Social Accountability: What Does the Evidence Really Say?

— Empirical evidence of tangible impacts of social accountability initiatives is mixed. This meta-analysis reinterprets evaluations through a new lens: the distinction between tactical and strategic approaches to the promotion of citizen voice to contribute to improved public sector performance. Field experiments study bounded, tactical interventions based on optimistic assumptions about the power of information alone, both to motivate collective action and to influence the state. Enabling environments for collective action combined with bolstered state capacity to respond to citizen voice are more promising. Sandwich strategies can help ‘voice’ and ‘teeth’ to become mutually empowering, through state–society synergy. 2015 Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).


INTRODUCTION 1
Social accountability strategies try to improve institutional performance by bolstering both citizen engagement and the public responsiveness of states and corporations. In practice, the concept includes a wide range of institutional innovations that both encourage and project voice. Insofar as social accountability builds citizen power vis-à-vis the state, it is a political process -yet it is distinct from political accountability of elected officials, where citizen voice is usually delegated to representatives in between elections. This distinction makes social accountability an especially relevant approach for societies in which representative government is weak, unresponsive, or non-existent. 2 Social accountability (SAcc) is an evolving umbrella category that includes: citizen monitoring and oversight of public and/or private sector performance, user-centered public information access/dissemination systems, public complaint and grievance redress mechanisms, as well as citizen participation in actual resource allocation decision-making, such as participatory budgeting. Yet amidst this diverse array of ongoing institutional experimentation (at both small and large scale), analysts are recognizing the differences between limited tools for civil society monitoring and voice on the one hand, and broader public interest advocacy reform initiatives on the other (e.g., Joshi & Houtzager, 2012).
Social accountability initiatives are multiplying in the broader global context of the booming transparency and accountability field, which also includes high-profile open government reforms and a proliferation of voluntary multi-stakeholder initiatives that attempt to set social and environmental standards, mainly for the private sector. 3 These diverse efforts are based on the assumption that 'information is power' -that transparency will necessarily leverage accountability. Yet widely accepted, normatively appealing theories of change, summed up as "sunshine is the best disinfectant," turn out to have uneven empirical foundations (Fox, 2007a). In response, both practitioners and policy analysts are increasingly posing the "what works" question -and the answer remains inconclusive. 4 Practice in the SAcc field continues to race ahead of empirical research, and relevant theory lags even further behind.
The diverse mix of institutional change initiatives that fall under the rubric of social accountability complicates efforts to draw broader lessons. Those who seek answers in terms of one-size-fits-all, easily replicable tools quickly confront the empirical reality that social accountability processes and outcomes are very context-dependent (Grandvoinnet, Aslam, & Raha, 2015;O'Meally, 2013). Calling for an evidence-based approach is not enough. Rethinking the growing body of evidence can advance the way we understand SAcc, which can help to inform realistic strategies.
This study reinterprets both the empirical evaluation evidence and the analytical concepts involved in SAcc, in order to help to address the "what next?" question. First, the paper identifies limits to the conceptual frameworks usually applied to SAcc. Second, a meta-analysis assesses the SAcc impact evaluation literature through new conceptual lenses. This exercise draws primarily on 25 quantitative evaluations, with an emphasis on field experiments that are widely considered to be influential in the field, based on their uptake by mainstream practitioners. Third, the study proposes a series of grounded conceptual propositions to analyze the dynamics of SAcc strategies, informed by the "state-society synergy" approach to institutional analysis (Evans, 1996). The article concludes with an emphasis on pro-accountability coalitions that bridge the state-society divide.
To preview the main argument, if one unpacks the impact evaluation evidence, it actually tests two very different approaches under the broad SAcc umbrella: tactical and strategic. Tactical SAcc approaches are bounded interventions (also known as tools) and they are limited to "society-side" efforts to project voice. Their theory of change assumes that access to information alone will motivate localized collective action, which will in turn generate sufficient power to influence public sector performance. Strategic SAcc approaches, in contrast, deploy multiple tactics, encourage enabling environments for collective action for accountability, and coordinate citizen voice initiatives with reforms that bolster public sector responsiveness. Reinterpreting evaluation evidence through this new lens, it turns out that the results of tactical approaches are indeed mixed, whereas the evidence of impacts of strategic approaches is much more promising. This interpretation points to the relevance of institutional change strategies that promote both "voice" and "teeth" (defined here as the state's institutional capacity to respond to citizen voice). The concluding proposition for discussion is that 'sandwich strategies' of mutually empowering coalitions of pro-accountability actors in both state and society can trigger the virtuous circles of mutual empowerment that are needed to break out of "lowaccountability traps."

RETHINKING CONCEPTUAL FRAMEWORKS FOR SACC
The SAcc field has outgrown conventional conceptual frameworks, and lessons learned from practice should inform new approaches. This section reviews the limitations of four widely accepted conceptual frameworks. All four were imported from other intellectual agendas, rather than developed with the goal of understanding social accountability.
The World Bank's 2004 World Development Report on public service delivery set a global agenda, framing service delivery performance problems in terms of accountability gaps and pathways (2003). Conceptually, the report emphasized the principal-agent framework (P-A) as the most relevant tool for understanding the relationship between citizen voice and public sector response. The P-A approach became conventional wisdom in mainstream development thinking, assuming that citizens are ultimately the principals -regardless of whether or not they actually live under electorally competitive regimes (e.g., Griffin, Ferranti, & Tolmie, 2010). Yet when the P-A framework is applied to governance, it implicitly assumes what it needs to demonstrate -that citizens are indeed ultimately in charge -the "principals." Moreover, this approach often makes the assumption that citizens-as-principals have relatively homogenous interests and goals. The main issue here is one of "conceptual stretching" (Sartori, 1970). The P-A model originally referred to two-way market relationships, such as shareholders-managers, managers-employees, or customers-service providers. When applied to politics, it originally focused on clear-cut, formal relationships of delegated authority. Social scientists then stretched the metaphor, applying it to more amorphous power relations involving mere influence rather than authoritative power, as well as multiple "principals." This diluted its parsimony. The model also has difficulty with non-hierarchical oversight relationships, as in the cases of mutual accountability inherent in partnerships, checks and balances institutions and informal accountability relationships -all of which are especially relevant for social accountability processes.
The 2004 WDR built on the P-A approach to propose another very influential metaphor for understanding different sets of power relations between citizens and public service providers. The "long route" has citizens exercising their "principal-ness" by delegating authority to political representatives, who then govern bureaucracies by choosing policymakers who in turn form compacts to manage front-line service providers. The "short route," in contrast, links citizens directly to service providers, through various oversight and voice mechanisms (as well as exit options, if available). The long-short route metaphor did not address the potential contributions of other public "checks and balances" institutions, such as legislatures, the judicial system, audit institutions, ombudsman agencies, or public information access reforms. In addition, the 2004 WDR's proposed short-route approach to addressing frontline service providers is also exclusively local, reflecting an assumption that institutional failures are primarily local, rather than distributed all the way up the governance "supply chain." A decade later, mixed results suggest that the "short route" may not be so short after all. Indeed, influential World Bank researchers recently concluded that there is no "short route" when the problem is what they call "government failure" -akin to market failure (Devarajan, Khemani, & Walton, 2014). They recognize that there is no way around the central issue of political accountability and the incentive structures that influence the degree to which elected officials are responsive to citizens.
