INSTITUTE. The Effectiveness of Democracy Aid to Different Regime Types and Democracy Sectors. Anna Lührmann, Kelly McMann and Carolien van Ham

INSTITUTE The Effectiveness of Democracy Aid to Different Regime Types and Democracy Sectors Anna Lührmann, Kelly McMann and Carolien van Ham January 2017 Working Paper SERIES 2017:40 THE VARIETIES OF DEMOCRACY INSTITUTE

Varieties of Democracy (V-Dem) is a new approach to the conceptualization and measurement of democracy. It is co-hosted by the University of Gothenburg and University of Notre Dame. With a V-Dem Institute at University of Gothenburg that comprises almost ten staff members, and a project team across the world with four Principal Investigators, fifteen Project Managers, 30+ Regional Managers, 170 Country Coordinators, Research Assistants, and 2,500 Country Experts, the V-Dem project is one of the largest-ever social science researchoriented data collection programs. Please address comments and/or queries for information to: V-Dem Institute Department of Political Science University of Gothenburg Sprängkullsgatan 19, PO Box 711 SE 40530 Gothenburg Sweden E-mail: contact@v-dem.net V-Dem Working Papers are available in electronic format at www.v-dem.net. Copyright 2017 by authors. All rights reserved.

The Effectiveness of Democracy Aid to Different Regime Types and Democracy Sectors 1 Anna Lührmann Post-doctoral Research Fellow, V-Dem Institute University of Gothenburg Kelly McMann Associate Professor of Political Science Case Western Reserve University Carolien van Ham Lecturer in Politics, University of New South Wales, Sydney Research Associate, V-Dem Institute 1 For helpful comments, we thank Agnes Cornell, Kyle Marquardt, Svend-Erik Skaaning and participants of the 2016 V-Dem Internal Research Conference, the 2016 Development Research Conference in Stockholm, the Effective Democracy Aid panel at the American Political Science Association 2016 annual meeting, and the 2016 European Consortium for Political Research General Conference where earlier versions of this paper were discussed. We gratefully acknowledge funding for this research project from the Riksbankens Jubileumsfond (grant M13-0559:1), the Knut and Alice Wallenberg Foundation (Grant 2013.0166), and the Australian Research Council DECRA funding scheme (grant number RG142911, grant name DE150101692). 1

Abstract Large-N studies suggest that democracy aid is effective, while multiple small-n investigations call such findings into question. This paper accounts for this contradiction and significantly improves our understanding of democracy aid effectiveness by disaggregating democracy aid into specific types and examining effectiveness in different regime types. We argue that a specific type of aid is more likely to be effective when the aid does not pose a threat to regime survival and when the aid matches the particular democratic deficits in a country. Analysis of OECD aid and Varieties of Democracy data for 119 countries from 2002-2012 supports our argument. Keywords: Democracy aid, democracy assistance, democracy promotion, democratization, electoral authoritarianism, aid effectiveness 2

Introduction Prior studies have reached mixed conclusions about the effectiveness of democracy aid. Large-N cross-national studies suggest that the overall effect of democracy aid on democratization has been positive. 2 However, case studies and qualitative accounts provide more critical assessments. 3 In particular, skepticism about democracy aid to authoritarian regimes is rising. 4 For example, Bush argues that democracy aid in autocracies is largely futile, because it does not threaten the survival of dictators. 5 Large-N studies of democracy aid have not adequately addressed such challenges, because they have mainly focused on average effects of democracy aid irrespective of the regime type of the recipient country. 6 Hence, we know little about how the effectiveness of democracy aid varies across regime types. Furthermore, democracy aid targets specific sectors elections, media, human rights and civil society. These different types of democracy aid might also yield divergent returns. To help address this weakness in the literature, this paper disaggregates democracy aid into specific types and examines its effectiveness in countries with different regime types. Theoretically, we argue that aid is more likely to be effective when two conditions are present. First, when aid does not pose a threat to the regime either because it does not challenge leaders survival strategy or because the country essentially lacks a regime aid is effective. When aid challenges leaders survival tactics, they will prevent it from promoting democracy. Second, when the type of aid matches the particular democratic deficits in a country, it is more likely to be successful. As a result, we expect aggregate aid to be effective when the threat to regime survival is low and democratic deficiencies are high. Empirically, we test this argument by using existing data in a novel way and by incorporating a new dataset. These empirical innovations allow us to examine the effectiveness of democracy aid in a more detailed, comprehensive, and accurate manner than previous research. We use data from the Organization of Economic Co-operation and Development (OECD) member states on specific types of democracy aid civil society, elections, media, human rights, rather than the blunter aggregate democracy aid measure 2 Finkel, Pérez-Liñán, and Seligson 2007; Kalyvitis and Vlachaki 2010; Scott and Steele 2011. 3 Burnell 2011; Peou 2007; Whitehead 2004; Zeeuw 2005. 4 Schlumberger 2006. 5 Bush 2015. 6 A notable exception is the work of Cornell (2013). However, she examines democracy aid effectiveness only at the authoritarian end of the regime spectrum. 3

