Counting Torture: Towards the Translation of Robust, Useful, and Inclusive Human Rights Indicators

ABSTRACT The turn to quantified measures is part of an attempt to produce more objective and comprehensive data on human rights violations. However, the turn to numbers has also been criticised for forcing human rights into the limitations of statistical capacities. This paper examines the methodological issues involved in trying to make human rights violations count, highlighting the cyclical process of translating between the experiences of human rights violations, quantified forms of measurement and human rights norms. It draws on the particular experience of conducting household surveys on the prevalence of torture in Nairobi, Kathmandu, and Dhaka. The paper argues that torture and ill-treatment can be made to count in ways that is robust, useful, and inclusive by developing indicators that are embedded in locally specific practices and forms of participation. This means treating the process of counting as a matter of contextualisation rather than abstraction. Doing so can help produce new understandings of the implications of human rights violations.


Introduction
The international human rights movement has increasingly called for more numbers. If human rights work traditionally focused on qualitative evidence from individual cases, there has been a growing interest in quantitative data from larger population samples over the past 20 years. 1 It is argued that qualitative case-by-case documentation can be anecdotal and biased, and it is hoped that quantification allows for a more complete, objective, and rigorous picture of the state of human rights. 2 This broad turn to numbers is also part of an attempt to understand the wider trends within which individual human rights violations take place, as well as the processes that facilitate accountability and compliance.
The promotion of quantitative data raises questions about the ways in which human rights violations more broadly can be made to 'count'. 3 As Sally Merry has argued, the search for seemingly objective measures can also lead to a focus on things that are relatively easier to measure, distorting the human rights project towards the vagaries of statistical capacity. 4 The turn to numbers, it is argued, can therefore exclude whole areas of experience and large numbers of victims or perpetrators. In this process, there is an additional risk that political struggles over what human rights mean get subsumed under technical questions of what and how to count. 5 Critics have, therefore, argued that the use of numerical data turns an exercise of judgment about the extent of human rights violations into one of technical measurement. 6 Counting human rights is, thus, far from straightforward.
In this article, we examine some of the methodological issues involved in making human rights violations count. We do so by discussing our own experiences in conducting mixed methods surveys examining the prevalence of torture and ill-treatment in lowincome neighbourhoods in Kenya, Bangladesh, and Nepal. In this way, the focus of this paper is directed at our reflections on carrying out the surveys, rather than simply specific empirical findings of the surveys themselves. Our surveys focused explicitly on the urban poor, as we have argued elsewhere that human rights work often under-perceives the extent of torture amongst the residents of informal settlements. 7 Such populations live in the political margins and their experiences are often not made to countin both senses of the word. There are, however, challenges in carrying out surveys amongst such communities particularly because the concept of torture more generally resists stable and consistent measurement. 8 The aim of this paper, therefore, is to use the experience of conducting these surveys as a point of entry to discuss the methodological issues involved in counting incidents of torture, rather than to present a step-by-step analysis of the surveys and their results.
In conducting the surveys, we took the widespread criticisms of attempts to count human rights violations seriously, not least the ways in which they can systematically leave out many experiences, behaviours, meanings and people. However, at the same time, we also recognised the limits of qualitative approaches. A case-by-case approach, for example, can prioritise high profile incidents, leading to seemingly less spectacular, but more widespread practices receiving less attention. 9 The question here is not whether violations can be countedfor which there is no simple yes/no answerbut rather, how can numbers be produced that begin to capture the variety and complexity of the ways in which torture is inflicted and experienced on the ground. 10 This is not an issue of methodological purity, as all methods have their blind spots and limitations. Instead, it is about thinking through how torture can be measured in meaningful ways. This, we think, is one of the fundamental challenges for human rights work in general.