By the latter part of the decade of the 2000s, official World Bank documents began to promote a third discursive frame for accountability issues, deploying the market metaphors that contrast the "supply" and "demand" for good governance. This reflected the World Bank's own internal organizational divisions, which separated staff dealing with inward-looking public sector reforms (the supply side) from those who promoted public interfaces and civil society engagement (the demand side). In contrast to the 2004 WDR, this approach does emphasize the potential contribution of checks and balances-type institutions, which fit under the "supply side" (anti-corruption bureaus, open budgeting, legislative oversight capacity-building, grievance redress mechanisms, etc.). Yet the market metaphor suggests that somehow demand will create its own supply, or vice versa. The implicit assumption that an invisible hand would bring them together is unrealistic.
A fourth conceptual framework for understanding accountability draws on spatial metaphors. Horizontal accountability refers to the mutual oversight embedded in the state's institutions of checks and balances -relatively co-equal relationships that do not fit easily into principal-agent models (O'Donnell, 1998). Vertical accountability refers to political accountability relations between citizens and their elected representatives (Mainwaring & Welna, 2003). This is a crucial concept for understanding where pro-accountability reformers come from, as well as whether their power base can help them to pursue institutional change. Diagonal accountability refers to hybrid combinations of vertical and horizontal oversight, involving direct citizen engagement within state institutions (e.g., Ackerman, 2004;Goetz & Jenkins, 2001;Isunza Vera, 2006;Paul, 1992). This can involve either participation in or direct management of official oversight bodies. Some of these official state-society power-sharing bodies are created from above, as in the case of "invited spaces" (Cornwall & Schattan Coelho, 2007). They can become remarkably participatory, like Brazil's national policy conferences (Pogrebinschi & Samuels, 2014). Other power-sharing institutions are created in response to broad-based citizen protest and advocacy, as in the case of the early days of Mexico's independent election administration (Avritzer, 2002;Isunza Vera & Olvera, 2006).
In the context of these spatial metaphors, social accountability efforts can be either vertical or diagonal. They are vertical when citizens make demands on the state directly, whether inside or outside of electoral channels (Peruzzotti & Smulovitz, 2006). These vertical and diagonal dimensions interact with each other, since the space for citizen power within official oversight bodies may be created in response to vertical pressures from below. Conversely, some argue that where horizontal accountability is weak, the underlying cause is flaws in the vertical accountability process (Moreno, Crisp, & Shugart, 2003). Where weak horizontal and vertical accountability systems reinforce each other, one can speak of "low accountability traps" (Fox, 2007b). Analysis of these accountability bottlenecks involves unpacking the state in terms of its often spatially uneven degree of institutionalization and efficacy (O'Donnell, 1993). Moreover, under some conditions, elected national authorities may have incentives to allow undemocratic subnational regimes to persist -or they may lack the clout to break them (Fox, 2007b;Gibson, 2005;Giraudy, 2013). Yet when social accountability efforts do have impact on the state, it is often through a pathway that involves triggering or empowering horizontal public oversight institutions to act (Fox, 2007a;Peruzzotti & Smulovitz, 2006).
Each of these four broad conceptual frameworks has their own strengths and limitations, yet they do not direct us to the kind of analytical tools that are needed to advance our capacity to categorize, measure, and compare the dynamics of the many different approaches that fall under the umbrella category of social accountability. A fresh set of conceptual propositions is needed, drawn inductively from actual reform experience, including the distinction between tactical and strategic approaches, the relationship between voice and teeth, forward vs. backward-looking accountabilities, "squeezing the balloon," vertical integration, and the sandwich strategy. First, however, a meta-analysis of the impact evaluation evidence is in order.

REREADING THE SACC EVALUATION EVIDENCE:
WHAT DO MIXED RESULTS MEAN?
How does one draw broader lessons from a body of empirical evidence that covers very diverse reform efforts, in a wide range of contexts? The SAcc impact evaluation evidence combines apples and oranges. As a result, it should not be surprising to find that "the evidence is mixed" -but this does raise the question of how to interpret the findings. Do specific cases of lack of SAcc impact "disprove" the broader concept? Do specific cases of positive impact "prove" the broader concept? This raises the broader question: what would "proof of concept" for SAcc look like?
The notion of "proof of concept" is very relevant for first addressing the "what works?" question -and then for reframing it. Widely used in scientific, medical, and engineering fields, "proof of concept (or principle)" refers to the demonstration that a proposed idea functions as predicted. 5 Moreover, the process of testing the possible validity of an idea is distinct from assessing its generalizability. In other words, there is a difference between demonstrating whether a proposition works at all, and showing that such a proposition holds across a wide range of conditions. In the case of SAcc, the general proposition would be that informed citizen engagement can improve the public sector's performance, especially if it bolsters the functioning of public oversight institutions. Yet tests of this general proposition under specific conditions would only provide hard evidence of whether that particular version of the idea works under those specific conditions. 6 For taking stock of the evidence related to SAcc, one of the most relevant observations from the experience with "proof of concept" in the biomedical field is that the pathway for translating a promising idea into practical, applied solutions is often long and difficult. For example, the "theory of change" behind vaccines originated in 1796. Now -centuries later -no one doubts the validity of that theory of change, yet vaccines still only work for certain diseases, to some degree, with problem-specific substances and doses that require very extensive experimentation to discover. The point of this analogy for assessing institutional change strategies is that even potentially "high impact" solutions to problems are likely to have only partial impacts, only under certain conditions, only for certain problems.
The "proof of concept" idea suggests reframing a common SAcc question: "does it work?" The problem with this formulation is that it implicitly sets up the answer in dichotomous, yes-or-no terms. 7 It is more relevant to frame questions in terms of the degree to which -and the conditions under which -an institutional change initiative would work. In addition, the criteria for assessing whether a change initiative "works" may well be contested. Especially in contexts in which the baseline is a complete absence of public accountability, even partial and uneven increases in accountability may be quite significant. For example, in the case of Mexico's regional Community Food Councils, at most one third of them managed to play their autonomous role of overseeing the performance of a large-scale rural food distribution program. Therefore, the program "failed" two-thirds of the time. Yet for those millions of low-income rural citizens whose interests were represented by the more autonomous councils, the program certainly did work (Fox, 1992(Fox, , 2007b. Moreover, the "does it work?" framing of the question also implies that a robust general answer can be drawn from what is still a relatively small impact evaluation literature that focuses more on measureable information-led pilot interventions than on large-scale, sustained national citizen voice or oversight programs. Perhaps most importantly, the "does it work?" framing of the question also implies that SAcc is expected to work all by itself, in the absence of other good governance reforms.