typically used. 7 To examine the effect of each type of aid, we measure specific components of democracy using new data from the Varieties of Democracy (V-Dem) dataset. In addition, our reliance on OECD data also distinguishes our work from most prior studies, which examine aid from only the U.S. 8 Donors other than the U.S. account for approximately twothirds of democracy aid and typically differ in their approach and focus, making it essential to examine the larger set of donors. 9 Also unlike most other studies, we focus exclusively on democracy aid, rather than mixing in governance and rule of law aid. 1011 Donors often report activities as aid for governance and rule of law even though they primarily serve other purposes, such as security and state-building. 12 Especially in countries with autocratic regimes this aid could be used for repression and cooptation, not democratization. Finally, whereas most studies on democracy aid rely on data about commitments and not the actual disbursements of democracy aid, we use disbursement data. 13 Only projects that actually have been implemented can be expected to have an effect. We also offer conceptual and methodological innovations that allows us to address a puzzle in the field of democracy aid: if leaders in a recipient country perceive democracy aid as a threat to their survival, why would they agree to let it in their country in the first place? Theoretically, studies that consider threats to regime survival do not consider the two different decisions regimes make whether to allow democracy aid into the country and whether to allow it to be implemented as intended. 14 Prior research has modeled this twostage process using selection models as we do, but it has not offered an answer to the underlying puzzle. 15 Our theoretical framework accounts for the puzzle by explaining how leaders can benefit from accepting aid and then not allowing it to be implemented. This explanation serves as the foundation for our main argument about threats to survival and democratic deficits, which account for the second decision regimes make whether to allow aid to be implemented as intended. Unlike prior research, this paper tackles the two stages of 7 OECD 2016. 8 Finkel, Pérez-Liñán, and Seligson 2007; Scott and Steele 2011. 9 OECD 2016; Pospieszna 2014; Youngs 2008. 10 Birch 2011; Bush 2015, 57; Dietrich and Wright 2015, 222; Finkel, Pérez-Liñán, and Seligson 2007, 434. 11 This is consistent with donor reporting. For example, USAID does not subsume all governance or rule of law- related activities under the label of democracy aid, but explicitly refers to the whole sector as democracy, rule of law and governance. 12 For example, the United States included large-scale counter-narcotics and security and law enforcement programs in countries such as Mexico, Afghanistan or Colombia as aid in the sector of legal and judicial development. OECD 2016. 13 See Birch 2007; Dietrich and Wright 2015; Cornell 2013; Kalyvitis and Vlachaki 2010. 14 See Cornell 2013. 15 Dietrich and Wright 2015; Finkel, Pérez-Liñán, and Seligson 2007. 4

democracy aid in a methodologically innovative way through both its theoretical framework and its models. Our findings are consistent with our main argument. We find that aggregate democracy aid is effective when overall the threat to regime survival is low and democratic deficiencies are high; this is supported by our results showing that aggregate aid is effective in electoral autocracies and countries essentially lacking a regime, but not in closed autocracies and liberal democracies. We also find that specific types of democracy aid are effective where they do not threaten the regime s survival and where a particular democracy deficit exists. Specifically, our results indicate that election aid is effective in closed autocracies; election and human rights aid is successful in electoral autocracies; aid to bolster civil society, media, and human right aid works in electoral democracies; and election and media aid are effective in countries essentially lacking a regime. We find that any type of aid tends to be ineffective in liberal democracies. In sum, this paper makes theoretical, empirical, and methodological contributions that enable it to account for the puzzling discrepancies between the findings from large-n and small-n studies. Whereas aggregate democracy aid tends to be effective, on average, globally, as large-n studies have shown, specific types of aid are less effective under certain regime types, as case studies have demonstrated. Our theoretical framework, data, and methodological approach demonstrate how these two sets of seemingly contradictory findings are complementary. This clarification significantly improves our understanding of democracy aid effectiveness and offers insights for policymakers and practitioners. The paper proceeds by first describing the theoretical argument and presenting the hypotheses. The second section uses data about democracy aid patterns to demonstrate the relevance of examining different types of aid to countries with different regime types. Sections three and four elaborate on the data and methods used and present the results, and the final section concludes with a reflection on the implications of our findings. I. Theoretical Framework Democracy aid is international development assistance with the specific goal [to] foster and advance democratization. 16 In examining democracy aid effectiveness, the outcome we are trying to explain is incremental movements toward more democratic practices and institutions, rather than a full transition to democracy. This definition of effectiveness is 16 Carothers 2015, 59. 5

