The EDUCO Program, Impact Evaluations, and the Political Economy of Global Education Reform

During the 1990s and 2000s, a policy known as Education with Community Participation (EDUCO) not only became the cornerstone of education reform in El Salvador but also became a global education policy, one which is known for decentralizing to rural families the responsibility for hiring and firing teachers. As is shown in this paper, its rise to fame was not only a product of the particular political-economic context in which it was borne, but was also a product of the impact evaluations produced by the World Bank, which served as the evidence base through which this and other international institutions could legitimately promote the neoliberal model of community involvement represented by EDUCO. Problematically, however, a reappraisal of these impact evaluations reveals, first, that their findings and conclusions around significant effects were not Education Policy Analysis Archives Vol. 24 No. 92 2 warranted and, second, that the entire impact evaluation enterprise is fundamentally flawed due to the financial-political-intellectual complex out of which these studies emerged and back into which they fed as they were used to advocate for market-oriented policy solutions around the world. Thus, in addition to explaining the reform dynamics that gave rise to EDUCO, this paper (a) systematically reviews the findings and limitations of each of the six impact studies that constitute the international knowledge base around this policy, (b) reconsiders what we can reasonably claim to know about EDUCO, (c) reflects on the national and international implications of the critical review presented here, and (d) remarks on the shortcomings of—and the alternatives to—impact evaluations as a means to produce policy-relevant findings.


The EDUCO Program, Impact Evaluations and the Political Economy of Global Education Reform 1
During the 1990s and 2000s, a policy known as Education with Community Participation (EDUCO) not only became the cornerstone of education reform in El Salvador but also became a global education policy (Edwards, forthcoming), in that it circulated around the world in the spaces where the international agenda for education reform is created and communicated (Verger, Novelli, & Kosar-Altinyelken, 2012). Indeed, since the mid-1990s, this policy, which decentralized the responsibility for hiring and firing teachers to parents in rural communities, has been studied by technical groups from at least 18 countries (Meza, Guzmán & de Varela, 2004a), 2 in addition to being promoted by numerous international organizations through high profile publications, including some of the most influential reports in the realm of international development (Edwards, 2013b). A select few key examples are:  The World Bank's World Development Report 2004, which focused on making public services more accountable (World Bank, 2003).  The report in 2005 of the United Nations task force on gender equity (Birdsall, Levine & Ibrahim, 2005).  A publication by the Brookings Institute (an influential think tank in Washington DC) in 2008 on effective development practices (Hartman & Linn, 2008).  The Global Monitoring Reports by UNESCO in both 2009 and 2011, which addressed education governance and education in conflict-affected contexts (UNESCO, 2008(UNESCO, , 2011. While many political and economic factors influenced its emergence and subsequent trajectory (Edwards, forthcoming), one important issue that previous studies of this program have tended not to highlight is the central role played by research and impact evaluations. The present manuscript attempts to fill this gap in the literature. It does this, first, by presenting a critical review of the evidence base that has been used to promote EDUCO, and, then, by discussing the meaning of that evidence-in light of the reappraisal presented here-within the political economic context that produced EDUCO and the impact evaluations that legitimated its dissemination and promotion globally. Going forward, this article has eight sections. The first establishes the predominance of international organizations in education reform, with a focus on the World Bank, its policy preferences and the ways that impact evaluations help to create a demand for the policy solutions it offers. The second section then defines what is meant by a critical analysis of impact evaluations within the political economy of global education reform. The third section explains the politicaleconomic context in El Salvador at the end of the 1980s and beginning of the 1990s. This period is the starting point because the dynamics which characterize it influenced the emergence of EDUCO in 1991 in important ways. The fourth section then shows how the strategic activity of key actors intersected with contextual constrains and then crystallized around certain policy priorities that favored EDUCO. Having clarified structural constraints and the vested interests in the second section, the fifth section systematically reviews the findings and limitations of six econometric studies that were integral to EDUCO's success and which have contributed to EDUCO's highprofile nature in the international realm. The sixth section contains additional discussion, both on how these studies should be understood as well as about what we can reasonably claim to know about EDUCO from a critical perspective. The seventh section considers the national and international implications of the critical review presented here, particularly as they relate to the political-economic dynamics of the education reform process that encompassed EDUCO's emergence and trajectory. The eighth section offers substantive concluding remarks on the shortcomings of-and the alternatives to-impact evaluations, particularly given the financialpolitical-intellectual complex out of which these studies tend to emerge and back into which they tend to feed as they are frequently used to advocate for market-oriented policy solutions in periods of system reform.

The World Bank, Global Education Reform, and Knowledge Production
In the literature on education reform in low-income countries, it has long been acknowledged that international organizations are influential (Berman, 1979(Berman, , 1992, with perhaps no other institution receiving more attention than the World Bank (see, e.g., Collins & Wiseman, 2012; Klees, Samoff, & Stromquist, 2012). Moreover, as has been acknowledged, this influence manifests or is exercised in various ways (Samoff, 2009), ways that demonstrate both the materialist and semiotic forms of power. While the former (i.e., materialist form) has arguably tended to receive more attention from scholars (e.g., when they focus on structural adjustment programs and the ways that the World Bank has coerced countries into enacting reforms by threatening to withhold loan funding otherwise), it can be said that these forms of power are inter-twined and that the latter is just as important. Indeed, part of the reason the World Bank is "the largest development research institution in the world" is precisely because of its financial resources (Gilbert & Vines, 2006, p. 49). And one implication of the fact that the World Bank operates in this way as a "knowledge institution" is that it can influence which reforms are seen as legitimate and desirable, through its ability to elevate and promote its research in strategic ways within the global governance of education (Verger, 2012). Interestingly, recent research has shown that actors in the field of global education policy have come to rely on the World Bank's research, to the point where they look for it and feel uncomfortable without it, even in spaces of consultation designed to be open to other perspectives (Verger, Edwards, & Kosar-Altinyelken, 2014).
Of course, an important aspect of the production of knowledge for development (like other forms of knowledge) is that it is not free from bias (Cox, 2002) and that its valuation is not free from the prevailing norms or standards of quality (Walters, Lareau, & Ranis, 2009). Scholars such as Broad (2006) and Klees and Edwards (2014) have addressed this issue in relation to the World Bank, pointing out that this institution, at its core, is guided by a neoliberal worldview and that the research it produces reflects that worldview by promoting policies and reforms based in market or quasimarket principles. That is, while the World Bank frames its findings and the policies it recommends as the product of scientific inquiry guided by the highest standards of social science (see, e.g., Khandker, Kookwal, & Samad, 2010;Verger, 2012), these claims are undercut by the fact that the general nature of these findings and the recommendations that follow are predetermined. While the coefficients might not be known a priori, we can be sure that the World Bank, in the realm of education reform, is going to demonstrate and emphasize the necessity of instituting policies for school competition, parental choice, teacher accountability, and, more recently, low-fee private schools, to mention a few examples.
However, in order to sustain its business model, the World Bank, like other international development banks, must be able to demonstrate that its policies are successful. Consequently, the World Bank not only promotes reforms that are-according to contemporary wisdom in the current context of neoliberalism (Spring, 2015)-accepted as appropriate to enhance such outcomes as quality and efficiency, but this organization also seeks out and conducts its own research on promising pilot programs or on-going reforms, even if it is also the World Bank that is financing those examples, despite the conflict of interest. If it can be shown through impact evaluations that a particular policy causes positive outcomes in terms of student achievement or student dropout, for example, then that policy will, on the basis of those evaluations, be promoted globally through World Bank publications and will be included in the short list of acceptable reforms that the World Bank is willing to support in other countries (Heyneman, 2003;Steiner-Khamsi, 2012). This state of affairs leads to the observation that the World Bank produces "policy-based evidence" as opposed to "evidence-based policy." The way this happens is a primary focus of the present article.

Critically Analyzing Impact Evaluations within the Political Economy of Global Education Reform
A critical analysis of impact evaluations within the political economy of global education reform means assessing a set of evaluative studies that have been carried out on a policy or program not only in relation to the methodological limitations of the studies but also in relation to the meaning or significance of those studies for policy, for practice, and for the vested interests that may be advanced (or not) on a global scale from their production and the findings they contain. Realizing this kind of review clearly means going beyond (a) a consideration of the stated results and methodological limitations of a corpus of studies, though this is an important step (as, e.g., in Bryk, 1981); it also means (b) understanding those studies within the multi-level (i.e., international, national, local) political-financial-intellectual complex out of which they emerged, (c) appraising the implications of those studies (in terms of their stated findings but also in terms of their political meaning) for the vested actors and institutions that facilitated, utilized, and/or otherwise benefitted from the studies; and (d) tending to the ways that the studies contribute to oppressing certain perspectives or policy preferences while elevating others at each level from the local to the global (Klees & Edwards, 2014;Lubienski, Weitzel & Lubienski, 2009;Robertson & Dale, 2009;Samoff, 1993Samoff, , 2012Verger, 2012). The present article engages with these steps in relation to the series of impact evaluations that were produced by the World Bank of the EDUCO program.
It is suggested that, ultimately, the insights gained through such an analysis are more significant than asking whether a program worked or whether a study's findings hold up to a critical review of their methodological shortcomings alone. The added benefit of taking a critical review of empirical findings (based, e.g., on their methodological basis) and placing those findings within the political economy of education reform and the political economy of research production is that it shifts our gaze away from an exclusive focus on research findings as if they were the product of a clinical and objective process to instead focus on how those findings may reproduce existing power structures and/or advance certain policy preferences over others; or, put differently, we are encouraged to look anew at how and why research findings, by virtue of how they are produced and the power relations that govern their production, may disadvantage certain groups, and with what implications (Water, 1998). Arguably, carrying out such reviews is becoming more imperative, particularly as the relationships among the policymakers, funders, evaluators, publishers, and promoters of research and "policy-relevant" findings continue to evolve and to blur on a global scale and in ways that can involve a conflict of interest wherein those who fund and carry out research are often also the beneficiaries of that research (Ball, 2009(Ball, , 2010Goldie, Linick, Jabbar, & Lubienski, 2014;Lubienski, Scott, & DeBray, 2014;Nambissan & Ball, 2010;Scott & Jabbar, 2014;Verger, 2012).
In practice, then, in order to follow through with the type of critical analysis described above, one must move from the macro to the micro and then from the micro back to the macro. That is to say, in going from the macro to the micro, one must understand: (a) the historicalstructural aspects of the context in question and, in particular, how they constrain the actors in that context; (b) the range of involved actors and the ways they are situated within the larger context, as well as their policy preferences; (c) the relevant processes or opportunities for policy change that emerge, in addition to the rules, requirements, and expectations that govern those processes; (d) the production and utilization of "legitimate" and "policy-relevant" knowledge (in this case, impact evaluations), including who generated and funded that knowledge; and (e) the findings of the relevant studies not only as stated by their authors but also in terms of what one can reasonably claim to know after considering the methodological and practical limitations that accompany the production of the studies in question. At this point, one not only has an understanding of the empirical findings of a body of research but also has a grasp of the political significance of that research and the ways that it has been employed. Subsequently, in going from the micro to the macro, one begins by (f) clearly problematizing that research in technical terms and then considering the implications of both the stated findings and their critical reinterpretation (that is, what one can reasonably claim to know in light of their limitations) for a number of issues, including: (g) processes of policymaking or educational reform, (h) the marginalization of certain political or alternative perspectives, (i) the advancement of political and institutional interests, (j) the emergence or prevalence of global reform trends, and, at the most broad level, (k) the maintenance or extension of the prevailing ideology or common sense in a certain area (in this case, educational governance). While other authors have arguably employed this approach when analyzing research findings (e.g., Edwards, 2012;Klees & Wells, 1983;Lubienski, Weitzel, & Lubienski, 2009;Samoff, 1993;Verger, 2012), it is hoped that specifying the steps in the process of analysis will make the process more clear and more accessible for scholars to utilize.
The above discussion thus indicates that the analysis presented in this paper is based on much more than a review of impact evaluations themselves. Indeed, the findings discussed here have resulted from on-going research that began in 2009 (see, e.g., Edwards, 2012Edwards, , 2013aEdwards, , 2013bEdwards, , 2015Edwards & DeMatthews, 2014;Edwards & Klees, 2012;Edwards, Victoria, & Martin, 2015).
To summarize, in addition to extensively reviewing the literature on EDUCO, the present piece is based on over 100 interviews with key actors from the Ministry of Education in El Salvador (MINED); the United States Agency for International Development (USAID); the World Bank; the United Nations Education, Science, and Culture Organization (UNESCO); and the Central American University, not to mention other organizations such as think tanks within and outside El Salvador as well as national and international non-governmental organizations. For data collection, the author also reviewed the archival collections of the organizations mentioned above (with the content of those collections covering the years of the 1970s to 2012) and participated as an observer in events related to the global promotion of EDUCO and its evidence base. Ten months of full-time fieldwork in El Salvador during 2012-2013 through a Fulbright Research Grant helped to ensure that data collection for this study was both broad and deep. On the basis of this information, it was possible to engage in the multi-step analytic process detailed in the previous paragraph.