The issue of proof of concept for SAcc is on the development policy agenda for very good reason -because a series of influential studies have documented cases that have led to little or no tangible development impact. Development practitioners are drawing at least three general "takeaways" from these evaluations: First, information is not enough. Specifically, impact evaluations have tested the proposition that local dissemination of service delivery outcome data will activate collective action, which will in turn improve service provider responsiveness. The studies that find no impact from information dissemination interventions include Banerjee, Banerji, Duflo, Glennerster, and Khemani (2010), Lieberman, Posner, and Tsai (2014), Keefer and Khemani (2012) and Ravallion, van de Walle, Dutta, and Murgai (2013), among others.
The second general proposition is that bottom-up monitoring often lacks bite. Here, an especially influential impact evaluation has tested the proposition that local oversight of public works, by itself, can limit corruption. Olken's field experiment involving community road-building in Indonesia's KDP program found that community monitoring had little impact on reducing corruption (2007).
Third, a growing body of research finds that official "community-driven development" programs are often captured by local elites (e.g., Mansuri & Rao, 2013;Platteau & Gaspart, 2003). This literature focuses on state-led ("induced") participation rather than on SAcc per se, but both approaches overlap to a degree, insofar as they share the goal of encouraging the under-represented to exercise voice in the use of public resources. These three propositions are quite compelling, so what do they mean for understanding SAcc? The interpretation of the empirical evidence of SAcc impacts is complicated by the fact that some of the most influential studies of SAcc nonimpact do not actually show what many think they show. It is instructive to examine three especially iconic studies in terms of the differences between how their findings are widely understood and what they actually show. The choice of these studies and the interpretation of how their findings are widely understood were based on 15 interviews with World Bank staff and consultants, carried out between December 2013 and March 348 WORLD DEVELOPMENT 2014. The interviews asked which evaluations they considered to be the most influential, both among their colleagues and for their own thinking about the strengths and limitations of SAcc approaches. First, consider Olken's methodologically elegant comparison of anti-corruption interventions in village public works in Indonesia (2007). Practitioners often interpret this study as supporting the more general claim that top-down central audits work, whereas community monitoring has little impact on corruption (though the same author later showed that because local authorities were skilled at hiding their corruption, the community-based monitoring lacked adequate project oversight capacity [Olken, 2009]). To "work" in this case meant a reduction of one third of the estimated leakage (down 8 percentage points from 24%). Yet the causal mechanism behind the audits rarely involved official penalties. It was mainly the threat of community responses to the promised local dissemination of the findings that gave the audits the clout to reduce corruption. 8 Moreover, all of the communities involved in the field experiment were already mobilized through their involvement in KDP, a national participatory rural development program. 9 As an architect of KDP put it, while the study itself clearly stressed the community read-outs of the audit findings, "for some reason the evaluation community at large didn't want to hear that part." 10 This study was subsequently quite influential in policy terms, leading the Indonesian government to scale up its application of central audits to more than 80% of the local development projects in 70,000 villages. Yet the project's official monitoring data do not indicate whether the community dissemination of those audit findings was also scaled up. 11 In spite of the imbalanced uptake of its findings, this study shows that top-down and bottom-up approaches were synergistic rather than dichotomous.
Second, consider Banerjee et al.'s influential field experiment focusing on village education committees in Uttar Pradesh (2010). Researchers collaborated with a prominent Indian education civil society organization to test approaches to the provision of information about schooling outcomes to parents and village education committees, in an effort to activate them to attempt to improve school performance. The CSO convened parent meetings that generated attendance, but no learning outcomes. Their findings show that "providing information on the status of education and the institutions of participation alone is not sufficient to encourage beneficiary involvement in public schools" (2010, p. 5). The study also documents the weaknesses of the official channels for community participation and oversight. In that state, the Village Education Committee is composed of the head teacher, the elected head of village government, and three parents chosen by local officials. They are therefore -by design -not independent oversight bodies (cf, Barr & Zeitlin, 2011). Moreover, the study found that a quarter of the parent members specifically denied being members, the vast majority of members knew little about the VEC, and 92% of villagers were unaware of the VEC's existence. Nevertheless, the study recognized that "citizens are unlikely to participate in collective action unless there is a concrete course of action available" (2010, p. 4). In this case, that action involved training parents to teach literacy outside of the public schools. This was the most intensive of the study's interventions, and offered a viable option to a minority of families -showing that the main constraint was not lack of parent interest in their children's education. However, "none of the interventions increased parents' involvement with the public school system" (2010, p. 21). This suggests that neither the existing channels for parent participation in schools, nor the intervention's attempt to activate them, managed to create an effective enabling environment for independent community oversight of the public schools. Indeed, the kind of information emphasized in the intervention focused on child learning outcomes rather than on teacher or school performance, limiting its actionability. Yet the study's title, "Pitfalls of Participatory Programs," implied that the official school oversight process was indeed participatory, in spite of the lack of parental collective action to hold the schools accountable. On balance, these results could be considered a "false negative" for the "does it work?" question, insofar as the "it" (participation in oversight) didn't actually happen.
Third, Mansuri and Rao's tour-de-force meta-analysis of almost 500 studies examines both community-driven development and local decentralization to address "the impact of large-scale, policy-driven efforts to induce participation" (2013, p. 2). Many within the World Bank concluded that the study found that participatory local development often does not work -that it is often captured by elites or leads to modest development impacts that are often socially biased. Yet the study explicitly limited its focus to top-down "local development" projects, and did not address bottom-up, "organic" participation. Moreover, many large-scale official development programs that ostensibly attempt to induce participation or that fund local authorities do not include substantive measures to promote accountability, either from above or below. 12 Local capacity to respond to potential openings from above may be limited -"civil society failure," as they put it. It should therefore not surprise analysts of participation that such interventions would be vulnerable to elite capture. The authors conclude that context-sensitive participatory efforts that are combined with the use of central authority to improve state responsiveness -and therefore accountability -are indeed quite promising. The study's conclusions are therefore very consistent with an emphasis on social accountability.
This exercise of re-reading three iconic impact evaluations helps to inform a more nuanced approach to different kinds of social accountability efforts. Many of the SAcc interventions that have produced meager results are based on key assumptions that turn out to be weak, such as "information is power, "decentralization brings the government closer to the people," and "community participation is democratic" and "community voice can (by itself) influence public service providers." Field evidence indicates that these propositions need to be further specified: First, what kind of information can empower the poor? Information needs to be perceived as actionable. 13 For citizens to be able to act on this information, an enabling environment needs to reduce fear of reprisals. 14 Incentives for information-led action increase with the likelihood that the state will actually respond to voice.
Second, what kind of decentralization can bring the government closer to the people?
Only those local governments that are pushed to be more democratic are likely to become more responsive when bolstered by the increased funding and authority that comes with decentralization. 15 Third, what kind of community participation is likely to represent the socially excluded?