empirically grounded: most studies have found incremental improvements rather than complete transitions. 17 For aid to have the possibility of making improvements, it must enter the country stage one and be implemented stage two. At each stage the regime of the recipient country has an influence over the aid. Stage One: Recipient governments have a decisive say in whether and which democracy aid enters their countries. National laws typically require government approval for aid to government institutions. Donors rarely take a confrontational approach by channeling aid without government approval directly to civic organizations or media outlets. 18 All OECD donors committed to the national governments ownership of democracy aid and to use country systems and procedures to the maximum extent possible in the Paris Declaration. 19 This strong commitment gives prejudice to working collaboratively with national governments. Evidence of this commitment comes from the OECD data about the channels through which aid is delivered. For the aid projects for which data are available, for 2002 to 2013, 23 percent channeled funds to governments themselves and 54 percent to multilateral institutions, which typically work closely with recipient governments. Only 21 percent of projects provided funds directly to NGOs. 20 The necessity of approval from recipient governments begs the question of why a regime would accept aid that could, at least at first glance, threaten its survival. For example, why would an autocrat agree to aid to promote free and fair elections? The reason is that even for non-democratic regimes there are international and domestic benefits to accepting democracy aid. Countries that accept democracy aid create an appearance of being supportive of democracy or some component of it. As a consequence, they can expect more international aid and foreign direct investments from democratic countries and organizations. 21 Domestically, this pro-democratic appearance can boost legitimacy and reduce pressure from opponents. So for authoritarian governments, accepting aid can provide international and domestic benefits. 22 17 See Carothers 1999, 304. 18 Bush 2015. 19 OECD 2008, 5. 20 Thirty-nine percent of all projects data are listed as other or missing. Hence, these data have to be interpreted with caution and are of limited utility. Because we did not find non-random patterns in the distribution of missing data and donor statements corroborate, it is safe to say that recipient governments have control of large shares of democracy aid. OECD 2016. 21 Hyde 2011, 114. 22 Of course, regimes sometimes do prohibit aid from entering their countries, particularly specific types of aid, such as civil society aid (Carothers 2015). The current increase in restrictions on foreign funding to nongovernmental organizations (NGOs) in authoritarian contexts, of which Russia and Egypt are two concrete 6

In some countries, as a result of extensive civil war or state collapse, for example, there is no intact political regime. Democracy aid cannot, of course, pose a threat to the survival of a non-existent regime. And, there is no regime to decide whether aid enters the country. In these situations, donors negotiate with different factions in society in order to provide aid. These factions typically have a strong interest in gaining additional legitimacy from international support. 23 Stage 2: The focus of our argument is on stage two why aid that enters the country is successfully implemented in some cases but not others. We argue that a specific type of aid is more likely to be effective when two conditions are present: 1) when the aid does not pose a threat to the regime either because it does not challenge leaders survival strategy or the country essentially lacks a regime, and 2) when the type of aid matches the particular democratic deficits in a country. These are necessary conditions for aid to be effective. When the overall threat is low and deficiencies are great then there is a positive effect of aggregate democracy aid. A threat to a regime is something that will likely weaken or destroy the government. Most countries have intact regimes, so the regimes are essentially gatekeepers to aid implementation. 24 For that reason regimes will prevent aid that threatens their survival from being implemented as it is intended. For example, a strong civil society poses a threat to certain types of regimes, so those ruling elites are likely to try and divert aid aimed at improving independent civic groups organizational capacity to more government-friendly groups. 25 When a specific type of aid challenges their survival strategy, leaders will ensure that the aid is squandered or they may redirect it for their own purposes. Not all democracy aid challenges the survival strategies of regimes. Leaders of democratic regimes, of course, accept democratic institutions and practices, so democracy aid aligns with their regime survival strategy. Democracy aid can also be compatible with the survival strategies of autocratic leaders. Considering the most extreme circumstance, democracy aid can reduce the risk of demise of the regime through armed conflict. 26 Also de jure democratic institutions are an increasingly important part of regime survival strategies for non-democratic political regimes. 27 Government leaders can actively use democracy aid to examples, bears witness to this practice (Dupuy, Ron, and Prakash 2016). Because our focus is on the implementation of aid once it is accepted by receiving countries, we do not focus on the prohibition of aid. 23 Lührmann 2015. 24 Tolstrup 2014. 25 Bush 2015, 61. 26 Savun and Tirone 2011. 27 Schedler 2002; 2013; Roessler and Howard 2009; Donno 2013. 7

strengthen their own positions. 28 For example, aid for electoral processes, parliament, and political parties can be helpful for authoritarian stabilization. Elections can be instrumental for signaling popular support, and a show of popular support can deter rivals within the national elite. 29 Elections, as well as parliaments, can be used to co-opt opponents and thus further secure the incumbents positions. 30 The establishment of political parties has been shown to lower the risk of forced removal from office. 31 Because much democracy aid consists of technical assistance, it can strengthen the capacity of governments to, for example, run elections and carry out judicial and bureaucratic processes, without enabling those institutions to become more independent or provide a more level playing field. 32 In sum, democracy aid that aligns with a regime s survival strategy is likely to be more effective because regimes allow it to be implemented as intended. When countries essentially lack a regime, democracy aid, regardless of type, is not a threat to the regime. There is no regime survival strategy for aid to be incompatible with, and there is no regime to prevent the implementation of aid. A democratic deficiency, the second component of our argument, is a political institution or practice that is not fully democratic. Aid targeted at a particular institution or practice is more effective the less democratic the institution or practice, because there is more room for improvement. In other words, the marginal utilities of aid decline as institutions or practices become more democratic. Typically more resources are needed to move from good to very good than from fair to good. Democratic deficiencies can result from the intentionally undemocratic design of political institutions as well as weak capacity to implement the complex processes that democratic governance requires. For instance, a clean election requires both the political commitment of the incumbent to refrain from manipulation and technical and financial resources. Countries that essentially lack regimes face significant democratic deficiencies as a government is not even present to run institutions and carry out processes. Where democratic deficiencies exist, aid can help build capacity for institutions and processes to operate democratically. The fact that democratic deficiencies can be due not only to weak capacity but also to lack of regime commitment to democracy underscores again the central role of regimes. For 28 Schlumberger 2006. 29 Hyde 2011; Birch 2011, 52. 30 Gerschewski 2013. 31 Wright and Escriba-Folch 2012. 32 Elklit 1999. 8