Historical Context and Structural Transformation in El Salvador Prior to EDUCO
To understand the emergence of EDUCO it is necessary to first grasp the political economic context of the 1980s in El Salvador. During this decade, El Salvador found itself mired in an ongoing civil war which had, on one side, the government, the landed oligarchy, and the armed forces and, on the other, a coalition of five rebel groups fighting for socialist reform of the land, economy, and social services (LeoGrande, 1998). Collectively, these rebel groups were known as the Farabundo Martí National Liberation Front (FMLN). 3 The impetus for the uprising of these five groups was the continual repression and persecution by the government and its death squads of the population, particularly in rural areas, which was increasingly unaccepting of the country's extreme inequality and its concentration of land ownership in the hands of a few (Montgomery, 1995). During 1980-1984 alone an estimated 70,000 civilians were murdered at the hands of the death squads (Meade, 2010;Orr, 2001).
Layered upon this state of affairs were three additional-and intertwined-aspects of the political economic context. The first was the involvement of the United States. With the Cold War as a backdrop, the socialist uprising in El Salvador (just as in neighboring Nicaragua) was an issue of particular concern. In response to FMLN offensives, the United States provided significant infusions of military and social aid. To put into perspective the magnitude of the aid provided, consider that, while the United States provided approximately $150 million in aid to El Salvador from 1963, it jumped to approximately $6 billion between 1980and 1992(Robinson, 2003. In 1985, military aid alone amounted to $533 million (Booth, Wade, & Walker, 2006, p. 105).
The second aspect to note was the international promotion of neoliberal ideology and reform by the Reagan administration internationally and by USAID within El Salvador. On this latter point, USAID perceived that the context of the civil war provided an opportunity to reconfigure the political landscape of the country in order to promote the adoption of its preferred economic and social policies (Robinson, 2003). In particular, USAID was interested to counter the policy orientation of President José Napoleón Duarte (1984)(1985)(1986)(1987)(1988)(1989), a populist who did not favor liberalizing the economy, but whom the U.S. Congress felt forced to support to ensure stability of the country during the civil war (LeoGrande, 1998).
In line with this strategy, USAID supported the creation of a think tank, the Salvadoran Foundation for Economic and Social Development (FUSADES, for its name in Spanish). This organization served as a key space through which to (a) host famous international economists, such as the "Chicago Boys" (Klein, 2007); (b) lend credibility to neoliberal reform principles; and (c) raise the profile of Salvadoran economists and businessmen who shared USAID's policy preferences (discussed further below). That USAID provided FUSADES with $150 million in contracts during 1983-1993 shows the importance of the latter to the former.
In 1986, USAID also facilitated the creation of the Business Foundation for Educational Development (FEPADE, for its name in Spanish). As a technocratic organization, FEPADE diagnosed the needs of the education sector and worked on USAID's education projects. It also provided training and capacity building to the MINED. Given its origins and orientation, FEPADE was a key source of technical capacity for USAID and the Salvadoran government, particularly during the 1990s, once the rightwing party assumed power. It is of note that the minister of education during 1990-1998-Cecilia Gallardo de Cano-was an education specialist at FEPADE before becoming the Minister of Education. Crucially, she also came from the core of the ARENA party and, as such, was closely networked with members from its highest echelon, including President Cristiani. Moreover, she was keen political player who aspired to become president one day.
Before proceeding with the developments of the 1990s, it is first important to highlight the third aspect of the political-economic context. This is that the eventual success (in 1989) of rightwing party known as the National Republican Alliance (ARENA) was due in part to the failure of the Christian Democrat party under the leadership of Duarte. Initially, after his election in 1984, Duarte was handicapped by the rightist majority in the National Assembly, and so could not move forward with the agrarian reform that had been initiated by the civilian junta that was briefly in power prior to 1984. 4 After 1985, even with control of the National Assembly, Duarte and his party were restricted by the military, USAID, and the conservative and reform-minded portion of the Salvadoran business elite. In the face of resistance by these actors, Duarte could neither bring about peace nor institute social welfare programs, as he had hoped to do. Duarte also failed to open the economy the way that many Salvadoran businessmen and the Reagan administration would have liked. Instead, he implemented economic reforms that ultimately accelerated inflation and imposed higher taxes on imports (LeoGrande, 1998).
Subsequently, and with support from USAID, the rightwing candidate for president, Alfredo Cristiani, won the presidency in 1989. Not surprisingly, Cristiani and "at least 17 business leaders and persons linked with … [FUSADES] became part of the new government," including the ministers of planning and finance and the head of the Central Reserve Bank (Segovia, 1996, p. 55;Eriksson, Kremer & Margaret, 2000). Thereafter, they pursued the economic reform package that was being promoted by FUSADES and USAID. Though the economic and public-sector reforms enacted were often grouped under the innocuous label of "modernization," it was clear that they represented the implementation of a neoliberal perspective on policy. To that end, as documents from the World Bank (1991-from dissertation) indicate, in the month following the assumption of power, Cristiani "initiated a major stabilization and structural adjustment program" in order to "create a more liberalized, private-sector led economy" (p. 5). The first structural adjustment loan from the World Bank would be approved in 1991, followed by a second one in 1993. 5 After his election, Cristiani also began to call for peace talks with the FMLN. At least in part, this was due to the fact Cristiani and other business elite realized that their investments in banking and industry would continue to suffer in the context of ongoing conflict. Peace Accords were finally signed in January 1992, with the help of the United Nations.