Enabling environments to actively encourage the voice and representation of those who would normally be excluded because of gender, ethnic, or class bias are necessary. Fourth, what kind of community oversight can address state failure?
Local voices that challenge un-accountable authorities, by themselves, are likely to be either ignored or squelched. Under what conditions can voice change the balance of power? Citizen action that has the backing of allies within the state who are both willing and able to get involved, or that has forged links with other citizen counterparts to build countervailing power, has a much greater chance of addressing impunity. In brief, exclusively localized, information-led "demandside" interventions -what can be called tactical approachestend to be based on unrealistic assumptions. In contrast, what can be called strategic approaches to SAcc combine information access with enabling environments for collective action that can scale up and coordinate with reforms of the state that encourage actual public sector responsiveness to voice.
The relevance of this distinction between tactical and strategic approaches to SAcc becomes clearer when one turns to the body of evidence that finds tangible positive development impacts. Table 1 synthesizes the findings from a wide range of countries and sectors. In terms of issue areas, this evidence of tangible development impacts clusters in the areas of education, participatory budgeting, and water management, in countries with at least nominally responsive elected governments. This snapshot of the evidence does not claim to be complete, and it is limited to quantitative studies, with an emphasis on field experiments. While space does not permit detailed analysis of this body of evidence, broader patterns of interaction between citizen voice and state response do emerge.
(a) Unpacking studies with evidence of SAcc impact Drawing on Table 1, it is useful to illustrate several cases of how social accountability processes can lead to tangible development impacts by spelling out their respective causal chains (Joshi, 2014). Note that the degree to which each case is "fully" strategic varies, and together they certainly do not constitute "proof" of specific generalizations that would hold up across diverse contexts. But the combination of the breadth and depth of this evidence supports the hypothesis that strategic approaches are more promising than tactical approaches for leading to tangible development impacts.

(i) Uganda education spending information campaign
Perhaps the single most influential study that demonstrates tangible positive impacts of "information for accountability" interventions is Reinikka and Svensson's analysis of the public dissemination of school-level funding information in Uganda (2004). Public spending tracking surveys had shown systematic, high rates of leakage, undermining efforts to invest more in education. An information campaign then tried to increase parental awareness of block grants for schools. The statistical analysis demonstrated a clear correlation between a school's distance to newspaper distribution and the fraction of school block grants that reached the school, sharply reducing the share of funds diverted. This experience was influential in informing the 2004 World Development Report's "short route" to more accountable service provision. Yet two key elements were not spelled out in the causal chain. First, the study assumed the role of participation rather than documenting or explaining it. 16 Second, subsequent analysis added the contextual "supply side" dimension to the explanation for the reduced share of funding diverted. The government was simultaneously prioritizing sharp increases in school enrollments and spending -which also got parents' attention (Hubbard, 2007, p. 3 (ii) Participatory municipal budgeting in Brazil A second example of the causal pathway through which SAcc can promote tangible development impacts is based on two decades of large-scale, nation-wide institutional practice (rather than on a field experiment). Numerous municipalities in Brazil have been practicing participatory budgeting (PB) for extended periods, beginning more than two decades ago (169 of 5,561 as of 2000, with 27% of the population). 17 Independently, two nationwide studies compared social indicators in Brazilian municipalities with and without this elaborate process of direct citizen input into municipal resource allocation decision-making (Gonçalves, 2014, Touchton & Wampler, 2014. Municipalities with PB allocated a larger share of funding to sanitation and health services, reducing infant mortality rates (holding per capita budgets constant). While Brazil's PB processes vary widely in practice, on balance their positive impacts are clear. The studies find that PB encourages authorities to provide services that meet needs of otherwise underrepresented citizens, and the deliberative process also creates frequent citizen checks on promised actions by municipal governments. This research also underscores the long time horizon and iterative pathways involved in reaching tangible development impacts.
(iii) Uganda community-based health clinic monitoring plus deliberative local compact Bjö rkman and Svensson's very influential field experiment in Uganda worked with civil society organizations to promote local compacts between communities and health workers in dozens of Ugandan villages. After extensive piloting, they tested a community monitoring process designed to encourage voice, to avoid elite capture and to facilitate periodic dialog with health workers ("interface meetings"). The impacts in treatment communities were dramatic, including reduction in infant mortality (33%), increased use of outpatient services (20%) and overall improvement of health treatment practices (immunization rates, waiting time, absenteeism). This was made possible by voice, expressed through inclusionary community discussion and assessment of service performance, bolstered by interlocutors who facilitated direct negotiation of expected actions with the service providers, informed by making public the contrast between health worker and community perceptions of performance. Social rewards and sanctions appeared to be key incentives, though the study did not address how they shaped health provider responsiveness. Years after this first study, a follow-up comparison of efforts to encourage beneficiary control with and without access to information about staff behavior, confirmed that such information was indeed crucial to enable stakeholder action to improve services (Bjö rkman-Nyqvist, Walque, & Svensson, 2014). This case indicates that not all ostensibly voice-led report cards are the same. For example, in contrast to the education intervention in Banerjee et al. (2010) discussed above, this experiment involved a primary focus on service provider performance, as well as both explicit, negotiated "community contracts" that specified how services were to be improved and apparently meaningful elected community representation in the subsequent oversight process. Indeed, "more than one third of the [previously ineffective] local oversight committees were dissolved and new members elected following the intervention" (Bjö rkman & Svensson, 2009, p. 747).
(iv) India's right to information law, applied to social programs In spite of the widespread optimism regarding the spread of public information access laws, few studies document how they can bolster access to public services. Peisakhin and Pinto (2010) tested India's Right to Information Act with a field experiment that compared different strategies for low-income citizens to apply for food ration cards. Bureaucrats ignored most applicants, but those who also filed official information requests about both the status of their application and district-level processing times were consistently successful. Only bribery produced comparable results. To understand the causal mechanism would have required a different method, however. With institutional ethnography, researchers could enter the black box of frontline agencies to analyze the determinants of the behavior of public sector workers (e.g., Lipsky, 1980;Long, 1984). In this case, the study hypothesizes that mid-level bureaucrats fear that non-compliance with the information access law may slow their professional advancement. India's RTI law is also unusual in that non-compliant administrators are potentially subject to nominal fines.
(v) Community-driven development and village public works in Indonesia First known as KDP, then PNPM, this nation-wide rural community development program followed a strategy that created enabling environments for community-level participatory budgeting and oversight, mainly for local public works and later for health and education programs. The program led to increased consumption and access to health care in poor households and reduced poverty in all the sub-districts where it operated, especially in the poorest and most remote communities -though members of marginalized groups did not benefit as much as others (PNPM, 2012). The program involved relatively low levels of corruption, especially compared to other state programs, and the causal factors include local transparency, informed participation, local trainers, central audits and extensive monitoring and evaluation (Friedman, 2013;Guggenheim, 2006). Levels of community participation were high, including women, though spillovers to improved access to information and governance involving other programs were low (PNPM, 2012).