a type of aid to be effective not only must a democratic deficiency exist, but the aid must not threaten the regime s survival. The extent to which specific types of aids are effective depends then on the specific regime type or, in the extreme case, on whether a regime even exists. The regime type (or regime s essential absence) helps us explain to what extent a particular type of aid is a threat and whether it addresses a democratic deficiency. When it is not a threat but does address a deficiency, it is effective. When overall the threat is low and deficiencies are high, democracy aid in aggregate is more effective. Below we examine four regime types closed autocracies, electoral autocracies, electoral democracies, and liberal democracies and the special case of countries essentially lacking a regime. We consider their interaction with four types of democracy aid: civil society, election, media, and human rights aid. Civil society aid includes aid for civil society organizations and civic education; it excludes aid to support elections. 33 Election aid is assistance to election management bodies and for election observation and voter education. Media aid includes activities aimed at improving the technical and editorial skills of media outlets and their overall quality. Human rights aid is assistance for official human rights bodies and mechanisms, human rights advocacy and human rights education as well as human rights related aid for specific groups such as indigenous groups, children, and minorities. Closed Autocracies hold no elections or formally restrict competitiveness of elections, such as one-party regimes without a choice on the ballot. In addition, regimes with elections only for the legislature fall into this category, because the head of the government remains unaccountable to voters. 34 Thus, central to this regime type is the formal exclusion of the head of state from electoral competition. Also common is repression of opposition yearround through control of the media and stifling of civil society; this ensures that these other institutions will not effectively challenge the formal electoral restrictions and will otherwise have difficulty holding the regime accountable. Violation of human rights typically deters extra-institutional means, such as mass street protest, to topple the autocrat. Considering these attributes, we expect that election aid, but not media, civil society, or human rights aid, to closed autocracies is effective. Election aid can be both compatible with leaders survival strategies and address a democratic deficit. Because of the formal 33 The information about the content of the specific aid categories comes from the official OECD list of Creditor-Reporting-System (CRS) purpose codes, which donors use to categorize their aid flows into thematic sectors (OECD. ND. The List of CRS Purpose Codes. Available at http://www.oecd.org/dac/stats/documentupload/2012%20crs%20purpose%20codes%20en.pdf, accessed 15 June 2016). 34 Schedler 2009, 309. 9

exclusion of the head of state from electoral competition the immediate threat of election aid is limited. The aid can improve the technical capacity of the regime to hold elections by funding the (regime-controlled) election management body for example, while not increasing the risk that the autocrat will be removed from office. While media, civil society, and human rights aid would address democratic deficits, they pose too much of a threat to the regime for it to allow the aid to be implemented as intended. Year-round suppression of these institutions and liberties is important for the survival of many closed autocracies. Democracy aid s substantial threat to regime survival outweighs the influence of multiple democratic deficiencies, making aggregate democracy aid ineffective. In sum, we expect only election aid to be effective. Electoral autocracies, unlike closed autocracies, allow for a wide variety of democratic institutions and process to exist and operate, but citizens cannot use them to fully contest. The survival strategy of these regimes is to do their best to appear democratic, while still applying authoritarian practices. 35 For example, they allow for de jure multi-party contestation, but elections are de facto not free and fair. 36 Leaders in electoral autocracies are vulnerable to election-related challenges if they allow opposition to build prior to the vote. 37 As long as the regime can control the media and civil society, the risk of removal from office through elections, as well as through mass protest, are minimized. 38 Without independent media and a vibrant civil society, it is difficult for opposition candidates to pose a significant challenge at the polls. Likewise without independent media and a vibrant civil society, mass protests are more difficult to organize and sustain. Mass protests are a possibility in electoral autocracies because some protection of human rights is compatible with electoral autocrats survival strategy of appearing democratic and remaining palatable to the public. Taking into account these characteristics of electoral autocracies, we expect that election aid and human rights aid will be effective but aid to civil society and the media will not. In all these areas democratic deficits exist, but only election and human rights aid do not undercut the regime s survival strategy. We expect aggregate democracy aid to electoral autocracies to be effective because of the significant democratic deficiencies and the overall compatibility of aid with the regime s survival strategy of appearing somewhat democratic. Electoral democracies, as their name indicates, exhibit strengths in electoral institutions and process, but show some deficiencies in one or more of the other components of 35 Schedler 2002. 36 Roessler and Howard 2009; Donno 2013. 37 Roessler and Howard 2009; Schedler 2009, 294. 38 McMann 2006, 178; Schedler 2013. 10