The paper highlights one of the key challenges in making human rights violations count in particular: finding ways to translate between the experiences of victims, social science forms of measurement, and human rights norms, in a way that is robust, useful, and inclusive. The most rigorous forms of measurement can be irrelevant for human rights purposes if they do not directly refer to human rights principles. 11 At the same time, human rights orientated research must pay attention to social science principles of robust measurement if it is to produce valid and reliableand therefore defensibledata about human rights violations. Furthermore, and perhaps most importantly, any knowledge about human rights violations must also reflect the ways that those violations are experienced by victims as well as the objectives and intentions of perpetrators or it risks losing its ethical and political authority. Finally, although the prohibition of torture is widely recognised as an absolute principle, the human rights jurisprudence also says that context needs to be taken into account when deciding whether a particular event is a violation. 12 Like all forms of translation, counting human rights violations involves complex decisions about what to include and what to exclude. The danger is that these essentially political judgements follow a path of least resistance, and in doing so systematically exclude whole areas of activity or categories of person, particularly amongst otherwise marginalised groups. The first key argument of this paper is that the development of context specific indicators is central to inclusive forms of translation. Given the contextually specific meanings associated with the concept of torture, the balance lies in producing indicators that respect the central concept being measured, but also take into account the specificity of the way the concept plays out in particular places. If human rights indicators are developed in a top down mannerstarting with human rights norms and simply translating outthe indicators will most often fail to capture the variety of ways in which torture is inflicted in practice, and exclude many victims and perpetrators. In reflecting on the potentials and limitations of our own surveys, we therefore arrive at the second key argument: that one way of developing more context specific indicators is by taking a more participatory approach to their design, implementation and analysis.
The rest of the paper is organised into six sections. In the first section, we outline some of the issues raised by use of indicators in the turn to numbers in human rights work, with a particular focus on measuring torture. In the second section, we set out the broad shape of the surveys we carried out in Nairobi, Kathmandu, and Dhaka. While we do introduce some of the conclusions from the studies, our main focus is on the methodological challenges, as per our main question in this article. In the third section, we discuss some of the methodological issues involved in translating between experiences of violence, quantifiable forms of measurement and human rights concerns. The fourth section moves onto a discussion of the ways in which 'concepts' and more significantly 'indicators' contribute to the process of translation, before examining the issue in relation to counting the involvement of 'public officials' in acts of violence in the fifth section. We end by asking, if counting human rights violations needs to be valid and useful: valid and useful according to whom? In doing so, we reflect on the challenges and potentials involved in more participatory approaches to count human rights violations that replace abstraction with constant contextualisation. As a result, the process of human rights translation between experiences of victims, forms of measurement, and human rights norms can itself produce new ways of understanding human rights work that must not be underestimated.

The Challenges of Counting Human Rights Violations
The United Nations' human rights monitoring system has been one of the central places where human rights relevant quantified data has been both requested and produced. 13 In particular, the UN hopes to use numerical indicators in order to monitor relative compliance of the states with international human rights obligations. The concern here is not with individual violations but with general patterns (over time and across states), processes and outcomes. In relation to torture, the Office of the High Commissioner for Human Rights (OHCHR) has suggested using data on the proportion of detained person in facilities inspected by an independent body, as well numbers on prevalence of death, and physical injuries amongst detained persons, reported cases of torture and numbers of victims who have received compensation, amongst other things. 14 These proxy indicatorswhere torture itself is not being countedcan be seen as a classic example of top-down global indicators, where human rights law is operationalised into numerical measuresand it is assumed that torture is caused and played out in general ways. 15 However, the use of such numerical indicators for human rights monitoring faces a number of challenges. At a basic level, states are often unable or unwilling to produce the specific statistical analysis requested by the UN system. This is especially the case in low-income countries, where the resource intensive management of statistical information can prove impossible. Furthermore, state agencies can be direct parties to human rights events, and therefore are an obvious sources of bias. Whilst such global human rights indicators might be useful to the UN system, they also raise important questions about validity, reliability and the exclusion of large numbers of violations.