The Emergence of EDUCO: Process Dynamics and Vested Interests
A first priority of the Cristiani administration was to address the state and management of the education system. In this area, the immediate challenge for the new Minister of Education, Cecilia Gallardo de Cano, who assumed her post in 1990, was to identify an acceptable policy to expand educational access. Countrywide, by 1990, at least 37% of children ages 7-14 were out of school, and in the conflict-affected areas government-provided education services were non-existent (MINED, 1990). Importantly, however, in searching for reform options, Gallardo had to take into account three constraints. First, the reform strategy had to fit with the larger focus on "modernization" that was a core characteristic of the Cristiani government, as already mentioned. For public services like education, this meant reducing bureaucracy and introducing new service provision arrangements that would ensure efficiency, effectiveness and accountability. Second, Gallardo had to circumvent the teachers' unions, who were not only anti-privatization but also allies of the FMLN. Third, the government saw it as necessary to incorporate the communities and schools linked with the FMLN into the official education system. 6 This third condition was particularly essential from the government's perspective because the civil war had not yet concluded at the time that Gallardo was searching for reform options, and thus the government was still looking for ways to subjugate the regions and schools controlled by the FMLN.
While the constraints specified in the previous paragraph and the preceding section flowed from the structural characteristics of the political-economic context of the 1980s, going forward, during 1990-1994, these constraints would, first, combine with the strategic action of key actors and, then, through that interaction, would lead to the entrenchment of policy priorities. Importantly for our purposes, the way in which vested interests crystallized around EDUCO in this period of transition ensured that EDUCO became a political priority for multiple actors-and political priority that would be furthered by the elaboration of impact studies during 1994-2005. The evolution of the period 1990-1994 can be distilled to three "moments." In the first moment, in mid-year 1990, a consultant from UNESCO was tasked with conducting an education sector assessment in order to understand the state of education in each of the country's 14 departments. This was a necessary task given that representatives of the government had not visited many rural parts of the country for 10 years due to the on-going conflict. By conducing this assessment, the consultant realized a reform model based in community management of education could be feasible because she observed first-hand that many communities (but especially in conflict-affected areas and in areas controlled by the FMLN) were already engaging in self-management of the education of their kids due to the lack of government-provided services (Gillies, 2010;MINED, 1990). In these areas, a community member offered to teach basic education in exchange for minimal, voluntary contributions (e.g., nominal payment or food donations).
Moreover, in FMLN areas, the schooling provided was grounded in the tradition of liberation theology and in the teachings of Paulo Freire in that it taught students to identify and to mobilize against the political-economic structures that ensured their oppressed position. For short, this approach is known as "popular education" (Hammond, 1998). But the larger point is that the Salvadoran government was threatened by the educational approach of the FMLN, as well as by the enduring low-level warfare.
The second moment spanned late 1990 to mid-1991. During this time, the suggestion from the UNESCO consultant was at first received with skepticism. The Minister of Education did not initially see this idea as viable in political terms because it was based on the management practices of FMLN-affiliated communities. For its part, the World Bank doubted that it would be possible to implement a community-management model when the majority of parents had low levels of literacy. However, after successfully testing this model in a pilot program-and thanks to the persistence of the UNESCO consultant-both the Minister of Education and the World Bank began to realize the opportunity that such a model represented. That is, it occurred to the Minister of Education of Education that this model met her constraints because it was innovative (i.e., it helped to "modernize" the MINED by dealing a blow to the central bureaucracy, which had traditionally been responsible to teacher management) and because it would allow her to incorporate and, thereby, to subjugate the FMLN communities. In addition, both the Minister of Education and the World Bank both saw a community model as a way to weaken the teachers unions because it would be possible to require that the teachers in this new program work on one-year contracts that would be renewable at the discretion of the community. And since these contracts would only affect those teachers working in this new program, the teachers unions did not put up sufficient resistance to block this reform model, as they had done with previously when the Minister of Education explored the possibility of adopting the voucher model from Chile (Edwards, forthcoming).
Separately, it dawned on the World Bank representatives working in El Salvador that this community-based model-if it were successful-would be a bankable policy in that they could sell it to other countries, especially since the international reform trends of the day heavily favored efforts towards the decentralization of central government functions and the "participation" of stakeholders in service provision (Edwards, 2012;Edwards & DeMatthews, 2014). But before moving ahead, two points are necessary here to illuminate the pressures and trends of the time. First, during the 1980s, and continuing into the 1990s, the World Bank other international organizations were preoccupied with the idea of decentralization (Edwards, 2012;Montero & Samuels, 2004), which was a popular idea in international development during the 1980s (Conyers, 1983). In the first of these decades, the World Bank had conceptual models that predicted increased efficiency and enhanced effectiveness from the decentralization of education systems (Winkler, 1989), though they did not have empirical evidence. Moreover, before the EDUCO program, it had not occurred to the World Bank that it would be possible to decentralize to the community level; they were instead focused on municipal or departmental levels. Second, internationally, in the late 1980s and early 1990s, much attention was drawn by specialists in in international development to the idea that "participation" was a central component of "good" development strategies (Cornwall, 2006;Leal, 2007), especially when it came to education policy (Edwards, 2012). This was reflected in March 1990 by the high-profile declaration of the World Conference on Education for All (WCEFA), which asserted that "partnerships at the community level … should be encouraged" because "they can help harmonize activities, utilize resources more effectively, and mobilize additional financial and human resources" (WCEFA Secretariat 1990, p. 58, as cited in Bray, 2003, p. 32). During this time, international organizations also highlighted the concept of participation. In particular, USAID raised the issue as a means to promote democracy and the effectiveness of public institutions (Montero & Samuels, 2004). The World Bank, in addition to emphasizing the issue of effectiveness, honed in as well on the potential of decentralization to result in more efficient and accountable service provision arrangements (Edwards, 2012).
For these reasons, by mid-1991, the Minister of Education and the World Bank decided to place the full weight of their support behind this model. This was a major turning point because it was impossible for any education reform to be successful without the backing of the World Bankthe institution that would finance the overhaul of the education sector's governance system. Thus, in May 1991, these actors agreed on a first loan for the education sector that included $10.3 million to solidify and extend the community management model, which, from then on, was known as "Education with Community Participation," or EDUCO, as previously noted. Over time, this program received support from the Inter-American Development Bank as well as from the World Bank. Indeed, these two institutions issued a total of $69.3 million in loans to ingrain, to scale up, and to improve the EDUCO program during 1991-2006. It is also notable that the decision to pursue this reform occurred well before the Peace Accords of January 1992, with a key implication being that this education governance reform strategy was selected before the war had ended and, concomitantly, before the official (i.e., public and democratic) post-conflict process of research and consultation had a chance to begin (Edwards, Victoria, & Martin, 2015).
The third moment extended from mid-1991 to 1994. In this period, through extensive and intensive World Bank technical assistance, the features of this program were refined. Not surprisingly, both the theory that guided the program's development and the practical arrangements that gave life to the program were guided by a neoliberal perspective on education management. To be specific, the key features of the program were as follows: a group of five parents from each community was elected to form a Community Education Association (Asociación Comunal para la Educación, ACE), an entity which was given legal status and which was charged with contracting and paying (with funds transferred from the MINED to the ACE via bank account) the teacher(s) that would work in their schools. 7 These ACEs were also thought to act as accountability levers that reduced teacher absenteeism and resulted in increased teacher effort in the classroom, especially since teachers were hired on one-year contracts. With the funds provided by the MINED, each ACE additionally had the responsibility of purchasing necessary didactic materials. These arrangements were assumed to reduce waste (thereby increasing the efficiency of provision) and to enhance student achievement (thereby increasing effectiveness). Given this arrangement, parents in disadvantaged rural communities were being asked to assume responsibilities traditionally borne by the State, and without compensation. In addition, community members were often expected to provide the time, labor, and materials either to construct or maintain their local school, particularly at the beginning of the program-and this was in addition to the informal school fees they already incurred to cover basic costs associated with schooling. The MINED, on the other hand, was responsible for the following: facilitating the creation of the ACEs; training the members of the ACEs in administrative and accounting procedures; setting the minimum criteria for teacher selection by the ACEs; designing and providing curricula; and overall coordination, supervision, and monitoring of the program as it was scaled up (Reimers, 1997;World Bank, 1994). 8 Thus, from late 1991 onwards, the Minister was committed to scaling up and widely implementing the EDUCO program, which soon became the only program through which the 7 While the EDUCO program was initially only intended as a strategy to provide education at the preschool level and in grades 1-3, it was subsequently expanded in 1994 to cover through grade 6 and then again in 1997 to cover through grade 9 (Meza, Guzmán, & Varela, 2004). After 2005, even some high schools became EDUCO schools (Gilles, Crouch, & Flórez, 2010). 8 For additional details on the technical aspects of the EDUCO program, see Gilles, Crouch and Flórez (2010).
Minister would allow educational access to be augmented. By 1993, the program was operating in all 14 departments of the country, and by 1995 the program had been institutionalized in the sense that the EDUCO office had been incorporated into the larger administrative structure of the MINED (Edwards, 2013b). In statistical terms, while the program began as a pilot with six schools in January 1991, it encompassed 114 schools by the end of that same year. By 2004, the program had 7,831 teachers and 378,208 students (Cuellar-Marchelli, 2003). In 2010, approximately 55% of rural public schools-which make up two-thirds of all schools in El Salvador-operated under the EDUCO program (Gillies, Crouch, & Flórez, 2010). The continued popularity and expansion of the program was due, among other reasons, to its perceived success and to the fact that the ARENA party-with its preference for neoliberal reform-retained the presidency until 2009. 9

The EDUCO Studies: A Critical Analysis
It should be clear that each of the involved actors discussed above had a vested interest in the success of the EDUCO program. The Minister of Education had bet her career on community-level decentralization and had made it the primary strategy by which she sought to reform the governance of the education system. For their part, the representatives of World Bank who were working on this program saw it as a major opportunity both to make a name for themselves and to develop a policy that could be useful to their organization more broadly, that is, a policy that could be sold as a best practice to other countries. 10 Yet, despite the excitement around this program during the early 1990s as it was being scaled up, the MINED and the representatives of the World Bank knew that they would need solid evidence which demonstrated that the program produced beneficial outcomes in order (a) to continue to promote the program as a central policy for education reform in El Salvador and (b) to be able to credibly promote the program internationally as a best practice (Edwards, 2013b). It was in this context of excitement and determination that the World Bank began to carry out evaluations of EDUCO.
With the above in mind, the remainder of this section aims to critically evaluate six key studies that were carried out between 1994 and 2005 by the World Bank on the EDUCO program. These six studies were selected for critical review because they represent all the studies that the World Bank produced as impact evaluations. In other words, in accordance with the methodology of this article, and given that no other entity or actor generated any impact evaluations on EDUCO, these six studies represent the entire body of "legitimate" and "policy-relevant" knowledge that was created in order to evaluate whether the program worked in causal terms. 11 While these studies were 9 See Edwards (2013b, pp. 54-55) for an update on the EDUCO program in the post-2009 time period. 10 On this point, it is necessary to recall the tremendous popularity of decentralization as a theme within the international development industry during the 1990s (Edwards & DeMatthews, 2014) as well the predominance of the World Bank during this time vis-a-vis other multilateral organizations, especially in the education sector (Mundy, 1998). 11 Going forward, it should be noted that there are additional EDUCO studies which were not included in the present review. These additional studies do not fit neatly into a pro-EDUCO trajectory and, whether due to their findings, methodology, authorship, timing or other reasons, they have not received the same kind of attention, nor have they wielded the same kind of influence. In contrast to the highly-cited and impactful studies critically re-assessed here, these other studies have employed methods which appear to be less rigorous and/or have offered findings which were less sanguine in nature. These studies include: Lindo Fuentes (1998) which presents findings from a qualitative case study of EDUCO funded by the World Bank; Sastry, Melamid and Ross (1995), which contains a cost analysis of the EDUCO program, also funded by the integral to the continued success of EDUCO-within and outside El Salvador-during the latter 1990s and 2000s, neither a comprehensive nor a critical analysis has been carried out of their methods and findings. In carrying out such an analysis, the goal, on one hand, to provide new insights into EDUCO's knowledge base; that is, the purpose is to further inform what is known about the program. This is useful insofar as EDUCO in particular and community-level decentralization in general continue to be strategies that are considered for implementation by development specialists and policymakers around the world. On the other hand, through the analysis below, the goal is also to engage in a larger discussion of the role, implications and limitations of impact evaluations as a key element in the political economy of education reform.
In what follows, the results of the critical analysis are presented. First, the context, methods and findings of each study are presented. Then, where appropriate, the limitations of each study are detailed. The subsequent section of this paper will reconsider what can reasonably be asserted with regard to the effects and lessons of EDUCO. The discussion below is summarized in Appendix 1. Just before the end of 1994, the World Bank produced a report that contained the results of a small evaluation of the effects of the EDUCO program. Despite the limitations of the study, the results received significant attention, given that they were the first quantitative data available which could claim to approximate the program's impact (World Bank, 1994). With a sample of 78 communities, the World Bank applied an ex-post comparison 12 between two groups (EDUCO vs. traditional rural public schools) as well as ordinary least squares (OLS) regression, all using data collected in 1993. According to this report, and with reference to the two-group comparison, the evaluation showed better math test results for students in grades one and two, but it is important to note that no statistical controls were applied. Background characteristics (and only very few) are taken into account in just one instance-related to preschool academic achievement, and in this case the EDUCO variable is not significant. What was significant, rather, was family income (with there being a positive relationship between this variable and student achievement). While the larger point here is that the study lacked rigor, this did not prevent the Minister of Education and the World Bank from pointing to these results as evidence of EDUCO's success. Not surprisingly, the appearance and reliance on these results coincided with the timing of a major political decision by the Minister of Education to continue and to further expand the EDUCO program through additional World Bank loans. Later years would see the production of five additional (and more rigorous) studies, and these would constitute the international evidence used to promote the program, as discussed below. Parandekar (2002), a quantitative study by a World Bank researcher that looks for lessons generally from EDUCO related to the decentralization of education; Reimers (1997), an independent study of the characteristics and implementation of the EDUCO program; and Cuéllar-Marchelli (2003), another independent study that assessed the overall experience of EDUCO along various dimensions.Each of which has been carefully reviewed in Edwards (2013b). Other, more recent reports and studies that document the operation of the EDUCO program are:  Bank (2009). All of studies cited here are available from the authors. Due to the present focus on impact evaluations, these other studies were not included here, though future research should further explore their insights. 12 Meaning that the EDUCO schools and the traditional schools were selected for comparison after the EDUCO intervention began.