democracy, in particular civil liberties and rule of law. 39 The survival strategy of these regimes is grounded in democracy: they have come to power and continue to govern thanks to democratic practices and institutions. Because electoral democracies exhibit deficiencies in sectors other than elections and because democracy aid does not pose a threat to their survival, we expect media, civil society, and human rights aid to be effective. Overall, we expect democracy aid to be effective. Liberal democracies are characterized by the integrity of their electoral institutions and processes and democratic institutions, such as a vibrant civil society, effective protection of civil liberties, and free and fair elections. While the threat of any type of democracy aid is low, democratic deficiencies are minimal. As a result no type of democracy aid is expected to be effective, and aggregate aid is not expected to be effective. Regimeless countries are those where formerly binding rules and institutions are no longer operational and a new order has not been fully institutionalized. 40 There is essentially no regime, so there is no regime survival strategy. Societal factions can be expected to collaborate constructively with democracy aid providers because they have a strategic interest in establishing a political regime with the help of legitimizing processes such as elections. 41 Democratic deficiencies are significant in each sector because institutions and process have dissolved and a regime does not exist to enforce rights. For these reasons, we expect that each type of democracy aid and overall democracy aid to countries essentially lacking a regime will be effective. Our predictions for the effectiveness of each type of democracy aid and of aggregate democracy aid to countries with each regime type appear in Table 1. 39 Diamond 2002; Schedler 2002. 40 Based on Fishman (1990, 428), we understand a regime as the formal and informal organization of the center of political power, and of its relations with the broader society. We use the term regime not only to denote such institutionalized patterns and norms but also, based on Schedler (2013, 17), the actors in formal positions of state power. However, as Fishman (1990, 428) rightly notes, regimes are more permanent forms of political organization than specific governments. 41 For instance in Libya in 2012, interim authorities were eager to achieve clean elections as a way to fill the power vacuum left by Gadhafi (Lührmann 2015). 11

Table 1. Predictions for Democracy Aid Effectiveness Aggregate Democracy Aid Civil Society Elections Media Human Rights Regime type Closed autocracies + Electoral autocracies + + + Electoral democracies + + + + Liberal democracies Regimeless countries + + + + + Based on our discussion above and Table 1, our hypotheses are: H1: In electoral autocracies, electoral democracies and countries essentially lacking a regime, aggregate democracy aid is effective H2: In closed autocracies, election aid is effective. H3: In electoral autocracies, election and human rights aid are effective. H4: In electoral democracies, civil society, media, and human rights aid are effective. H5: In regimeless countries, all types of democracy aid are effective. Before testing these hypotheses, we examine global patterns of democracy aid to justify our lines of inquiry. II. Democracy Aid Patterns Democracy aid data underscore the enormity of this enterprise and thus the importance of understanding when money is most effectively spent. The data also reveal that aid goes to different democracy sectors and to countries with different regime types, lending credence to our investigation of how democracy sectors and regime types influence how aid effectiveness. The OECD data we present are from 2002 to 2012, the complete period for which aid disbursement data are available. From 2002 to 2012 a considerably sum of democracy aid 24.1 billion USD was distributed, as indicated by the solid line in Figure 1. Aid went to promote civil society, human rights, elections, and media. Aid for civil society was greatest (Figure 1). Spending on human rights aid has been the next highest for most of the time period. Election aid falls 12

slightly below human rights aid for most years, and aid for media is the lowest among these categories. 42 Figure 1. Total and specific democracy aid per year (2002-2012, in million USD) 0 1000 2000 3000 2002 2004 2006 2008 2010 2012 Total democracy aid Elections Human rights Civil society Media Source: OECD 2016. Democracy aid is distributed in countries representing the full range of regime types. This is evident from Table 2 and Figure 2 below. Electoral democracies and electoral autocracies receive the most aid, when it is measured as total aid received between 2002 and 2012 (Table 2, column 1). Regimeless countries receive the most aid (followed by electoral democracies and electoral autocracies), when it is measured by aid per capita (Table 2, column 2 and Figure 2). Closed autocracies and liberal democracies receive considerably less aid. 43 42 Note that the OECD data include information on specific aid spent on women empowerment as well as aid spent on legislatures and political parties. Since aid for women empowerment includes many projects that target generic development rather than democracy, we leave this category out of our analysis. Data for aid targeting legislature and political parties are only available from 2009 onwards and was integrated in the other democracy aid sectors before that time. This type of aid amounts for 2.2% of the overall democracy aid in the studied time period. Therefore, we refrain from analyzing this category individually, but include it in our calculation of the total amount of democracy aid. 43 Examples of liberal democracies that received democracy aid between 2002 and 2012 include Cape Verde, Costa Rice, and Uruguay. 13