Alongside human rights monitoring, quantified human rights data have been widely used in political science. 16 Particularly striking is the research that tackles questions around whether human rights treaties make a difference to human rights compliance. 17 However, questions have been asked about the datasets that are used in much of this analysis. Indicators of torture, for example, have sometimes been inferred directly from 13 Merry (n 1); Office of the High Commissioner for Human Rights (n 11); Rosga and Satterthwaite (n 6). 14 Office of the High Commissioner for Human Rights (n 11). Amnesty International or US State Department Human Rights reports. Doing so can largely ignore the limitations in the ways in which the data is collected, that is, they can fail to adequately take into account the political context within which these reports are produced. They may also fail to adequately take into account systematic forms of underreporting that disproportionately affect marginalised populations. 18 Those victims who never come to Amnesty's attention or are deemed irrelevant to the campaigns of the organisation, for example, will simply be left out of the picture. As we have argued elsewhere, human rights documentation techniques can systematically misrepresent many forms of human rights violations and therefore do not necessarily serve as a reliable source of data. 19 More nuanced approaches have tried to make statistical adjustments for patterns of underreporting or paid particular attention to potential bias when coding human rights reports. 20 Much of this work focuses on new statistical techniques, rather than directly considers ways in which more inclusive data might be generated. Furthermore, it often treats torture as a self-evident social and legal category, rather than a contextually specific and normatively thick category. And as such, important debates remain about the inclusivity and validity of the datasets used in much of this analysis. 21 As political scientists and the UN treaty bodies have turned to quantifiable indicators around human rights, NGOs have also found quantified human rights data profitable. There has, for example, been an increased use of epidemiological work in relation to refugees and victims of torture. 22 Surveys are designed to provide baseline studies for larger health-based interventions, rather than to understand specific cases. 23 There are some attempts to tie public health surveys to human rights norms. In the case of torture, this has meant trying to measure specific outcomes such as pain and well-being in relation to acts of self-reported torture. 24 In general, broad terms such as 'human rights abuses', 25 or 'grave violations' 26 are used, but there is seldom an attempt to define what counts as a specific human rights abuse, or to relate these definitions to human rights standards.  The limits of surveys are well recognised, not least by the NGOs and social scientists that carry them out, and these issues are particularly relevant to our work in Nairobi, Kathmandu, and Dhaka. Large-scale surveys can be very resource intensive, and can therefore be impractical. Rates of torture may be too low to be reliably measured by anything but very large and very expensive surveys. There are also problems in producing valid and reliable data on sensitive and contested issues. Survivors might, for example, be unwilling or unable to report their experiences to enumerators. Furthermore, there are important ethical problems involved in carrying out such surveys. Interviewing people about torture can expose them to considerable additional risk. Issues of stigma and fear may mean that people are reluctant to report their experiences to interviewers. Finally, if your primary concerns are human rights, rather than health, it can be very difficult to produce reliable data about perpetrators, which is a crucial part of international human rights definitions of torture. 27 Household epidemiological surveys can therefore significantly underreport or misrepresent experiences of torture. Such surveys can be useful though, in the sense that they seek to shape health interventions, but questions remain about their robustness and direct relevance in human rights terms.
In sum, the counting of human rights violations has raised a number of key questions around validity, reliability, usefulness, and inclusivity. It is in this context that the authors of this paper carried out the surveys on torture and ill-treatment in low-income settlements. The challenge was: to count torture in a way that included contextually specific meanings of torture and the experiences of as wide a range of survivors as possible, particularly those from marginalised groups; to do so in a valid and reliable way; and to make sure that these numbers were directly applicable to human rights norms.

Surveying Torture in Dhaka, Kathmandu, and Nairobi
The three surveys were carried out in the context of broader research into the documentation of torture in low-income neighbourhoods. 28 The working hypothesis behind the research was that current human rights documentation practices systematically underreported the experience of torture amongst the residents of informal settlements. Social and geographic distance, as well as a focus on places of detention, and the idea of the 'good victim' could all mean the extent of torture amongst the poor was systematically underperceived by human rights organisations. The surveys were, therefore, an attempt to begin to get a sense of the prevalence of torture amongst such populations. These three different surveys were conducted in collaboration with local research institutes and NGOs. In Bangladesh, we worked with the Department of Peace and Conflict Studies at the University of Dhaka; in Nepal, a research institute called Social Science Baha; and in Kenya, a human rights organisation called Independent Medical and Legal Unit (IMLU).
Broadly speaking, victimisation surveys tend to ask questions around perceived and experienced levels of violence, justice seeking behaviour, and health and financial consequences of violence. The problem with this approach is that it assumes in advance that we know what the relevant issues are, who the most likely victims might be, what types of 27 Deportee and others (n 22). 28 Jensen and others (n 7). violence they might experience and who the perpetrators are. These assumptions are often found wrong. In order to mitigate these concerns, our survey questions were generated in collaboration with the research team from Edinburgh, Dignity, local research partners, and human rights organisations.