Umanzor et al. (1997). El Salvador's EDUCO program: A first report on parents' participation in school-based management. Working Paper Series on Impact Evaluation of Education Reforms, No. 4. World Bank.
The second study was conducted by the World Bank and a few staff from the MINED. It was an impact evaluation and it relied on data collected in 1996 (most future studies would return to this data as well) through a random sample of 311 schools. The population of schools from which the sample was drawn had to have a third grade section with at least 10 students. At each school, in addition to administering language and math tests to third-grade pupils, the team interviewed and gave a questionnaire to the director (or person in charge), the third-grade teacher, five randomly selected fifth-grade students, their parents, and two members from the ACE. The questionnaire collected background data. Although the sample initially included four types of schools -i.e., EDUCO, mixed, 13 traditional public, and private -the 1997 impact study only looked at EDUCO schools and traditional public schools located in rural areas.
Underlying this data is an important feature to note before continuing with the study's results. Though the EDUCO sample was identified in a random way, this is negated by the fact that the MINED initially chose the communities which would participate in a non-random way. That is, as discussed previously, the MINED purposefully selected the poorest and most disadvantaged communities for the program (economists refer to this strategy as endogenous program selection), thus indicating that selection bias would be a problem inherent to the quasi-experimental studies conducted later.
With this data, Umanzor et al. (1997) ran a series of ex-post matched comparisons between the EDUCO schools and the traditional public schools on a number of characteristics of interest. They focused on associations between school and family characteristics and the outcomes observed in the school. They found both similarities and differences in general characteristics across groups. For example, on average, and to a statistically significant degree, EDUCO teachers had a higher level of education, but less experience. This is because, at the outset, ACEs were required to hire teachers who had at least a university level teaching certificate. The excess supply of university-trained teachers in the early 1990s also contributed to this. EDUCO schools also had less access to electricity, but more of them had classroom libraries (most likely because a separate USAID program funded classroom supplies). Despite better libraries, EDUCO schools did not have more textbooks. Additionally, EDUCO students were found to live in much more rudimentary homes (as in, without electricity or piped water, but with dirt floors), yet parents were not found to help with homework any less than parents in the comparison group. In terms of school-community interaction, ACE members felt that they had more influence in school decisions, though they did not actually make more decisions in the school's management (Umanzor et al., 1997). ACE members also visited their community's school more often than did parents on the school councils of traditional rural schools. Likewise, EDUCO teachers met with parents more often and contacted parents more frequently when students were absent. Lastly, with regard to the variables that have 13 Until 1999, the MINED allowed EDUCO and traditional schools to co-exist through what it called 'mixed' schools (Lindo Fuentes, 1998). Here, a traditional school would also receive funding from the EDUCO program and the community would establish an ACE. With the funding received, the community would hire a teacher for lower grades (grades 1-3). The school would then use its non-EDUCO budget to pay a teacher to provide grades 4-6. As mentioned, the EDUCO program was initially targeted to communities where educational services were not provided during the civil war. However, as the MINED began to scale up the program it did so by creating 'mixed' schools, as well. received the most attention, there were differences in neither student absenteeism, dropout, grade repetition nor student achievement on either the math or language tests (Umanzor et al., 1997).
Some of these differences are cause for optimism -as with ACE parents being more involved in the school and EDUCO teachers meeting more with parents. Nevertheless, educational interventions are assessed primarily in terms of their ability to increase student achievement. Moreover, in this case, the theory of the EDUCO program is in part based on the notion that community-level management would result in increased teacher effort, which would lead to better student learning. Although the comparison of the treatment and control group in this study did not show that EDUCO students had significantly better achievement than students in rural traditional public schools, these students also did not score significantly worse than their counterparts in public schools. Umanzor et al. (1997) took this as a positive sign, given the relatively worse economic conditions of EDUCO students, and thus conclude that EDUCO students thus exhibit "better than expected academic performance" (p. 26). To be sure, and as will be discussed, the conclusions by Umanzor et al. (1997) are the most justifiable of the studies reviewed here. That said, it may be that the positive results observed were not the result of the EDUCO program, per se, but rather a result of the pre-existing social capital that had accumulated during the civil war in the EDUCO communities (many of which were FMLN communities, at that time), when they were forced by their circumstances to work together to provide basic services, such as education (ADES, 2005;Aguilar, Becerra, Burgos, Castro ,& García, 1993;Alvear Galindo, 2002;Cruz, 1994Cruz, , 2004MINED, 1990MINED, , 1994.