Table 2. Levels of aid allocated to different regime types (2002-2012) Aid type Aid overall Elections Media Human Rights Civil Society Regime type Total M N T M N T M N T M N T M N Closed autocracies 2570 0.91 35 282 0.07 28 222 0.06 32 708 0.25 35 1302 0.50 34 Electoral autocracies 8087 1.25 61 1697 0.21 58 383 0.06 59 1706 0.27 61 4116 0.68 61 Electoral democracies 9060 2.39 65 938 0.32 63 402 0.14 63 2228 0.48 65 5296 1.39 65 Liberal democracies 993 0.54 15 62 0.05 14 45 0.02 12 271 0.11 15 610 0.35 15 Regimeless countries 3895 3.77 16 1423 0.90 15 261 0.26 15 540 0.72 16 1567 1.81 16 Total / Average 24605 1.65 129 4401 0.24 122 1314 0.09 127 5453 0.35 129 12891 0.93 129 Source: OECD 2016. Note: T = Total levels of aid is the sum of aid received between 2002 and 2012 (in millions of USD). M = Mean aid per capita received between 2002 and 2012 (in USD). N = the number of countries that received aid in this regime category in this period. The total is larger than the number of countries included in the sample because some countries switch between regime categories. Figure 2. Democracy aid per capita by different regime types 2002-2012 (in USD) Source: OECD 2016. Each regime type has received aid in each specific category (Table 2). In fact, most countries in the OECD database have received each type of aid. The amount of aid per capita, however, varies with regime type. The highest amounts of election, human rights, civil society, and media aid per capita go to countries essentially lacking a regime. When measured in total amounts of aid, the highest amount of election aid goes to electoral autocracies, but 14

the highest amount of media, human rights and civil society aid goes to electoral democracies. Here we used OECD data to underscore the importance of understanding democracy aid effectiveness and understanding the influence of democracy sectors and regime types. In the next section we provide more detail about this dataset and the V-Dem dataset in preparation for testing our hypotheses. III. Data and methods To test our hypotheses about democracy aid effectiveness, we use new data and employ existing data in novel ways. These approaches allow us to examine aid effectiveness in a more detailed, comprehensive, and accurate manner than previous research. Dependent Variables To examine the impact of specific types of aid, as well as aggregate aid we rely on data from the new Varieties of Democracy dataset. 44 We use V-Dem mid-level indices to test the effects of specific types of aid. We use the Clean Elections Index, Alternative Sources of Information Index, Civil Liberties Index, and Core Civil Society Index as our measures of the quality of elections, media, human rights, and civil society, respectively. 45 We use the V-Dem Electoral Democracy Index to measure the impact of aggregate democracy aid. 46 All V-Dem 44 Coppedge et al. 2016a; Coppedge et al. 2016b; Coppedge et al. 2016c. 45 The civil liberties measures the absence of physical violence committed by government agents and the absence of constraints of private liberties and political liberties by the government and includes such indicators as freedom of movement, freedom from torture and freedom from forced labor (Coppedge et al. 2016a). Therefore it is a good measure of the extent to which human rights are respected. We do however, check the robustness of our results using an alternative measure of human rights, namely the more narrowly defined physical violence, that measures to what extent physical integrity is respected. Results of these analyses are substantively the same and reported in the appendix. 46 The mid-level and Electoral Democracy indices include variables requiring expert knowledge to characterize. For these variables V-Dem uses country experts. Potential country experts are identified based on their reputations as known by members of the V-Dem team and as demonstrated by their publications and other works. More than 80 percent of all experts hold Ph.Ds. or M.A.s and work at some type of research institution. Most of the experts are nationals or permanent residents of the country they are coding. For each indicator, five country experts are enlisted, resulting in five separate codings. A coder provides ratings for only those topics in his or her areas of expertise. A measurement model aggregates the data generated by the experts so that one data point exists for each country-year. The measurement model adjusts for systematic bias in coders answers by examining patterns in expert disagreement, for experts own reported confidence in each response they give, and for the tendency of domestic coders to provide less favorable evaluations. For additional details on country experts and the measurement model, see Pemstein et al. 2015; Coppedge et al. 2016c. The Electoral Democracy Index also includes variables that use extant data, such as information about whether or not elections take place. Extant data are factual in nature, and collected by V-Dem staff from existing sources. Full details about each variable that comprises the mid-level and Electoral Democracy indices can be found in the V-Dem codebook (Coppedge et al 2016a). The V-Dem Electoral Democracy aims to capture Dahl s concept of Polyarchy and is explained more in detail in Coppedge, Teorell, and Skaaning 2016; Coppedge et al. 2016a. 15

