OSCE/ODIHR report lists twenty-two polling stations where ballot-box stuffing was witnessed in the first round and more than seventy polling stations in the second round. In the case of Armenia, election-day fraud deterrence meant that the international observers slowed down the rate of fraud in the polling stations that they visited. The precise mechanisms by which this took place are unknown, but a plausible explanation is that the presence of international observers caused polling-station officials to reduce the rate of intended election-day fraud, either because they were instructed to stop fraudulent activities in front of international observ- ers or because many were worried about being caught. What explains the persistent deterrent effect of observers on fraud in round 2? It is possible that in the first round of the election, the [133.9.85.196] Project MUSE (2024-10-17 07:11 GMT) Waseda University Library polling-station workers were uncertain regarding the reach and qual- ity of the international observers. Prior to the election the commission charged with administering the election had undergone significant re- structuring. The restructuring was such that most of the polling-station administrators were selected just before the 2003 presidential elections, making them relatively inexperienced with international observers. Af- ter being observed in the first round and witnessing the mission’s nega- tive assessment of the election, those officials monitored in the first round may have updated their beliefs regarding the quality and reach of international observers and decided they were unlikely to get away with election-day fraud. Those that were not observed in the first round were less likely to update their beliefs on the reach of international observers but may have still been somewhat deterred in the second round by the OSCE’s negative assessment of the first round of elections. Election observation in the first round decreased round 1 fraud and to a lesser degree also decreased fraud in the second round. The second-round reduction of fraud likely occurred in three ways: those that were monitored 62 W O R L D P O LI T I C S in the first round committed less round 2 fraud, those that were moni- tored in the second round committed less fraud in the second round, and because the observers demonstrated that they were likely to give a negative