Jimenez, E., & Sawada, Y. (1999). Do community-managed schools work? An evaluation of El Salvador's EDUCO program. The orld Bank Economic Review, 13 (3), 415-441.
The third study -by Jimenez and Sawada (1999) -emerged two years later. 14 Here, Jimenez and Sawada (1999) employed more advanced econometric techniques to assess whether EDUCO had an effect on either student achievement or student absences. In order to answer these questions, they used the same 1996 data as Umanzor et al. (1997) and applied regression analyses along with the Heckman two-step correction (Heckman, 1979) in order to attempt to correct for the sample selection bias inherent in the EDUCO treatment group. 15 In their study, they successively added 14 It should be noted that Emmanuel Jimenez was at that time a member of the World Bank's influential Development Economics Vice-Presidency, and was well respected among his peers. The fact that he took up the cause of EDUCO by doing research on it attracted the attention of others within the Bank. 15 The Heckman two-step correction is employed where there are issues with sample selection bias. Sample selection bias is normally a problem for regression analysis because, when the treatment and comparison groups systematically differ on a range of characteristics, we cannot be sure that the effect observed for the treatment (e.g., EDUCO) is unbiased. That is, the effect (or lack of it) could be due to unobserved variables related to the treatment (labeled by economists as endogeneity). Under endogeneity, we can also not be sure of the extent to which the estimates for the other variables in equation are biased as well (Berk, 1983). The best way to address this issue is to first draw a representative random sample of participants from the population of interest. In most cases this is not possible, especially in social science research. Such was the case with EDUCO, where participating communities were not chosen randomly and where participating communities differed systematically from a nationally representative sample.
Thus, to deal with this common issue, James Heckman developed his now famous correction technique which treats sample selection bias as specification error (Heckman, 1979). Per this technique, a separate estimator term is added to the principal regression equation of interest (i.e., the "substantive equation"). This term is arrived at by estimating through a (probit) selection equation the likelihood of a student, for example, being placed in an EDUCO school, given a set of selected characteristics. This controls for student, family, teacher, classroom, school, and community participation characteristics. Throughout, the comparison group is traditional rural schools.
In their discussion of student achievement, the authors state: "EDUCO's … effect on language test scores is positive and significant, while its effect on math performance is positive and not significant" (Jimenez & Sawada, 1999, p. 428). Yet, this statement is quickly followed by a concession: "Our measure of EDUCO's advantage in language may be imprecise. The estimate of the EDUCO coefficient is sensitive to the specification of the participation equation" (p. 428). What they allude to is that their more completely specified models clearly show that the results for both language and math achievement are not significant. Nevertheless, the authors make inferences in the concluding section to their paper from models missing controls for school inputs or community participation. They state: "[EDUCO] has improved language scores" (p. 440).
In addition to the specification issue, there is the matter of endogeneity stemming from sample selection bias. That is, the results provided by Jimenez and Sawada (1999) are confounded by the fact that the students who attended EDUCO schools also possessed unobserved characteristics that correlated with student achievement outcomes. Jimenez and Sawada (1999) recognize this when they write, "The error terms of the participation and achievement equations are negatively correlated. This means that EDUCO students have unobserved characteristics that are negatively correlated with achievement test scores" (p. 428). Under conditions of endogeneity, one cannot truly parse out the impact of the intervention on student achievement because the participation of the student in the intervention is correlated with other factors which are also correlated with student achievement.
Additionally, alternative conclusions can be drawn from the student achievement regression results, keeping in mind their limitations. In math, for example, although the EDUCO variable is not significant, what does contribute to improved test scores is the availability of sanitary services. With regard to language test scores, it is having a classroom library that contributes positively. For both math and language, more frequent visits to the school by ACE parents is positively associated with probability, then, is fed back into the substantive equation as part of a ratio (the inverse Mills ratio) that represents the probability that a child with given characteristics will be excluded from the sample from which results are generalized, conditional on participation in EDUCO (Berk, 1983). Ideally, then, the added estimator is a sample selection correction term that adjusts for the fact that certain characteristics are overrepresented in the sample being examined. From the regression output, one can then interpret a statistically significant result for this term's coefficient as indicating that there is sample selection bias. That is, a statistically significant coefficient signals that certain characteristics are over or under-represented.
Fundamental issues surface in practice, however. For a thorough review of the many issues that can and do arise in practice through the use of the Heckman two-step correction, see Bushway, Johnson and Slocum (2007). For example, if the same variables appear in both the selection equation and the substantive equation, the issue of multicollinearity arises (Bushway, Johnson & Slocum 2007). That is, in concrete terms, if the same variables used to predict a student's participation in EDUCO are also used to predict a student's test scores, then the correction term added to the equation for test scores may well co-vary with the other independent variables. What's more is that this technique assumes that the error terms for both the selection and substantive equation are jointly normal, meaning that they are independent of each other (Heckman, 1979). However, the error terms will be correlated if there are unobserved characteristics (omitted variables) relevant to the dependent variables of both the selection and substantive equations. It should be noted that, in the case of EDUCO, this is almost certainly the case. The reason is as follows: Socioeconomic conditions determined who participated in EDUCO; these same poverty related characteristics always influence student achievement. Ultimately, then, although this fix works well in theory, in practice it actually worsens issues of misspecification because there are now two linked regression equations to specify correctly, and which themselves must not be correlated.
higher student achievement. The data thus support the conclusions that schools in which parents are involved and which possess basic resources encourage better student performance. Given that the focus of the evaluation was to determine EDUCO's impact, these findings are not highlighted.
The second question posed by Jimenez and Sawada (1999) is whether there is an EDUCO effect on student attendance. They use the same data and controls as with their models for student achievement, changing only the dependent variable to indicate the number of days the student was absent in the month prior to the survey. Here, they find that EDUCO significantly reduces the number of days a student was absent. They thus conclude that the accountability relationships instituted between parents and teachers which EDUCO are responsible for reducing student absences.
However, if we decompose the effect of the EDUCO variable according to when the EDUCO schools were built, the picture changes (Jimenez & Sawada, 1999). If the EDUCO school was built between 1991-1995, there was no effect; however, if the school was built in 1996, it significantly decreased student absences. As opposed to attributing the effect to relationships of accountability, the result could simply be due to the fact that students received a new school in the same year that the survey was administered. The presence of a new school -where there had not been one previously -could significantly increase student attendance.
Nevertheless, the authors chose to highlight a particular conclusion, one that resonated with the theory of community-level accountability relationships and their ability to lead students and teachers to perform better (Jimenez & Sawada, 1999). These conclusions are powerful, particularly given the fact that they are offered by the World Bank and the fact that they were derived from methods that are widely seen as rigorous and objective. To that end, the literature shows that other authors have repeated these conclusions, as did Barrera-Osorio, Fasih, Patrinos, and Santibañez (2009) in their review of the effectiveness of school-based management in developing countries. In this way, over time, conjecture can contribute to consensus around the idea that community-level decentralization can lead to fewer student absences. The fourth study was conducted by Sawada (2000) and attempted to determine if the EDUCO program had an effect on student achievement, teacher pay, teacher effort, or classroom inputs. Numerous models were estimated using largely the same data and controls as were employed by Jimenez and Sawada (1999), the difference being that Sawada (2000) uses instrumental variables in an attempt to remedy the endogeneity which results from the nature of the sample. 16 In the end, four conclusions were drawn, each of which is reviewed here.
The first conclusion stated that community participation was positively related to teacher wages. However, although Sawada (2000) runs eight different models to test this, under the most complete specification neither the community participation variable nor the EDUCO variable is significant. 16 Instrumental variables are employed in regression analysis under conditions of endogeneity. The idea is that one seeks to "find a variable (or instrument) that is highly correlated with program placement or participation but that is not correlated with unobserved characteristics affecting outcomes" (Khandker, Koolwal, & Samad, 2010, p. 87). Instrumental variables must be selected thoughtfully, as a weak instrumental variable -one that is correlated with unobserved characteristics or omitted variables -can worsen the bias of coefficient estimates. Ideally, the instrumental variable serves as a proxy for program participation while eliminating issues of endogeneity. See the previous footnote for more on endogeneity.
The author refers in his conclusion to an OLS regression which did not control for teacher characteristics. Moreover, in this particular model, while it is true that community participation (defined as the number of times parents on the ACE visit the school monthly) is positively associated with increased wages, what Sawada (2000) chooses not to mention is that EDUCO is significantly and negatively related to wages. Thus, while Sawada (2000) finds a significant effect of community participation, this is not related to the EDUCO program. In contrast, the EDUCOrelated finding -that EDUCO teachers receive less compensation than their traditional school counterparts -is not highlighted.
Second, Sawada (2000) highlights the relationship between teacher effort and community participation in EDUCO schools. Here, the model does indeed show a statistically significant relationship between school visits by ACE members and the number of hours teachers spend each month meeting with parents. However, the latent constructs, as measured and specified here, experience simultaneity and, as such, it is not possible to separate the effect of community participation from teacher effort. This is not to say that there is not a relationship between the two variables. To be sure, parents spending more time at school would be intertwined with the number of hours teachers spend meeting with parents. What remains unclear is the nature and directionality of the relationship. Nevertheless, Sawada (2000) concludes: "Community participation seems to enhance the teacher effort level" (p. 26).
For the third finding, Sawada (2000) reports that there is a connection between community participation and teacher-student ratio. In particular, the finding is that the former causally decreases the latter. This is a curious findings because the relationship between community participation and teacher-student ratio is not clear. That is, one wonders why teacher-student ratio would be a function of community participation. While Sawada (2000) does not offer an explanation of why this would be, it is suggested here that this may be the result of the nature of the sample: EDUCO schools were located in particularly rural areas, where there tended to be fewer children per teacher.
Fourth and finally, Sawada (2000) speaks to student achievement. Although the EDUCO variable was not in any of the model specifications significantly related to test scores, the study is bookended by statements which, at first glance, indicate the contrary. At the beginning and at the very end of the paper, Sawada (2000) writes, "We observe consistently positive and significant EDUCO participation effects on standardized test scores" (pgs. 1, 25, emphasis added). What he refers to, however, is not a variable that represents parental participation in EDUCO, but rather a variable for parental participation generally. The problem with this wording is that the parental participation variable which was shown to be associated with student achievement is not specific to EDUCO. Instead, it is a variable that represents participation in schools generally -by either parents of EDUCO students or parents of students in traditional rural public schools. 17 Such statements can filter through to future reviews of empirical work on EDUCO and decentralization more generally.

Jimenez, E., & Sawada, Y. (2003). Does community management help keep kids in schools? Evidence using panel data from El Salvador's EDUCO program. Discussion Paper, Center for International Research on the Japanese Economy. Tokyo, Japan: University of Tokyo.
In the fifth study, Jimenez and Sawada (2003) attempt once more to uncover an EDUCO effect. This time, they return to the question of whether participating in EDUCO is associated with staying in school over the two-year period 1996-1998. To answer this question new data collected in 1998 (from student learning tests and teacher and parent surveys [Lindo, 1998, p. 21]) is added to the 1996 data and then run through regression analyses (specifically, they use probit models) which utilize the Heckman two-step correction for sample selection bias and the same controls as their other studies (Jimenez & Sawada, 1999;Sawada, 2000).
Once again, the authors acknowledge the inherent limitations they face due to sample selection bias and the effects of this on their results. They reveal that the error terms for the selection and substantive equations are negatively correlated. As a result, Jimenez and Sawada (2003) note that "unobserved characteristics regarding the children, households, and communities might positively affect the likelihood of selecting an EDUCO school and might negatively … affect a student's decision to continue in school" (p.21-22). The implication is that their findings may suffer from endogeneity and, as such, may not be valid. Nevertheless, the findings of Jimenez and Sawada (2003) can be found cited in the literature on education decentralization in developing countries. One example is Barrera-Osorio, Fasih, Patrinos and Santibañez (2009), who cite results from model 1 in the study by Jimenez and Sawada (2003). While the results cited indicate that attending an EDUCO school increases the probability of remaining in school for two years by 64%, we cannot really be sure due the issue of endogeneity.

Sawada, Y., & Ragatz, A. (2005). Decentralization of education, teacher behavior, and outcomes: The case of El Salvador's EDUCO program. Incentive to improve teaching: Lessons from Latin America. (pp. 255-306). Washington, D.C.: World Bank.
The sixth and final study came about in 2005, this time by Sawada and Ragatz (2005). It builds on Jimenez and Sawada (1999). In this incarnation, the study asks how decentralization affects administrative processes, teacher behavior, and student achievement. In addition to using the Heckman two-step correction procedure, it also makes use of propensity score matching -a strategy through which the compared treatment and control communities are matched along a series of characteristics to ensure that they are as identical as possible, given the communities in the sample. In their analysis, Sawada and Ragatz (2005) eliminated the mixed schools included by Jimenez and Sawada (2003), thus reducing the sample to 37 EDUCO schools and 96 traditional schools. In some models, the same controls as before were used. Those instances where the controls were different are discussed below.
The results are interesting in that they are not what the authors expected. In terms of administrative processes, the only activity to occur more often at the school level in EDUCO schools concerns hiring and firing teachers (Sawada & Ragatz, 2005). Neither determining teacher salaries, giving teachers incentives, evaluating teachers, nor spending school money occurred more often at the school level in EDUCO than in traditional public schools. Not surprisingly, the results show that teachers' association relations occurred more often in traditional rural public schools. This was the case because teachers hired by ACEs are non-union, and thus there was no teachers' association activity to observe.
For teacher behavior, interpreting the results is like reading an Ouija board. There are models for 19 different dependent variables. Each dependent variable tells us something different about teacher behavior -for example, how many hours per week the instructor teaches, how many hours s/he meets with parents, or how often s/he is absent. Further complicating the picture is that each model is then run four different ways -using (a) standard (that is, Ordinary Least Squares) regression with 42 independent variables, (b) the Heckman two-step procedure, (c) revised OLS regression with 32 independent variables, and (d) propensity score matching with 32 independent variables. 18 The problem across these 76 different combinations is that there is tremendous variability in terms of what is significant. Those results that are significant under propensity score matching with 32 controls are then nonsignificant in the OLS regression with 42 controls, for example. Elsewhere, the Heckman models return significant coefficients, but, in these cases, none of the other models do. In the end, there is not one outcome variable that is significant across all four forms of regression. Sawada and Ragatz (2005) still report, however, that "many key measures survived the bias tests, including meeting with the parents, teacher absence, and hours dedicated to teaching" (p. 295).
In addition to the above finding, the authors offer two additional conclusions based on the regression results. The first is as follows: The results indicate that community participation seems to enhance the teacher effort level, possibly because of intensive monitoring of teacher behaviors and the implicit threat that exists because hiring and firing of teachers occurs at the community level (Sawada & Ragatz, 2005, p. 295).
And, second: "Community participation not only uses relevant information that outside government agencies are not likely to have, but also imposes commitment on teachers, which leads them to exert greater effort" (p. 295).
It is suggested here, however, that these assertions are not borne out by the regression results, given the variability and inconsistency across models discussed above. It seems that the authors were either confident in the regression results or had a predilection to explain the results in terms of accountability relationships that resonate with the World Bank's theory of community-level education management decentralization, or both. In the end, and in contrast with the interpretation by Sawada and Ragatz (2005), it is suggested here that the true effect is not known for any of the significant variables presented by the authors. As Klees and Edwards (2014) assert: Regression analysis theory … offers no guidance as to how inaccurate the resulting regression coefficients are under real-world misspecification conditions… . With so many alternative specifications from which to choose, regression analysts can almost always find their favorite variable to be significant, and thus the empirical literature becomes a debate over who has the best specification (p. 34).
The final question of this study pertained to student achievement. This time, the results look promising, for language scores. We are presented with the output for a handful of coefficients across four specifications. Three of the four models show significant effects of being in an EDUCO school compared with being in a traditional rural public school. What the models do not show, however, are the authors' controls, though the footnote to the table does note that child and household characteristics were taken into account. Yet what the authors choose not to mention is that many of the available controls were not applied. That is, Sawada (2000) and Jimenez and Sawada (1999) ran the same models with additional controls for teacher quality, classroom quality and school quality, and under those circumstances they did not find an EDUCO effect on student achievement. One wonders if Sawada and Ragatz (2005) purposefully specified their models to generate significant effects. There is no way to know for certain. Nevertheless, the fact that they chose not to use the available controls puts in question the conclusion they offer: That "EDUCO program governance leads to better effort of teachers, which improves educational outcome" (Sawada & Ragatz, 2005, p. 299). 19 Many of the interpretations offered by Sawada and Ragatz (2005) raise concerns. Nevertheless, one conclusion suggested by the data is that ACE parents, compared with parents on the school councils of traditional rural public schools, have and use the ability hire and fire teachers more. Given the provisions on which the EDUCO program rested all along, this is to be expected. Other findings from this study are more controversial and uncertain, and seem to be intended to positively skew the perception of EDUCO.