indices score political regimes and institutions on a continuum varying from 0 (very autocratic) to 1 (very democratic). In our basic models, we measure all dependent variables in the year after which aid was spent. Independent Variables of Interest To achieve a more comprehensive and accurate picture of democracy aid effectiveness we rely on data about specific types of democracy aid disbursements from the OECD Development Assistance Committee (OECD/DAC) database, rather than aggregate commitment data from only the U.S., as many prior works have done. All members of the OECD/DAC record their Official Development Assistance (ODA) in this database. 47 Disbursement information is only available from 2002 onwards, which is why we start our analysis in that year. 48 Using the database s detailed purpose codes, we are able to exclude non-democracy aid, such as that for governance and rule law, and also examine aid to specific democracy sectors civil society, elections, media, and human rights. We use aid per capita logged in all our models to correct for population size and the skewed distribution of democracy aid, as is common for democracy and, more generally, foreign aid studies. 49 To capture the interaction between specific aid types and regime types, we use Schedler s basic classification of regimes types and add regimeless countries as an additional category. 50 We use V-Dem data to measure regime attributes, in order to place each regime in a particular category. We begin by defining regimes as democratic if they achieve basic principles of electoral democracy (freedom of association, suffrage, clean elections, elected executive, and freedom of expression) and hold somewhat free and fair multiparty elections. 51 Among these regimes, those that respect the rule of law and civil liberties, have checks and balances that constrain the executive, and do not constrain the formation of civil society organizations are coded as liberal democracies. Those that do not constrain the executive and do constrain civil society are coded as electoral democracies. All regimes that do not meet these criteria but hold multiparty elections for the head of the executive are coded as electoral autocracies. Countries are categorized as closed autocracies, if they do not hold multiparty elections for the head of the executive. 52 In our final category regimeless countries we sort 47 Countries that not reporting to the OECD/DAC tend not provide democracy aid, such as China and Russia. 48 OECD 2016. 49 Wright 2009, 566. Population data are from World Development Indicators. World Bank. 2016. "World development indicators, 2016." Available at http://databank.worldbank.org/data/reports.aspx?source=wdidatabase-archives-%28beta%29, accessed April 2016. 50 Schedler 2002. 51 Lindberg 2016. 52 Schedler 2009, 309. 16

those countries that the Polity IV dataset did not categorize into autocracies or democracies. Those regimes received a Standardized Authority Code which indicate if a polity is interrupted (e.g. due to foreign occupation), in a state of complete collapse of central political authority, or in a substantial transition process. 53 This coding scheme is laid out in Table 3 and is described in detail in the appendix. Closed Autocracies Table 3. Operationalization of regime types and case examples Electoral Autocracies Minimal democratic principles not achieved No multiparty elections for the head of the executive Multiparty elections for the head of the executive Electoral Democracie s Liberal Democracies Minimal democratic principles achieved Executive is not constrained Civil society is constrained Executive is constrained Regimeless Countries Polity interrupted, collapsed, or in substantial transition Civil society is not constrained Jordan 2012 Nigeria 2011 Bolivia 2012 Ghana 2012 Libya 2012 Control variables Factors other than democracy aid also shape how democratic specific institutions and practices or a country are. To capture this, we include control variables that have commonly been found to be associated with democratization in previous research. 54 These represent the major alternative explanations for increasingly democratic institutions, processes, and regimes. We include GDP per capita (logged) and trade (trade as % of GDP) as levels of democracy are likely to be higher in wealthier countries and countries that are more open to trade. 55 In addition, we include Hanson and Sigman s state capacity measure as levels of democracy might be higher in regimes with higher state capacity. 56 This captures the level of state capacity on a scale from -4 (low levels) to 4 (high levels). Post-conflict contexts are controlled for with a binary variable, internal conflict, signifying an armed conflict during the five years prior to receiving democracy aid based on data from the widely used UCDP/PRIO Armed Conflict Dataset. 57 Post-conflict contexts are commonly viewed as 53 Regimeless countries in the time-period between 2002 and 2012 are for example Afghanistan, Somalia, Bosnia and Herzegovina and Iraq from 2003-2009 (Marshall, Gurr, and Jaggers 2014, 19). 54 Przeworski et al. 2000; Lindberg 2006; Teorell 2010. 55 Przeworski et al. 2000; López-Córdova and Meissner 2008. 56 Hanson and Sigman 2013. 57 Gleditsch et al. 2002; Teorell et al. 2016. 17

challenging for democratization and democracy promotion. 58 We also control for the proportion of the countries GDP that is based on rents from natural resources, natural resources, 59 as we also expect this to lower levels of democracy. 60 Election year is a binary variable indicating if an election took place during the year of democracy aid disbursement, because elections may facilitate democratization processes. 61 Three additional controls address effects of aid separate from the relationship we are studying. We include the variable total aid minus democracy aid, which measures the amount of non-democracy aid the country received and is based on OECD data. 62 We expect nondemocracy aid to also positively affect the level of democracy. 63 We also include the control aid dependency measured as aid as a proportion of the countries GDP, which we expect to have a negative effect. 64 Finally, in the specific aid models we also include a control variable measuring the amount of other types of democracy aid the country received in order to disentangle the effect of the specific type of democracy aid from the effects of other types of democracy aid. Summary statistics of all dependent and independent variables are available in the appendix. Methods We test the effects of aggregate and specific types of democracy aid on levels of democracy using time-series cross-sectional regression models. As we are essentially interested in dynamics within countries over time, we use country fixed effects. Models include all the controls discussed in the previous section. The dependent variable is measured in the year after aid was disbursed, while all independent variables are measured in the same year in which aid was disbursed. Hence all independent variables are effectively lagged one year. Model diagnostics suggest that heteroskedasticity and serial autocorrelation are present in our data. We therefore check the robustness of our results using panel-corrected standard errors and a first-order autocorrelation correction. Results for all robustness checks are reported in the next section and shown in the appendix. We also verify the robustness of our results to selection effects. Democracy aid is not allocated at random. Donors will seek to allocate aid based at least in part on a prior expectation about aid being successful and hence not accounting for such selection effects 58 Lindberg 2006; Zeeuw and Kumar 2006. 59 From World Development Indicators (see note 49). 60 Lindberg 2006; Ulfelder 2007; Teorell 2010. 61 Lindberg 2006. Based on V-Dem variables v2eltype* (Coppedge et al. 2016a). 62 OECD 2016. 63 Wright 2009. 64 Knack 2001. 18