(vi) Social audit hearings in India
The incorporation of community public oversight hearings into India's national rural right-to-employment law is one of the most significant examples of a grassroots SAcc initiative that influenced national policy. Because of India's federal system, states exercise a high degree of autonomy in their interpretation and implementation of this law. In the state of Rajasthan, for example -the home of the grassroots social audit -local politicians blocked state government efforts to implement the law (Pande, 2014). Andhra Pradesh became the only state that committed itself to institutionalize the social audit strategy, bypassing local government and politicians, using a relatively disciplined bureaucracy to create the enabling local environment needed to have widespread, repeated public hearings to oversee the rural employment program (Aiyar & Mehta, 2015;Maiorano, 2014). This process led to improved performance of the rural employment program, compared to states where the social audit process was captured or not implemented (Shankar, 2010).

DISENTANGLING TACTICAL AND STRATEGIC APPROACHES
To sum up, this exercise of reinterpreting the empirical evidence of both strong and weak SAcc impacts indicates that the wide range of change efforts that are pursued under the SAcc umbrella are not all pursuing the same theory of change. Instead, this rethinking process suggests reframing the basic propositions that inform SAcc into two quite distinct categories: tactical and strategic (see Table 2). These two terms warrant explicit definitions. At the most general level, strategies link coordinated actions to goals, with a macro view of the overall process, whereas tactics refer to specific micro-level actions. "Strategic" is defined in this context as an approach with a theory of change that takes into account the relationship between pro-change actions and eventual goals by specifying the multiple links in the causal chain. A "tactical" approach is limited to a specific link in the causal chain. 18 The argument here is that a tactical approach to SAcc, which emphasizes local-level dissemination of information on service delivery outcomes and resource allocation to under-represented stakeholders -an exclusively "demandside" intervention -is based on two unrealistic propositions. The first assumption is that people who have been denied voice and lack power will necessarily perceive vocal participation as having more benefits than costs (if the costs are recognized at all). The second assumption is that even if locally bounded voices do call for accountability, their collective action will have sufficient clout to influence public sector performancein the absence of external allies with both perceived and actual leverage.
Strategic approaches to SAcc, in contrast, focus on disseminating information that is clearly perceived by users as actionable, in coordination with measures that actively enable collective action, influence service provider incentives and/or share power over resource allocation. This proposition also suggests that SAcc strategies that manage to scale up voice and collective action beyond the local arena, while bolstering the capacity of the state to respond to voice (i.e., teeth) are more promising.
This tactical-strategic distinction has major implications for how one assesses evidence. Localized, voice-only, tactical interventions test extremely weak versions of SAcc. In treatment and control terms, this could be considered 'underdosage.' To recall the earlier analogy, if a small dose, or an insufficient number of doses of a vaccine fails to prevent a disease, that does not rule out the possibility that larger or more numerous doses could be more effective. Critical mass is needed, and this may require significant lead time (as in Brazil's diverse participatory budgeting experiences). The pathway to impact may also be quite discontinuous, perhaps following a J-curve (Woolcock, 2013). Moreover, information-led, voice-only approaches tend to focus on the symptoms rather than the underlying causes of state failure (e.g., systemic teacher or nurse absenteeism). Insofar as the prospects for SAcc strategies to transform state-society interfaces depend on bolstering the state's capacity to respond to voice, voice needs to find synergy with other governance reform strategies -such as bolstering the autonomy and capacity of oversight agencies, as well as the access to the rule of law more generally. 19

SOCIAL ACCOUNTABILITY PROPOSITIONS FOR DISCUSSION
The tactical/strategic distinction is not the only analytical proposition that can help to shed light on both the opportunities and obstacles for bolstering SAcc impact. So far, mainstream development thinking about how to close accountability gaps has taken a deductive approach, importing concepts like "principal-agent" theory that were not designed to address SAcc's checks and balances and mutual accountabilities. A more inductive approach to concept development may be more appropriate to inform future research into the causal dynamics that drive SAcc impacts. To provide context for the tactical/strategic distinction, here follow a series of additional analytical and conceptual propositions for debate and elaboration, initially developed inductively from both top-down and bottom-up SAcc efforts in Mexico over more than two decades (Fox, 2007b).

(a) Information needs to be user-centered to empower
The tactical approach to SAcc interventions tends to assume that external actors can predict what kinds of data are going to be most relevant to spark and guide collective action. It would be more useful to draw on the concept of "targeted transparency," which focuses specifically on accessible information that is perceived as useful and actionable by stakeholders, and can be integrated into their routines (Fung, Graham, & Weil, 2007). In this view, information disclosure informs action by changing actors' perceptions, mediated by a political economy analysis of the different interests involved. In other words, it is unrealistic to assume that information that is not linked to credible pathways to change will overcome well-known obstacles to collective action. In other words, targeted transparency helps to identify when information can redistribute power. This user-centered emphasis on actionable information contrasts sharply with widespread optimism that larger quantities of publicly accessible data will necessarily promote good governance.

(b) Voice needs representation as well as aggregation
The SAcc literature tends to refer to voice without defining it. In practice, voice can have many different modalities, ranging from weak to strong, from small to large-scale, from socially biased to more inclusionary. Some policymakers may consider local-level "beneficiary satisfaction" surveysthe aggregation of individual responses to questions determined from above-to "count" as citizen engagement. Public interest groups, in contrast, would tend to understand voice in more collective, scaled-up terms. Widespread experimentation with social media has made the "scaling up" of voice easier in often-inhospitable contexts. Yet while ICT-enabled voice can certainly play an agenda-setting role, crowd-sourced voices have limited capacity to negotiate with authority about what to do about those new agendas. If and when the political space created by voice makes it possible This raises the issue of how to address the challenge of what Mansuri and Rao call "civil society failure" (2013)social contexts with limited capacity for autonomous, pro-accountability collective action. Where traditions of scaled-up self-organization are weak, freedom of association is limited, or cultural and linguistic differences complicate the projection of voice, the role of interlocutors becomes central (Tembo, 2013). Interlocutors are facilitators of two-way communication, and their role is often crucial for bridging cultural and power gaps. In contrast to tactical approaches that assume that information will by itself motivate action among subordinated people, strategies that emphasize interlocutors recognize that for the voiceless to exercise voice effectively requires support -as well as cross-cultural translation and bridge-building. The proposition here is that in a SAcc context, voice is most usefully understood as involving both the aggregation and the representation of the views of otherwise-excluded citizens.