Discussion: What Do We Actually Know About EDUCO?
The previous section clearly calls into question the findings that have been asserted in the most widely-known and widely-cited studies of EDUCO. In light of the critical review presented above, these studies can interpreted as repeated attempts to find evidence that supports the beneficial effects of a particular version of decentralization. It was shown that World Bank researchers and publications, in many instances, offered interpretations which were questionable (based on the limitations of their methods and data and the possible alternative explanations). In view of the reappraisal above, which is summarized in Appendix 1 to this paper, it is argued here that not one of the studies reviewed definitively shows that the EDUCO program produced positive and significant results on the variables of most interest to the World Bank and the MINED. To the contrary, it is concluded that the EDUCO program does not lead to improved student achievement, student attendance, student retention, or teacher absenteeism when the available controls are applied.
The one finding which stands out across the studies is that EDUCO schools did no worse (in statistically significant terms) on student achievement than traditional rural public schools. EDUCO proponents have consistently interpreted this as a positive finding: since EDUCO schools were located in the most disadvantaged areas of the country, one might have expected them to perform much worse than their public school counterparts. However, given that EDUCO schools performed just as well (i.e., no worse), EDUCO advocates conclude that, under the EDUCO model, we can introduce an arrangement that is more accountable without sacrificing quality.
Yet one cannot be sure that the effect is due to the EDUCO model. First, it must be remembered that the data for these studies were collected during 1993-1998; second, it must be remembered that EDUCO schools were, at the outset, purposefully targeted to FMLN communities (Reimers, 1997) as a way to undermine the system of popular education that had developed during the civil war (Edwards, 2013b, p. 224). It should be noted that the popular education system was not only present in the areas controlled by the FMLN but it was also substantial in its reach: by the early 1990s, this system included approximately 1,000 teachers and 13,500 students (Alvear Galindo, 2002, pps. 189, 200), although according to Reimers (1997) this system served as many as 17,000 students. The implication is that, during the time when data was being collected for the reviewed studies, the communities included in the sample were in fact FMLN communities that had previous experience with self-organization through the system of popular education as well as through cooperative and democratic forms of agricultural production, communication, housing, and community government (ADES, 2005). This realization is problematic for the conclusions drawn by EDUCO supporters because it means that the findings which they have attributed to the EDUCO model could rather be the result of the practice with self-organization that these communities had during the civil war as part of their efforts to prepare for the reality of a socialist and cooperative society after the war, assuming the FMLN won (Montgomery, 1995). Put differently, as opposed to attributing the results to EDUCO the results could just as likely be attributed to the pre-existing social capital possessed by the communities which participated in EDUCO.
Given the plausibility of this alternative explanation, there are now no results which suggest the superiority-or even the equivalence-of EDUCO to traditional rural public schools, particularly in terms of such outcomes as student achievement, student attendance, student retention, or teacher absenteeism. With regard to parental participation, Umanzor et al. (1997) did show that parents in EDUCO schools visit their child's classroom more times per month (5.7 times vs. 1.6 times) and spend more time meeting with teachers (EDUCO teachers spent 4.9 hours/week meeting with parents, versus 2.9 hours/week for non-EDUCO teachers). But again, these findings may be spurious because the groups of parents being compared were likely not the same in terms or their social capital, as explained above. The results may have been equivalent for EDUCO and traditional rural schools if the parents from the latter group had the same pre-existing social capital.
Interestingly, however, in terms of lessons learned, our reappraisal of the evidence actually suggests a contrary conclusion to that offered by the authors of the studies. To reiterate, if we recall that the "EDUCO communities" in the impact evaluations are FMLN communities that had adopted the EDUCO program, then any positive benefits observed may be a reflection not of the design of the EDUCO program but rather a reflection of the experience of cooperative selforganization and the social capital that consequently accumulated with the parents who, under EDUCO, served on the school councils. Seen in this light, the impact evaluations of EDUCO are, ironically, evidence in favor of the opposition, that is, evidence in favor of those same communities that the government sought to undermine, and which the government did in fact undermine through the requirement that EDUCO schools could only hire certified teachers, which necessarily prevented the ACEs from hiring popular education teachers, since they had low levels of education without the credentials specified under EDUCO's regulations. But the fact that EDUCO successfully subjugated FMLN communities to the official system-despite proposals from popular educators to allow for two parallel systems (Aguilar, Becera, Burgos, Castro, & García, 1993)-does not erase the social capital that accrued to parents nor their prior experiences with self-organization, both of which informed and arguably improved their work under EDUCO when it came to hiring and supervising teachers and managing the school budget among other tasks. Thus, the findings by Umanzor et al. (1997) that EDUCO parents visited their children's classrooms more often and spent more time meeting with teachers is likely the result of efforts at community organization that go back to the 1970s, when rural Salvadorans began to mobilize in response to governmental repression and in accordance with the critical teachings of liberation theology (Montgomery, 1995). These same communities later became affiliated with the FMLN and put into practice critical teachings through their self-managed community schools and the methods of critical educators such as Paulo Friere. Ultimately, then, this critical review suggests that the experience of EDUCO supports forms of community education management that are rooted in broad, community-wide cooperative organization, solidarity and popular education, as opposed to a more narrow, neoliberal version based on accountability, student test scores, the ability to fire teachers, and abstract notions of efficiency. These two versions of community participation in education governance have been referred to as progressive and neoliberal, respectively (Edwards & Klees, 2015).
In sum, then, the same studies which have supposedly shown the benefits of the EDUCO program can be interpreted differently to offer alternative conclusions. These alternative conclusions clearly cast doubt on the advisability of EDUCO as a public policy. Furthermore, by placing our critical review in the political economic context of the 1990s, it is suggested that the alternative interpretations offered here indicate that EDUCO, on the whole, was: (a) a program which did not improve key indicators of quality, (b) imposed costs on the rural parents (who volunteered their time on the ACEs and who helped to build and repair the schools in their communities), (c) sought to weaken the teachers unions (by instituting one-year contracts for EDUCO teachers), and (d) was directed at incorporating FMLN communities into the official education system (by replacing their popular education teachers and schools with the EDUCO program and with teachers from outside their communities). Of course, this alternative interpretation highlights the political motivations behind EDUCO. But it needs to be remembered that EDUCO was always a political decision. Moreover, and not surprisingly, it has been in the realm of politics that the EDUCO program has continued to have relevance, thanks in no small part to the studies reviewed here, as is discussed in the next section.

National and International Implications
Within the particular political-economic context of the 1990s and 2000s, these studies have contributed to numerous developments, both inside and outside El Salvador. In what follows, the implications are discussed in national and then international terms. Keep in mind that the implications discussed below are not presented as being the result of the EDUCO studies alone; rather, it is suggested that the EDUCO studies are one important factor among many within the historical context characterized earlier in this article.

National Implications
First, within El Salvador, these studies initially influenced the decision in 1994 to scale up the program countrywide, and then later served to maintain the focus of the MINED on EDUCO, even after the transition to a new political administration in 1998 and again in 2003. Gillies (2010) has also noted this latter point. Of course, the other side of the coin is that these studies, and the justifications they offered, constituted significant obstacles to overcome for those who opposed the program, particularly because of their econometric nature, and thus the perception that their findings were definitive, not to mention the difficult-to-unpack nature of the methodology itself. At once, then, these studies provided political capital to those in power while also raising barriers to reform for those who sought to modify or repeal the program.
A second (and ironic) implication is that these studies-which popularized EDUCO for enhancing teacher accountability, for reflecting a neoliberal logic with regard to efficiency and effectiveness-actually helped to extend a model with roots in the FMLN's approach to development, which was based in solidarity, community empowerment, and popular education (ADES, 2005;Edwards & Ávalos, 2015). However, as noted, the FMLN model was sanitized and reframed according to market-based principles. Amid the focus on this new management model, what received less attention was the fact that, through EDUCO, a second-class system of schools was created that continued to marginalize large portions of the student population. This outcome resulted from the fact that, in addition to hiring less-experienced teachers, EDUCO communities, which already tended to be located in poor rural areas, were also expected to voluntarily contribute to the construction, maintenance, and management of their community school (note that none of the parents on the ACEs received compensation for their time).
More generally, and perhaps more perniciously, these studies contributed to the development and general acceptance of myths around the EDUCO program. In El Salvador, during the last 20 years, it has become common to espouse the notion that EDUCO is a program which introduces teacher accountability and which produces better student achievement, among other things. By repeating these assertions, researchers, political actors and education commentators have helped to spread a kind of "common sense" around EDUCO. The present article is an attempt to provide an alternative narrative to the "common sense" that heralds market-based forms of accountability.
Third, in that EDUCO was the education reform superstar of the ARENA party, it stands to reason that this program-and its positive evaluations-contributed to this party's continued success in national elections, even if only as one element among many.