could lead us to find aid to have an effect, whereas in fact prior causal factors explain the level of democracy in the year after aid was spent. 65 Selection effects can be conceived of as a special case of omitted variable bias. Oster has developed a post-estimation test to evaluate the degree to which model coefficients are affected by omitted variable bias. 66 If coefficients change very little, this is an indication that selection bias is not significantly affecting the results. We test to what extent our findings are robust to correcting for omitted variable bias using this method and find that model coefficients change very little. In addition, we also verify the robustness of our results using a two-stage-least-square regression model to explicitly model selection and account for the selection process in our main models. In order to account for selection effects most studies use either treatment or Heckman models where the dependent variable in the selection model is a dichotomous variable indicating whether the country received aid or not. However, this does not apply in our sample: in 90 percent of country-years aid was received. This high percentage is due to our exclusion of OECD donors, who do not receive democracy aid, from our sample and the substantial increase in aid spending during the period we investigate. Hence, the selection process that needs to be accounted for is how much aid countries received, not whether they received aid or not. Therefore, a selection model with a continuous dependent variable is needed. We use a two-stage selection model, modeling selection using Tobit regression to account for the censored nature of the dependent variable, and subsequently including the residuals of the selection models into the main regression to control for selection bias. 67 Results are robust to these alternative specifications and are available in the appendix to this article. A final set of robustness checks uses different dependent and independent variable measures. For the dependent variable aggregate aid, we check our results using the V-Dem Liberal Democracy Index as an alternative measure of the overall level of democracy. We also ran robustness checks using alternative measures for the quality of elections, media, human rights, and civil society, specifically: the Electoral Component Index, the Freedom of Expression Index, the Physical Violence Index, and the Civil Society Participation Index. 68 We also experimented with lags of two and three years following aid disbursement. Results of these robustness 65 Cornell 2013. 66 The test can be performed in Stata (psacalc) and is described in detail in Oster (2013). Note that omitted variable bias is evaluated under the assumption that the selection on the observed controls is proportional to the selection on the unobserved controls. Results are substantively the same and are available upon request from the authors. 67 Based on Baser 2006 and Baser 2015. Results for the selection models are reported in Table 3.0.b. in the appendix. The dependent variables for each selection model differ, in some cases overall democracy aid and in others the specific aid types that we test. 68 See the V-Dem Codebook for a full description of these indices (Coppedge et al. 2016a). 19

checks are available open request. Our main findings hold in these robustness checks as well and will be discussed more in detail in the next section. 69 IV. Results Using these data and methods, we first test whether aggregate democracy aid has an effect on democracy, across all regime types. The purpose of this test is to demonstrate that our data and models produce findings about aggregate aid similar to those of prior large-n research. To explore whether an aggregate effect might occur because of the positive influence of specific types of aid, thus lending credence to our disaggregated approach, we next test the effect of specific types of aid on their targeted institutions and processes, independent of regime types. Then, we turn to the focus of our paper the effectiveness of aggregate aid and specific types of aid in countries with different regime types. Here we test our five hypotheses. Like much prior research, we find a statistically significant positive main effect for aggregate democracy aid across countries regardless of regime type (Model 1, Table 4). This effect does not seem to be driven by a particular type of aid. All sub-types of aid have a statistically significant positive effect on the V-Dem reflecting the area they target (Model 2, 3, 4 and 5; Table 4). All these models include the controls described above. Models 2-5 also include a control for level of other democracy aid. Full model results are presented in the appendix. The robustness checks for these models confirm our findings. The positive effects of aggregate aid and all specific types of aid hold with different model specifications, as well as different lag structures for the dependent variables, and different dependent variables to measure overall level of democracy and the specific dimensions of democracy. Only when the dependent variables are measured 3 years after aid was disbursed are the main effects of aid no longer significant, suggesting that the effects of aid are relatively short-term and are strongest in the first and second year after aid has been disbursed. The results for these robustness checks are available open request. Overall, the findings about aggregate aid and specific aid types are encouraging. They show that democracy aid not only has an effect at the aggregate level, but also that specific types of democracy aid improve the sectors they target. These specific associations provide support for the notion that democracy aid might indeed have a causal effect on levels of 69 Full replication data and documentation can be found on [author s website]. 20