In Nairobi, initial qualitative research had indicated a close relationship between fears of violent crime and state violence within slums. 29 In Dhaka, the literature review had indicated a relationship between service delivery, political parties, gangs, and violence. 30 Finally, the literature review and scoping with local partners in Kathmandu suggested that state violence in slum areas was often linked to the threat of evictions among squatter settlements along the city's rivers. 31 These differences were one of the reasons why we did not demand that the same methods be implemented across the surveysto do so would have failed to take into account the specific local context within which violence was inflicted. We were mindful that this might impact our abilities to compare levels of torture across the three sites, but agreed that context and local buy-in had to take precedence. The principal aim was to explore methodologies, not to make inferential claim about torture and its correlates beyond individual local settings.
There were local differences in both sampling techniques and the organisation of the research in the three sites. These were partly the result of differences in institutional research capacity in the three research sites, as the team in Dhaka, especially, had longer experience in carrying out high-quality quantitative data collection. Dignity and Edinburgh were more active especially in Kenya, where the collaborating partner was an NGO. In Nairobi, a consultant was hired to carry out the interviews developed by IMLU and the research team. Five professional enumerators were employed who were teamed up with members of a local paralegal organisation to help interpret results and manage security.
Differences in the survey teams' make up were also the result of the political conditions of access in the three research sites. In Kathmandu, the local research team sought the assistance of the local squatters' associations for field assistance. This was seen as beneficial to the survey, but was also a demand from the squatters' associations to allow access to a highly politicised field. In Bangladesh, a team of researchers employed at the University of Dhaka carried out the research in an equally politically tense environment. Here, the reputation of Dhaka University helped gain access.
In total, the three survey teams interviewed around 2000 residents of urban informal settlements (usable sample sizes were 500 in Nairobi, 600 in Kathmandu, and 998 in Dhaka). After having administered the surveys, we carried out qualitative research in the three sites, following up specific cases that had been identified by the survey.
The bald numbers generated by the surveys reveal the wide prevalence of torture and ill-treatment. In Kathmandu, 5% of respondents reported experiencing torture and ill- treatment at the hands of state officials. The survey also showed that 18.2% of respondents felt that they or their family were at risk of being subjected to torture or ill-treatment. The police were perceived as the main source of risk by nearly 70% of the residents of the settlements alongside the river, which were under threat of eviction. Interestingly, just 14.7% of respondents said that they would report incidents of torture and ill-treatment to human rights organisations. Of the 21 incidents of torture and ill-treatment identified in the survey, 11 were reported to take place in their own homes, 7 in the 'community', and the remainder in a variety of police premises. 7 of these incidents were linked to 'investigations' and 10 to 'forceful eviction'.
In Dhaka, we identified 79 victims of violence over the previous 12 months from 998 respondents. Neighbours accounted for 19% of the reported perpetrators of these incidents, local 'goons' (gangsters) and 'extortionists' accounted for 17%, political activists for 10%, family members for 7%, and the police for 7%. Unidentified assailants carried out the remainder of incidents. Almost all the reported incidents took place inside the slum, a third within the victims' own homes, and the remainder on the street or in the market place.
In Nairobi, 41% of respondents reported that they or a household member had been the victim of a violent incident over the previous 12 months. Of these incidents, police officers were the perpetrators in 26% of cases. The police were perceived as the 'main perpetrator of violence' by 19% of respondents, behind 'criminal gangs' at 61%. Police violence included shootings, beatings, threats, and extortion. The most common place of violence was the victim's own home. Only 36% of all incidents of violence were reported to the police and only two victims had reported their experiences to NGOs.
These numbers reveal a pattern of torture that largely falls outside human rights documentation techniques. As we have argued elsewhere, this is in part because human rights organisations have a relatively limited presence amongst poor populations. 32 Another reason is, as shown in the surveys, that people living in low-income neighbourhoods are highly unlikely to report their experience to human rights organisations, because they have not heard of them, they cannot reach them, they are too scared to report their experiences, or the forms of redress provided by human rights groups were not their priorities. As such, they show the dangers of relying solely on data produced by human rights organisations.
While these numbers show how common the experience of torture and ill-treatment is amongst residents of slums in all three cities, on their own, they are almost meaningless and need to be heavily contextualised. It will already be evident, for example, that different terms were used in all three surveys in categorising the violence experienced by victims because they were generated locally, and therefore, different indicators used to count torture and related forms of violence. In large measure, the differences in the categories used was the result of specific judgments that needed to be made in order to translate between experience of violence, numbers, and human rights concerns. 32 We discuss these issues in detail in Jensen and others (n 7).