International Implications
Internationally, the reviewed studies have been used to promote the implementation of a model that not only transfers to the community part of the responsibility for education governance, but also transfers part of the cost of educational provision. That is, while studies such as those on EDUCO have provided the justification for countries to experiment with community-level decentralization, in practice, these studies have served as political cover for the introduction of decentralization policies that require (whether officially or unofficially) that community members (a) make donations to the school, (b) help with building and repairing school buildings and other structures (such as latrines), (c) contribute to school management (e.g., through monitoring teachers and principals), and (d) participate in administrative or supplemental tasks (e.g., the development of school improvement plans or the provision of school snacks for students) (Edwards & DeMatthews, 2014).
The second international implication, which relates to the first, is that these studies have promoted the belief among development practitioners that EDUCO-style community decentralization is a preferred reform for education systems around the world. A prominent example is the World Bank. As the studies reviewed here began to be produced, they garnered attention outside of the staff members who initially worked on the EDUCO program, including prominent staff within this organization. Indeed, as Gershberg, Meade, and Andersson (2009) note, based on personal observation, " Jimenez and Sawada (1999) was influential in convincing World Bank staff that EDUCO was a successful model" (p. 198). 20 Moreover, the evidence base constituted by both Jimenez and Sawada (1999) and Sawada (2000) grabbed the attention of the authors of the World Development Report 2004(WDR 2004, which is the flagship publication of the World Bank (World Bank, 2003). In this Report, which has come be regarded as a guide for how to introduce relations of accountability in public sector reform in developing countries, the EDUCO program was highlighted as the key model to emulate in the education sector. By being included in the WDR 2004, EDUCO's reach among development professionals was thus amplified, thereby increasing the influence of this model on the global education reform agenda. Were it not for the findings offered in the EDUCO studies, this would not have been possible. For additional examples of the ways that EDUCO was invoked in the 1990s and 2000s, after these studies began to come out, see Appendix 2.
Third, in raising the profile of the EDUCO program, the EDUCO studies also helped to raise the profile of the personnel from the MINED who were instrumental in institutionalizing and implementing the EDUCO program during the 1990s. For example, following Cecilia Gallardo de Cano, who was the Minister of Education (1990Education ( -1998 when EDUCO began, three of the four ministers of education during 1998-2009 (after which point the presidency passed from ARENA to 20 Alec Gershberg served as a senior education economist for the World Bank during the late 1990s. the FMLN) previously either worked for the EDUCO program or otherwise contributed to its furtherance. 21 Of course, without the positive evidence presented by the EDUCO studies reviewed here, these actors would have lacked the ability to claim success, or to have success attributed to them, thereby making it more difficult to generate the political capital necessary to assume such high-profile positions. This not to say that it would have been impossible for such actors to reach such prominent positions, only that it would have been less likely.
Fourth, these studies helped to generate interest from other countries. Governmental study groups came from, at least, Argentina, Bangladesh, Brazil, Chile, Equatorial Guinea, Honduras, Niger, and Thailand, as well as additional study groups from Africa, the Caribbean, and Europe (Meza, Guzmán & de Varela, 2004a). MINED representatives themselves also went to numerous other countries. MINED interviewees in Edwards (2013b) mentioned that, over time, they were sent by the World Bank to such countries as Bolivia, Brazil, China, Colombia, Guatemala, Honduras, Mexico, Spain, and the United States. Not surprisingly, the World Bank played a key role in financing these study trips, which was not an unusual practice during the 1990s. Separately, these impact evaluations-together with the movement of World Bank staff-have influenced the design and adaptation of very similar models at least in Afghanistan, Burkina Faso, Guatemala, Honduras, Nepal, Niger, Senegal y Uganda, although it should be noted that there may be more cases of the direct influence of the reviewed studies, only that it is not possible to identify them because the individuals inspired or unknowingly influenced by the evidence on EDUCO do not always cite it publically as a point of reference (Najjumba, Habyarimana, & Bunjo, 2013;Sawada et al., 2014;Shibuya, 2013). To that end, an education specialist from the World Bank with over 20 years of experience in El Salvador and other countries explained how the lessons from EDUCO have become part of "international knowledge," such that development practitioners and governmental functionaries in other countries are unaware that they are following the example of El Salvador: Most development agencies now have learned about those models that happened, you know, more than 20 years ago, but, 20 years ago, actually, to go from something that was a natural process [in El Salvador, in the absence of governmental support] and then saying, ok, first we need to set up an organization to create a community body and let's give some formal role to parents and then they need to plan and create a school improvement plan and they need to have committees where students come together with teachers and parents and they execute those plans, and we need a mechanism to transfer resources from the central level to the school …, and then how do we provide accountability of those resources-all those things … 20 years later are well known processes, not only in the Bank but in the United Nation's Children's Fund, USAID, the Japan International Cooperation Agency-we know about those processes. Twenty years ago, they evolved from El Salvador, and they have become now part of international knowledge that some of our younger staff, the newer countries that are developing community participation programs don't know that that evolved from El Salvador. My staff worked with Afghanistan, you know, 21 These three ministers were: Abigail de Perez (Minister during 1998(Minister during -1999, formerly Vice-Minister and Director of Planning for the MINED), Evelyn Jacir de Lovo (Minister during 1999(Minister during -2003, formerly Director of the Modernization Unit within the MINED, later working for the Organization of American States as the Director of the Departamento de Desarrollo Social y Empleo), and Darlyn Meza (Minister during 2004, formerly Vice-Minister, Director of Education, and Coordinator of the EDUCO office). Only Rolando Marin, who briefly served served as Minister during 2004, did not have a previous association with the EDUCO reform. they don't know. The ministry of education didn't know that what they call community participation schools and all these operational processes, that, you know, in organizational management, in financial transfer, in planning, that it evolved from this tiny nation. (INTACT34) Fifth, by imbuing EDUCO with an aura of success, these studies have created a foundation of legitimacy that allows the World Bank to call on the former champions of the program for support. According to Edwards (2013b), one key functionary, who worked with the EDUCO program for many years early on mentioned that she would eventually "know all of Latin America … and a great part of the world because of EDUCO" -about 25 to 30 countries in all (p. 247). This functionary goes on to say, "people wanted to know how it was possible that illiterate rural people could manage, could hire, could fire teachers, how it was that the poor could be interested in school. That is was people had a hard time understanding" (Edwards, 2013b, pps. 247-248). More recently, in 2011, this same interviewee was sent by the World Bank to Afghanistan to discuss the EDUCO model. For the World Bank, such Salvadoran education specialists have played a key role in EDUCO's promotion. In the words of the above-mentioned interviewee, "I helped a lot with the community theme. When people [(i.e., governments)] were scared to give money to the parents, I helped to make the school boards and transfer the funds to the schools … in many places" (Edwards, 2013b, p. 248). This former Salvadoran education specialist served as reinforcement when officials in other countries were reluctant to experiment with the EDUCO model. Here again, actions such as being called on to defend and spread the program would not have been possible-or arguably would not have occurred to the same extent-without the track record of success outcomes that had been established through the studies discussed in this paper.
The sixth, final, and perhaps most direct international implication is one that still has relevance: the evidence base discussed here has become a body of literature that exists independently of the EDUCO program. Appendix 3 contains extensive examples. In terms of the World Bank, staff there continue to make use of the evidence base produced during the height of attention to EDUCO. In more recent publications, as they review and write on findings related to decentralization in practice, they draw on EDUCO studies, despite their limitations, and in so doing signal what is, from their perspective, the enduringly positive and strong example of EDUCO (see, e.g., Barrera-Osorio, Fasih, Patrinos, & Santibañez, 2009;Bruns, Filmer, & Patrinos, 2011;Di Gropello, 2006Umansky & Vegas, 2007). As with the World Bank, so too do other institutions continue to draw on the EDUCO studies reviewed here. Although this is evident from the list found in the introduction and in Appendix 3, included here are a few key examples (apart from the World Development Report 2004, discussed above) that demonstrate that institutions as diverse as the Brookings Center, the Global Partnership for Education, and UNESCO have recently found the lessons highlighted in EDUCO evaluations to be relevant.
 Brookings Institution, 2008: A report on improving development effectiveness spotlights EDUCO and the fact that the school council "hires, monitors, retains or dismisses teachers" (Hartman & Linn, 2008, p. 47 (Bruns, Filmer, & Patrinos, 2011).  World Bank, 2012: A typical example is found in a book written by World Bank specialists about education reforms for sub-Saharan Africa. They state: "the authority of communitymanaged EDUCO schools to hire and fire teachers in El Salvador had a beneficial impact on student outcomes compared with traditional schools serving similar populations" (Majgaard & Mingat, 2012, p. 149).  Save the Children, 2013: In a report on community participation, EDUCO is cited in relation to what mothers can do to enhance student outcomes (Save the Children, 2013).
It should be noted, though, that the above mentioned institutions have highlighted EDUCO for a variety of other reasons that do not have anything to do with the original goals of the program. In looking at the evidence on EDUCO, they have found-or have chosen to see-lessons related not only to instituting community involvement in education (Global Partnership for Education, 2009), but also to extending system coverage rapidly (UNESCO, n.d.), achieving educational and gender equality (Birdsall, Levine, & Ibrahim, 2005), providing education in post-conflict contexts (Beleli et al., 2007;UNESCO, 2011), and ensuring the provision of education as a human right (UNESCO, 2007). In these ways, one sees, on one hand, how the evidence base created by the World Bank continues to echo throughout the international development landscape, and, on the other hand, how various institutions have adapted the lessons of EDUCO to their own purposes. In other words, the evidence on EDUCO has become symbolic; these studies have a reputation for being rigorous, or are often otherwise perceived by their readers as such, and so development professionals invoke them as a way lend credibility to the points they wish to advance. Going forward, this will likely be one of the primary legacies of the EDUCO program internationally, especially as development trends change, and as community-level decentralization becomes less popular, replaced by such hot topics as public-private partnerships (Patrinos, Barrera-Osorio, & Guáqueta, 2009).