(c) Recognize that voice can be constrained by the "fear factor" Tactical, information-led interventions tend to be based on the implicit assumption that participation has more benefits than costs -and that the people who are expected to participate also perceive the benefits as being greater than the costs. These assumptions tend to ignore well-founded fears of reprisals. 22 External allies can reduce the risks inherent in challenging impunity from below, as well as help to identify actionable pathways through which collective action can leverage responses from power-holders. That is the substantive meaning behind the technocratic-sounding term "enabling environment." The fear factor is why anonymity can be crucial for enabling voice. To take a larger scale example -India's widely hailed right-to-know law -violent reprisals against information-requestors have been significant. The Indian media reports the assassination of at least 50 information-requestors, as well as threats and injuries to hundreds more (Pande, 2015). There are significant opportunities for synergy between SAcc and legal empowerment here, with the latter field's focus on alternative legal defense approaches, such as community paralegals, in contexts where the rule of law is weak (Gauri, 2013;Gauri & Brinks, 2008;Maru, 2010). Indeed, until the fear factor is addressed and grievance mechanisms have more capacity to provide meaningful redress, many SAcc initiatives will fall short of a rights-based approach.
(d) Unpack accountability goals in terms of reactive vs. preventative approaches One of the foundational questions in the emerging field of accountability studies involves the two core elements of the term's definition -answerability and sanctions (Schedler, 1999). Is "answerability" enough to "count" as accountability, or does the concept necessarily require the inclusion of the capacity to sanction as well (Fox, 2007a)? Looking forward, it would be useful for analysts to address how the relative weight given to sanctions may vary across different account-ability efforts. The strategies that prioritize responding to past problems are often distinct from approaches that emphasize preventing future abuses. This is the classic challenge that faces promoters of "transitional justice" around the world, as they attempt to build democratic institutions following authoritarian regimes. The political dynamics of the possible tradeoffs between forward and backward-looking accountability efforts are rarely explicitly addressed in the research literature on social accountability. Yet frontline accountability campaigners, operating in institutional contexts that combine high risk with little means of recourse or redress, are likely to be quite strategic about investing their limited political capital primarily in forward-looking, preventative approaches.
(e) "Teeth" for public accountability refers to the state's capacity to respond to citizen voice -a process that includes both negative sanctions and proactive reforms These two dimensions of "teeth" draw directly from the previous distinction between reactive and proactive approaches. This use of the metaphor is not as broad as state capacity in general, nor is it limited to capacity to sanction abuse or inefficiency. State capacity refers to a very broad range of activities, whereas this focus is specifically on capacity for public accountability and responsiveness. The use of the "teeth" metaphor here is therefore not quite the same as its intuitive understanding. For example, the notion of "teeth" is also associated with pressure from below, generated by protest. For the purposes of this discussion, protest can be framed as an especially vigorous form of voice. The idea of teeth is certainly intuitively associated with state capacity to apply negative sanctions (legal or administrative), as in the case of investigating, verifying and responding to citizen complaints and grievances. Indeed, the capacity to sanction is central to many definitions of accountability. The use of the metaphor here goes further, since the idea of capacity for accountability here goes beyond reactive, punitive approaches to include proactive responses to public accountability demands. State capacity for positive institutional responses can take a more preventative approach by addressing underlying causes of accountability problems, in contrast to an exclusive focus on symptoms (as in the case of sanctions). Such capacities would include changing public sector incentive structures to discourage abusive or wasteful behavior, deploying preventative measures to reduce opportunities for corruption or abuse -such as open government reforms and citizen oversight -or bolstering state capacity to follow citizen recommendations that emerge from participatory budgeting. The reason for including both positive incentives and negative sanctions in this definition of teeth is that they often need to be deployed together in order to have maximum impact (hence the term 'carrots and sticks'). To summarize, the proposition here is that for the purposes of analyzing SAcc, teeth refers to state capacity to respond to voice.

(f) Bring vertical accountability back in
The proposition here is that accountability strategies need to address mutually reinforcing linkages between non-accountable politicians and bureaucrats. According to the 2004 World Development Report's conceptual framework, the long and short routes to accountability are separate. Yet in practice, both public sector managers and frontline service providers are rarely insulated from electoral politics. 23 Indeed, in many contexts the politicized delivery of public services is widely used as a tool of electoral control (e.g., Fox, 2007bFox, , 2012. Moreover, less-than-democratic elections produce political leaders who are motivated to restrain the same public oversight agencies whose actions are crucial to give teeth to SAcc initiatives (e.g., anti-corruption agencies, grievance redress mechanisms). In addition, the combination of partisan manipulation of access to social programs with the politicization of horizontal oversight agencies can undermine fair elections, which leads to vicious circles of self-reproducing "low-accountability traps" at both national and subnational levels. This problem suggests the need to complement the vast "transitions to democracy" research of the 1980s and 1990s with new analytical frameworks that can account for the inherently uneven and contested processes of "transitions to accountability" within regimes that are widely considered to be at least formally democratic (Fox, 2007b).
Reviewing the SAcc evaluation evidence a decade after the 2004 WDR, the short route to accountability has turned out to be much more indirect than initially postulated, and its success may depend on making the long route more responsive as well. For example, Bangalore's famed Citizen Report Cards had their most significant impact on public sector performance only after a responsive Chief Minister was elected (Paul, 2006) -thereby finding synergy between voice and teeth. This issue underscores one of the missing links in the discussion of the "short route" to accountability -it often needs the "long route" of responsive elected authorities to work. The proposition here is to identify obstacles to SAcc by recognizing the interdependence of vertical, horizontal, and diagonal accountability relationships, since blockages in one arena can spill over to the other. Unless the mutually reinforcing linkages between non-accountable politicians and bureaucrats can be broken, however, their resistance to SAcc efforts will likely succeed. Conversely, not only is attention to vertical accountability crucial for addressing the obstacles to other kinds of accountability, citizens' capacity to exercise power over elected officials -whether through the ballot or protest -is also essential for explaining why policymakers might choose to invest their political capital in promoting pro-accountability reforms in the first place.
(g) Voice and teeth need each other Once the shorthand concepts of voice and teeth have been defined, the next step to address the dynamic processes through which they interact. How do these two different approaches to deploying power to shift accountability relationships reinforce each other, in order to break out of low accountability traps? This question suggests identifying causal pathways. Though they will always be context-dependent, one could argue that such pathways will vary in terms of whether they are either more voice-led or more teeth-ledas illustrated in Figure 1. (h) SAcc strategies need to address the 'squeezing the balloon' problem The targets of citizen oversight may well adapt by reconfiguring their corruption or diverting advocacy attention to other agencies or levels of the state. The corrupt are flexible, so their corruption can be fungible. For example, in some large-scale community oversight programs, like India's social audits or Indonesia's KDP, it appears that corrupt officials respond by inventing new and less visible ways to divert funds, shifting from wage theft to manipulation of materials billing Figure 1. Diverse pathways out of low accountability traps.

354
WORLD DEVELOPMENT (e.g., Olken, 2009;Shankar, 2010). Indeed, these two very large-scale national social accountability programs do much to project voice but little to bolster teeth, in the form of official willingness and capacity to sanction corrupt officials from above. Moreover, if citizen oversight efforts only address local, front-line service providers, this leaves out the rest of the "supply chain" of governance. Program monitoring that is partial or exclusively local in scope may well manage to change the shape of the "corruption market," but not necessarily its size (the amount of leakage -see Zimmerman, 2014).