Translating Human Rights
It requires a complex process of translation to move from the ways in which violence is experienced and inflicted on the streets and neighbourhoods of Nairobi, Dhaka, and Kathmandu, to discrete eventsthat is, distinct occurrence that can be counted and then used as part of human rights work. Further, it is a process of translation that involves moving through specific domains of knowledge. In this section, we shall raise some challenges of such human rights translations, before suggesting that these translations can be productive of new insights of their own.
Translation is a process of creating equivalence to allow the transfer of meaning across epistemic domains. 33 Core principles have to be defined in order to decide what exactly needs to be translated. 34 When counting torture, for example, specific acts of violence need to be made commensurate in order to be translated into numbers and human rights concerns. As Merry and Wood argue such acts of commensuration ' … requires simplification and decontextualisation in order to render concepts such as … [torture] … countable'. 35 The risk is always that translation so distorts an object being translated that it loses meaning. 36 A universalising human rights understanding of torture, for example, can have a complex and indirect relationship with the ways in which violence is actually experienced in specific places.
Translating from the experience of violence inevitably strips acts of violence away from the complex relationships through which they take shape. Translation also involves normative judgments about what is important to translate and in which ways, including judging what type of perpetrators and victims should be prioritised and what types of violence should be emphasised. In this process, there is the danger that not only particular types of victims and perpetrator will be left out, but also that experiences and behaviours with specific local significance are ignored. These processes are never neutral, and they take place in a field of unequal power, where some actors and frames are made to count more than others.
What about the more specific translation between quantified forms of measurement and human rights norms? At first glance, there seems to be a close affinity between human rights work and quantified forms of measurement. Both, for example, privilege binary structures. Just as quantified research has to decide whether something is or is not an example of a concept being investigated, human rights practice decides whether an incident is or is not a violation of human rights legal frameworks. Furthermore, both can work according to primarily deductive forms of reasoning, in that they seek to start by applying already established categories or principles to empirical events. 37 But there are also important differences between human rights work and quantified forms of measurement. Binary forms of deductive decision-making are only one part of the human rights process. Human rights work also involves the inductive investigation of abuses. This part of human rights work involves the interpretation of clues and shards of evidence to make claims about otherwise unobserved events. Human rights practitioners, for example, are not there when torture is inflicted, but make claims about its existence from the clues left behind, such as the scars left on survivor's bodies. 38 Such work sees inductive reasoning to the fore. 39 Translating human rights violations therefore requires combining inductive judgments and inferences with the careful deductive consideration of principles. In this process, as Fariss and Dancy have argued, translation errors are often ' … subtle and difficult to detect'. 40 This is especially the case in a context where the meanings of torture can change in time and place. Howeverand this is a crucial pointalthough these translations require complex judgements, they are not necessarily bound to fail. Given that all translation is a process of communication, the very fact of translation assumes that we can speak across boundaries. 41 Translation can also enlarge original meanings and increase the audience who are willing and able to listen. In the world of human rights, turning violations into numbers and then translating those numbers into human rights concernsand back againcan expand the reach of human rights campaigns, enable new arguments to be made, and address new publics. 42 The challenge therefore is to translate across domains in a way that expands rather than restricts meaning, and does so in a way that does not exclude otherwise marginalised actors.

Concepts and Indicators
Translation is spread throughout the entire survey procedure, from question design, to sampling methods, enumerator training, codification, and onwards. However, we focus on one narrow aspect of the process in the shape of 'concepts' and 'indicators', as it is here that the meanings and shape of the object to be translated are primarily formed. 43 A concept describes the thing we want to measure. An indicator is the sign that we will use to mark a concrete instance of the concept we want to measure. Developing appropriate and dynamic concepts and indicators allows movement between human rights norms and empirical incidents in a robust, useful and inclusive way.