Conclusion: Limitations and Future Possibilities
The present manuscript began by examining the historical context within which EDUCO emerged in order to explain the dynamics that surrounded the program from the outset. This was the necessary starting point because the studies reviewed here were, in addition to being produced by and for institutions and individuals with vested interests, also borne into a very receptive politicaleconomic context, that is, a context in which notions of decentralization were already popular and in which the idea of community management of education was received with excitement because it represented the next step in experimentation around transferring responsibilities away from the central ministries of the State. Subsequent sections of the article then engaged in a detailed reevaluation of the widely-known and cited studies on EDUCO, followed by a consideration of what-in light of this re-evaluation-can reasonably be claimed to be known about the EDUCO program. Finally, a discussion was presented of the various implications of these EDUCO studies and the ways that they intersected with the political-economic context of the 1990s and 2000s.
In this review, it has not only been demonstrated that the evidence used to promote EDUCO is fundamentally flawed, but also that one cannot be sure of any of the outcomes claimed by the studies reviewed here. Moreover, as opposed to attributing the observed outcomes to the EDUCO program, it was showed that there is reason to believe that the outcomes could be the result of pre-existing social capital. Recall that the EDUCO program was initially targeted to communities which were affiliated with the FMLN and which, as such, had experience with various forms of self-organization, including the system of popular education schools that had developed during the civil war. Given these experiences, it may be that the EDUCO communities were well prepared to engage in the kind of community-level management that the EDUCO program entailed. Put differently, there is reason to believe that the EDUCO communities included in the studies reviewed here had relatively more social capital than the communities surrounding the traditional rural public schools. To the extent that this is the case, the control group is dissimilar and cannot be compared to treatment group of EDUCO communities for the purpose of determining the impact of the EDUCO program.
Returning to the studies reviewed in this study, they are generally problematic for two reasons that go beyond their specific findings. The first is the econometric nature of the studies, while the second is the political-financial-intellectual complex from which these studies were borne and back into which they went as they furthered the interests of that complex. On the first point, one must note that econometric methods, while useful for indicating the general nature of relationships or correlations among variables, cannot by their nature offer causal explanations. On one hand, this is because data do not speak for themselves, and so researchers and practitioners must interpret the relationships among variables based on their guiding frameworks, whether they be implicit or explicit (Kvernbekk, 2013). On the other hand, this is because econometric studies look at inputs and outputs without being able to unpack the "black box" of how a policy works (Pawson, 2006). That is to say, econometric methods are limited in their ability to explain how and why each sub-component of a program-such as EDUCO, for example-works or does not work in practice. These methods restrict their focus to what goes in and what comes out, thereby failing to account for both the inner-workings of a program as well as the many layers of context that impinge on how the program operates in implementation (Pawson & Tilly, 1997). In the words of Dumas and Anderson (2014), "Achieving high levels of rigor in quantitative studies involves limitations that include social and cultural decontextualization, ahistoricity, [and] the creation of randomization that is seldom generalizable to real life settings" (p. 5). A fundamental point is that, while quantitative data are considered ideal for empirically justifying the implementation of education policies (that ultimately have and always had political motivations), these data are not sufficient.
It is for the above reasons that in-depth, thorough, and analytic qualitative studies are necessary for answering policy-relevant questions and for providing policy-relevant findings. While Dumas and Anderson (2014) discuss a range of possibilities, a few approaches are highlighted here that hold promise for informing policy. Given space constraints, it is only possible to mention these approaches briefly; this is done with the hope that they may spark interest among researchers as useful ways to investigate complex educational issues. These approaches are vertical/horizontal case studies (Vavrus & Bartlett, 2006), process tracing (Bennett & Elman, 2006), realist case study (Pawson, 2006), systems thinking (Gillies 2010), ethnographic evaluation (Whitehead, 2002), participatory evaluation methods (Upshur, 1995;World Bank, 2011), and narrative research (sometimes also known as life history research) (Lieblich, Tuval-Mashiach, & Zilber, 1998;Moen, 2006). The common characteristic that these approaches share is that they that complement and expand the findings that econometric analyses can offer, for example, by illuminating the lived experience and emic perspectives of stakeholders (narrative research, participatory project assessment), by revealing which policy mechanisms work in which contexts (realist evaluation), by indicating how policies are implemented and affect the daily lives of people in practice (ethnographic evaluation, participatory action research), by providing insight into how the same policy can be interpreted and implemented differently across multiple levels and multiple locations (vertical/horizontal case study), by showing various sub-systems and various aspects of the education system impact one another (systems thinking), and by illustrating how multiple actors and multiple events influence one another in the process of policy making and policy implementation (process tracing). As can be seen, through these methods, researchers are frequently able to arrive at more nuanced and more explanatory findings than would be the case through the exclusive use of econometric methods.
Going back to the second issue mentioned above-i.e., the kind of political-financialintellectual complex that tends to produce impact evaluations and the consultants who carry them out-the work of Samoff (1993) is relevant. As a scholar who focuses on African education reform in the context of international development efforts, and often under external pressure, Samoff (1993) has witnessed conditions similar to those which engulfed El Salvador in the late 1980s and 1990s. He characterizes the dynamics of reform in such contexts in this way: Having concluded that local decision makers have fundamentally mis-managed their responsibilities, the external agencies offer general and rapidly disbursed support in exchange for broad control. This broader form of external assistance impels a demand for broader expert knowledge. The expertise required goes well beyond, say, curriculum development and teacher training. The relevant experts are those who can understand and manage production, finance, international exchanges, and national planning as well as social services. This call for broader expertise comes at a moment of severe economic crisis, precisely the time when African countries are least able to supply expertise and experts at the level and scale and with the credibility and legitimacy the external agencies demand. (Samoff, 1993, p. 186) This description is reminiscent of the context and sequence of events that President Cristiani experienced after being elected in 1989. Not only was El Salvador subject to external pressure from USAID and the World Bank to enact certain reforms for both the economic and social sectors, but Salvadoran entities such as FUSADES, the MINED and the Ministry of health also had to work with international consultants and researchers from UNESCO and the World Bank, among other foreign institutions, who possessed the necessary clout and technical capacity to credibly guide the reform and to carry out the research that would be used for decision-making around the reform of social services, such as education.
But Samoff (1993) makes an additional observation about the relationship between national and international actors that is key for our purposes. In his words: "what is most powerful and most insidious in this relationship is the internalization within Africa of worldviews, research approaches, and procedures for creating and validating knowledge that effectively perpetuate Africa's dependence and poverty" (Samoff, 1993, p. 187). The point here is that, over time, certain approaches to research-typically the econometric and advanced statistical approaches utilized by the consultants-are seen as the only or the most legitimate methods of investigation, particularly for the purpose of policymaking (Samoff, 1996). The implication in practice is that local researchers strive to replicate technically complex methods, in order to for their research to achieve the same legitimacy, though often researchers in middle and low-income countries are hindered by a lack of training in these methods and by a lack of funding to carry out costly studies. One consequence of this situation is that local researchers continue to rely on international consultants and international organizations for both financial and technical resources, in addition to relying on them for their reputation and status as capable knowledge producers. A second consequence is that qualitative methods, such as those mentioned above, are seen as less desirable and less appropriate when it comes to their ability to inform education sector reform, despite the fact that quantitative methods suffer from numerous shortcomings themselves (Erickson & Gutierrez, 2002;Klees & Edwards, 2014;Leamer, 1983;Samoff, 1991Samoff, , 1996Quade, 1970;Schroeder, Sjoquist, & Stephan, 1986;Verschuren, 2001), though they are often unrecognized or purposefully overlooked, as well as the fact that the relevance of qualitative research to policymaking has been thoroughly defended (Dumas & Anderson, 2014).
Both of the above-noted consequences have certainly played out-and continue to play out-in El Salvador. Going forward, it will be necessary to work in a number of different areas to address this situation, especially to the extent that the goal is to ensure that there are well-trained Salvadoran researchers who can work within and outside the Salvadoran government to produce policy-relevant research from numerous theoretical perspectives and through various research methods. Based on this case from El Salvador, it is also suggested that an additional goal should be to incorporate and to prioritize research that is conducted through the lens of social justice. A few broad, interrelated strategies to work towards these ends include: first, reducing the power imbalance that has allowed external actors and external studies to hold such influence; second, seeking alliances with domestic and international organizations who themselves prioritize principles of social justice in their research studies and in their policy recommendations; third, developing research centers and the research capacity necessary to counter mainstream (often neoliberal) studies and rhetoric around policy; fourth, connecting these centers and local researchers with international think tanks, international scholars, transnational social movements, and networks of activists in order to provide alternative and/or novel perspectives on currently relevant issues. These strategies would also help to counter, where necessary, the policy messages that are being promoted around the world and in El Salvador by dominant (often neoliberal) knowledge institutions. The importance of these suggestions cannot be understated, as making progress in these areas would drastically transform how education policies are made (and how, or whether, they go global), particularly in low-and middle-income countries, and particularly given the entrenched nature of neoliberal frameworks used for understanding and for reshaping the provision of social services. Importantly, as has been highlighted in this paper with regard to the World Bank, these frameworks underlie the methods used and shape the findings and recommendations offered, with the implication being that there is an ideological agenda within the World Bank that is not only protected from research findings but which itself informs the findings produced through ostensibly objective research. In the worst case scenario, World Bank researchers use empirical data to justify those policies that fit with the organization's preferences (Broad, 2006;Verger, 2012). At the global level, one indicator of the timeliness of the suggestion to increase the research capacity of southern knowledge agents is the fact that it has been proposed by actors involved in the process of setting the post-2015 goals (Quint & Lucas, 2014).
In the end, it is hoped that the present paper will help to take a step forward to address the issues highlighted in this concluding section. As has ben argued, when it comes to research on education policy, new methods and new political-institutional-financial arrangements are needed, a point which the present review of the impact evaluations of EDUCO has made clear. -Not the case in more fully specified model-Moreover, in models where variables for (a) community participation and (b) the presence of an EDUCO school are significant, the latter is negatively related to teacher salary. In other words, the presence of an EDUCO school corresponded with lower teacher salaries -Community participation (i.e., ACE member school visits) enhances teacher effort (i.e., meetings) with parents -Simultaneity; not possible to separate effects or directionality of influence between variables. That is, both variables (participation and teacher effort) are inter-related and, as such, cannot indicate a concrete and independent relationship with the dependent variable -Community participation in EDUCO associated with lower studentteacher ratio -Spurious relationship; EDUCO schools in sample have fewer students.

Appendix 1 Summary of the Findings and Limitations of the World Bank's Evaluative Studies of the EDUCO Program
-EDUCO participation positively and significantly effects student achievement -EDUCO resulta logros académicos mejores (en términos estadísticos) -The participation variable with an effect is not specific to EDUCO, but represents participation in by parents in councils of traditional and/or EDUCO schoolsand, as such, we do not know if the observed effect is due to the EDUCO program -76 models estimated, variables inconsistently significant -EDUCO leads to increased language scores -Additional controls were available which were not used in the models presented with statistically significant variables; when applied, language scores were no longer significant Key: EDUCO = Education with Community Participation Program; OLS = ordinary least squares; PSM = propensity score matching Notes: *Data from 1993 come from a sample of 78 schools. This sample include 33 sections of preschool and 45 sections of grades one and two. In each section for grades and and two, 17 students took the language and math tests. Basic information was also collected about the family to use as controls. Data from 1996 is based on random sample of 311 schools with a third grade section of at least 10 students. The surveys administered included language and math tests, in addition to a household survey and questionnaires for the director, thirdgrade teacher, parents, and two ACE members. Data from 1998 added new student learning tests and teacher and parent surveys to the information collected in 1996. The comparison group for all quantitative studies was traditional rural public schools.