(i) As a result, civil society policy monitoring and advocacy needs vertical integration The premise here is that corruption and social exclusion are produced by vertically integrated power structures. Insofar as multiple links in the chain of governance facilitate the deflection of civil society oversight and advocacy, effective responses require parallel processes that are also vertically integrated. 24 Vertical integration of local, regional, and national civil society oversight can begin to mitigate the "squeezing the balloon" problem. Yet there are often missing links between local community voice and national citizen policy/oversight. Given the often well-coordinated coalitions between anti-accountability actors across scale, vertically integrated civil society monitoring and advocacy is likely to be only as strong as the weakest link in its chain. Clearly, this is a tall order, though examples may be more numerous than the literature on development policy and service delivery suggests. In practice, CSOs around the world have extensive track records with vertically integrated/horizontally broad oversight and advocacy strategies -in the field of election monitoring, for example. Overall, institutional analysis of the density and dynamics of the local-national linkages that ground civil society advocacy campaigns in the global South has lagged. Indeed, after two decades of extensive research that highlights local-global civil society relationships, scholars are still in the relatively early stages of "bringing the national back in." 25 (j) Sandwich strategies can shift power with state-society synergy This proposition is grounded in the "state-society synergy" conceptual framework for understanding institutional change (Evans, 1996). 26 This process of mutual empowerment across the state-society divide is also called "co-production" or "cogovernance" (Ackerman, 2004;Ostrom, 1996). The specific theory of change here is that the construction of accountability is driven by coalitions of pro-accountability forces that bridge the state-society divide -acting to offset anti-accountability Figure 2. The sandwich strategy: opening from above meets mobilization from below. forces that are also often linked across the state-society divide. The term "sandwich strategy" is shorthand for these coordinated coalitions among pro-accountability actors embedded in both state and society (Fox, 1992). 27 The sandwich strategy's point of departure is that anti-accountability forces, deeply embedded in both state and society, are often stronger than pro-accountability forces. To break these "low-accountability traps," resistance is likely and therefore conflict would be both expected and necessary -as indicated in Figure 2. While initial opportunities for change are necessarily context-driven and can be opened either from society or from the state, the main determinant of a subsequent pro-accountability power shift is whether or not pro-change actors in one domain can empower the others -thereby triggering a virtuous circle (e.g., Fox, 1996). In this scenario of mutual empowerment -as illustrated in Figure 2 -reformists within the state need to have actual capacity to deliver to their societal counterparts, by providing tangible support and the political space necessary to provide some degree of protection from the likely reprisals from vested interests.
This process of openings from above led by reform champions that meet collective action from below represents one only of many possible strategic approaches to pro-accountability change, using the term in the sense described above. Many other kinds of accountability campaigns are led primarily by pressure from below, which may or may not find and empower counterparts within the state.
(k) Because context matters, the subnational comparative method is necessary to capture variation The comparative method has a great deal to offer the "what works, and why" research agenda, but it has been persistently crowded out by the dominant qualitative-quantitative debate. In the emerging field of accountability studies, few analysts recognize that the comparative method is a broad logic of inquiry within which quantitative and qualitative approaches are both sets of tools whose relevance and relative strengths depend on the question. Among those who do use the comparative method, nation-states are often the main unit of analysis. Yet the study of SAcc requires more nuanced approaches that can address their inherent uneven-ness within states and societies. In practice, any large-scale change initiative is likely to unfold in diverse ways across districts, provinces, and sectors. Empirically, the subnational comparative method can reveal patterns of variation that otherwise would be hidden by homogenizing national averages. Analytically, this approach allows researchers to hold many key contextual factors constant, which can help to highlight the impacts of specific institutional change strategies (Snyder, 2001).

CONCLUSIONS
This study reconsidered the empirical evidence through new conceptual lenses, in order to transcend the impasse associated with the notion of "mixed results" and to inform more strategic approaches to social accountability. The main elements of the argument include: First, the SAcc umbrella of diverse citizen-state engagement interfaces involves two qualitatively distinct sets of approaches: tactical and strategic. The impact evaluation evidence indicates that while the tactical approach has led to mixed results, strategic approaches are more promising. Tactical approaches are bounded, localized, and information-led -yet information alone often turns out to be insufficient. More innovation, experimentation and comparative analysis will help to determine what kinds of information are most actionable for pro-accountability stakeholders, as well as the channels for dissemination that can motivate collective action, empower allies and weaken vested interests.
Strategic approaches to SAcc, in contrast, bolster enabling environments for collective action, scale up citizen engagement beyond the local arena and attempt to bolster governmental capacity to respond to voice. So far, however, both SAcc advocates and skeptics have tended to assume that citizen voice, by itself, is supposed to be able to do the work of the state's own horizontal accountability institutions. Yet few voice-led initiatives are well-coordinated with relevant public sector reforms that encourage responsiveness (i.e., audit/ anti-corruption investigative bodies, information access reforms, grievance redress mechanisms, ombudsman agencies, access to courts, etc.). At the same time, ICT-led SAcc initiatives are increasingly framed in terms of "closing the feedback loop" -in other words, getting institutions to listen to citizen voice (Gigler & Baijur, 2014). Yet institutional response capacity often remains elusive; in practice feedback loops rarely close.
Second, now that the SAcc field has generated a substantial body of practice, this is a timely moment to take stock. Research lags significantly behind practice, and theoretical and conceptual work lags even further behind research. In response, this study concludes with a series of grounded conceptual propositions intended to inform higher impact SAcc strategies -with an emphasis on the potential synergy between "voice" and "teeth" -with the latter defined as governmental capacity to respond to voice.
Third, both practical and analytical work on SAcc needs to take scale into account. When voice spreads horizontally, the excluded can gain representation. When voice is projected vertically, it can gain clout. When authorities listen, they can both build trust and create incentives for more voice. Yet this process is easier said than done, and the dynamics that drive it will not be well-understood if mainstream development agencies continue to treat state failures as a strictly local, "end-ofthe-pipe" problem. 28 This underscores the relevance of combining vertical integration with the horizontal spread of civil society oversight and advocacy capacity. This combination of scaling up with "scaling across" can make possible the combination of voice with representation that is crucial for significantly changing the terms of engagement between excluded citizens and the state.
What does this emphasis on scale have to do with the vast and growing on-going array of SAcc initiatives around the world? At the local level, many small-scale pilots may be ready for more scaling up and horizontal expansion -though only fine-grained contextual analysis can determine which ones, when and how. Already-existing large-scale governmental SAcc reforms need more support from other governance reforms in order to both broaden and deepen the openings they have created -for example, in the cases of India's social audits in the National Rural Employment Guarantee Program and the community participation in Indonesia's PNPM rural development program. Both innovative reform initiatives potentially reach tens of millions of people, both build in openings for voice for the poorest -yet state capacity to respond remains limited, uneven and contested, and grassroots stakeholders have yet to scale up their monitoring capacity and gain a seat at the table.
To sum up, the challenge facing social accountability strategies is how to break low-accountability traps by triggering vir- 356 WORLD DEVELOPMENT tuous circles in which enabling environments embolden citizens to exercise voice, which in turn can trigger and empower reforms, which can then encourage more voice. That is, voice needs teeth to have bite -but teeth may not bite without voice.

NOTES
1. This is a revised and abridged version of a working paper released as Fox (2014 2. Some political theorists have also argued persuasively that electionsusually considered the public accountability mechanism par excellenceare actually quite blunt instruments for holding authority accountable, insofar as voters are often more prospective than retrospective in their behavior (Przeworski, Stokes, & Manin, 1999). See also Peruzzotti and Smulovitz (2006).
3. So far, social accountability efforts focused on the public and private sectors have evolved independently of each other, with development policymakers focused almost exclusively on the public sector. In the early 1990s, the multilateral development banks' own safeguard policies included pioneering social accountability processes in their environmental impact assessment and indigenous peoples policies, as well as the World Bank's Inspection Panel (Fox & Brown, 1998).
4. This was a key conclusion of an agenda-setting body of research commissioned by the Transparency and Accountability Initiative (a donor consortium) and completed by the Institute of Development Studies in 2010 (Carlitz, 2013;Gaventa & McGee, 2013;Joshi, 2013;Mejía Acosta, 2013 9. Scott Guggenheim (email communication, Jakarta, Feb. 3, 2014). He also notes that field experiment was bolstered by the shock effect of the first-ever promise of audits and their dissemination. This effect may have been temporary, insofar as local authorities subsequently learned that the audits were rarely followed by official sanctions. In addition, the auditors themselves -who previously rarely visited the villages -were especially diligent because the researcher directed the field experiment. As a result, "the government auditors for the first time got paid on time, they received their travel allowances, they were supervised in the field, and somebody checked their work." 10. Scott Guggenhiem (Jakarta, email communication, Jakarta, April 21, 2014).
12. The author pointed this out to World Bank Mexico country staff in 1996, noting that their support for decentralized rural municipal fundingthough positive in the state of Oaxaca -was simultaneously bolstering authoritarian local governments in states like Chiapas (Fox, 1997).
13. Kosack and Fung add an important distinction -between those information-led interventions that offer citizens data about public service performance, vs. those that involve indicators that specifically allow citizens to compare their services with others (2014). In addition, Loewenstein, Sunstein, and Golman bring the insights of behavioral economics to the analysis of the impacts of information disclosure (2014).
14. The term "enabling environment" is rarely defined with precision. For the purposes of this paper, the term refers to actions by external allies that have two characteristics. First, they reduce the actual and perceived risks and costs often inherent in collective action. Second, they bolster the actual and perceived efficacy of collective action by increasing the likelihood and/or degree of positive institutional response. Thanks to Tiago Peixoto for requesting a more precise definition of the term. For a study that operationalizes enabling environment by documenting the actual degree of application of World Bank safeguard policies in the context of 10 rural development projects, see Fox and Gershman (2000).
16. The one exception is a sentence in a later article, which reported that of the half of the schools that did not receive their grant after the information campaign, "47% complained or protested to some formal or informal authority that could transmit the complaints onwards or act on them" (Reinikka and Svensson, 2011, p. 959). Their causal model explicitly assumes that because information makes protest possible, it is the access rather than actual collective action that mattered: "since both SOCIAL ACCOUNTABILITY: WHAT DOES THE EVIDENCE REALLY SAY?
an actual protest and the threat of voice may have discouraged the local political elite from diverting resources intended for the schools, in equilibrium, there is no reason to believe that the incidence of voice and local diversion of funds should be correlated" (2011, p. 959). It is not clear whether this assumption was tested.
17. Participatory budgeting can be considered to be a social accountability strategy insofar as it includes processes to monitor and encourage governmental responses to citizen input.
18. Thanks to Anuradha Joshi for suggesting that these definitions be made more explicit, and to Jeff Unsicker for feedback.
19. This last point underscores two of the many limitations of the vaccine analogy mentioned above. First, vaccines need to be consistent, whereas SAcc efforts may vary. Second, when vaccines work, they work by themselves -in contrast to most SAcc initiatives, whose effectiveness may depend on complementary governance reforms.
20. As Rosie McGee put it: "Aggregation can just be about numbers and scale, and is technical. Representation implies mediation and framing, and is political" (email communication, June 18, 2014).
21. On the ebbs and flows of leadership accountability within scaled-up mass membership organizations, see Fox (2007b).
23. For example, if one wants to understand why teacher absenteeism rates may be very high in a given context, then a tactical approach to SAcc would focus on monitoring their behavior and then applying sanctions for non-compliance and/or incentives for positive performance. If one took a strategic approach, then one would also ask who decides which teachers to hire in the first place, based on what criteria -and one may find that political patronage at higher levels of the system plays a major role. When ghost nurses or teachers have political backing, their capacity to resist local parent or NGO oversight is likely to be very high. This is the implication of Banerjee, Glennerster, and Duflo (2008). On the clientelistic politicization of a common SAcc approach, community-managed schools, see also Altschuler (2013).
24. For the initial formulation of this proposition, developed in the context of independent CSO monitoring of World Bank projects, see Fox (2001). Specific examples of fully vertically integrated CSO oversight of public service delivery include the work of SEND-Ghana (Dogbe & Kwabena-Adade, 2012), Textbook Count in the Philippines (Guerzovich & Rosenzweig, 2013), and the work of the Slum/ Shackdwellers International (D'Cruz, Fadrigo Cadornigara, & Satterthwaite, 2014).
26. Additional relevant work on state-society synergy includes, among others: Ostrom (1996), Fox (1992Fox ( , 1996Fox ( , 2004Fox ( , 2007b, Tendler (1997), Goetz and Jenkins (2001), Heller (2001) 27. The sandwich strategy would describe two of the largest-scale SAcc strategies now under way in the global South, Indonesia's PNPM rural village development program (formerly known as KDP) and India's National Employment Guarantee Program social audits in the state of Andhra Pradesh (Aiyar & Mehta 2015;Friedman, 2013;Guggenheim, 2006;Maiorano, 2014). Mexico's Community Food Councils also pursued this strategy for almost three decades (Fox, 1992(Fox, , 1996(Fox, , 2007b, and their broad-based, vertically integrated social oversight constituencies survived for many years after the purge of their reform allies within the state. The councils' autonomy eventually eroded in the absence of allies within the state, reducing their scope to a few regional enclaves.
28. The environmental policy metaphor of "end-of-the-pipe" refers to the difference between "source reduction" -reducing the use and emissions of toxicsvs. trying to clean up the pollution once it has been generatedas in the case of putting scrubbers on smokestacks. This is analogous to the medical distinction between mitigating the symptoms vs. treating the disease. In SAcc terms, for example, this would point to the difference between monitoring service providers' absenteeism in order to sanction non-compliance, vs. transforming service providers' relationships with the community they are supposed to serve -as well as changing the hiring, training and firing process in order to hire motivated service providers in the first place.