The development of concepts is, above all, a process of making distinctions between what is and what is not being measured. However, violence does not exist in discreet, self-evident blocks. Instead, violence is produced and experienced through overlapping and multi-layered political, economic, and cultural processes. For instance, an act of sexual violence, while apparently objective and evident, can become rapeor not - depending on a complex system of interpretations laced with power relations. 44 We, therefore, have to make distinctions within the continuum of overlapping events. 45 At one level, it would appear that international human rights already have exactly the tools necessary for developing the concepts needed to make torture count, and which can apply as much in Dhaka, Nairobi, and Kathmandu, as anywhere else. There is a long history of human rights jurisprudence on the definition of torture, which has aimed precisely at developing universal concepts about what counts as torture universally. The UN Convention Against Torture (CAT), defines torture as: … any act by which severe pain or suffering, whether physical or mental, is intentionally inflicted on a person … when such pain or suffering is inflicted by or at the instigation of or with the consent or acquiescence of a public official or other person acting in an official capacity … 46 As such, levels of suffering, forms of intention, and the identity of the perpetrator are all central to the way in which the human rights concept of torture distinguishes between different forms of violence. Human rights norms appear, therefore, to provide us with the appropriate concepts for measuring torture.
The next difficult step is to produce indicators through which events can be measured. This means deciding what does torture look like when you see it in Nairobi, Dhaka, and Kathmandu? Crucially, an indicator of a concept that is valid in one context might not be valid in another. Torture does not necessarily look the same in all places; it is not carried out in the same ways, by the same type of people, in the same type of places across the world. Torture in Bangladesh, for example, does not necessarily manifest itself in the same ways as torture in Nepal or Kenya. We must, therefore, develop indicators that respect the central concept, and take into account local specificity of the ways in which torture is practised. The point, though, is not just that torture is carried out differently on a physical levelthat is, the torture practice of beating feet may be favoured in one place, and electric shocks in another. 47 Rather, it is also that any act of violence can only be described as an incident of torture once it has been given meaning in ways that are both locally significant and resonate with the globally circulating human rights norms.
In moving from the normative concept of torture to developing specific indicators, we need to make at least four key decisions: (1) what counts as severe pain or suffering; (2) what counts as evidence of intentional infliction with a specific purpose; (3) who counts as a public official or other person acting in an official capacity; and (4) how these three elements relate to one another. This process raises a number of challenges. To begin with, it is hard to measure the perpetrator's intention. Thus, proxy indictors have to be designed to infer whether a perpetrator intended to intimidate a victim to confess. As numerous studies have shown, the ways in which we read off possible intentions has to take into account culturally embedded forms of communication. 48 Measuring levels of suffering is also notoriously complicated, due to the complex relationship between the experience of pain and its external indicators. Different people can show and talk about their suffering in very different ways, making it extremely difficult to produce valid and reliable data. 49 Determining the level of involvement of public officials is equally complex, as we will discuss in detail below.
Above all, the development of indicators involves treading a fine line between being too wide and including incidents that are wider than the concept, and being too narrow and therefore risk missing observations. Thus, any decision about what counts as pain, intention, or a public official requires complex judgment and interpretive work that only makes sense within specific contexts. Attempts to count human rights violations, therefore, face specific challenges in moving between the concept of torture and its indicators. And it is here that one of the key tensions between human rights and quantifiable forms of measurement comes into play. In much social science research, if the indicator proves unreliable and produces invalid data, we can look for other indicators. There is also space to change concepts, if the empirical research does not align with findings out there in the world. This is much more difficult when it comes to human rights or other norm-bound research. The concepts and indicators we are working with are derived from international human rights jurisprudence, and although they are not set in stone, they are nonetheless much less amenable to redefinition by researchers alone.
The next section of the paper will, therefore, examine the challenges of keeping human rights norms in place whilst counting torture in a robust manner, without excluding large areas of experience of specific types of victim. We do so by zooming in on one part of the measurement process: developing indicators on the involvement of people acting in an official capacity.

Counting public officials
What counts as an indicator of an act 'inflicted by or at the instigation of or with the consent or acquiescence of a public official or other person acting in an official capacity'? 50 At a conceptual level, there is a large amount of jurisprudence discussing the precise meanings and implications of suffering, intention, and purpose in the UN CAT definition of torture and ill-treatment. 'Public capacity' has been relatively less contentious, although still debated. In practice, 'public officials' is translated into 'state actors'. There has also been a move in the human rights jurisprudence away from focusing on the specific acts of identifiable public officials, towards a more normative focus on acts for which the state can ' … legitimately be held responsible'. 51 Acts carried out by public officials in a 'private capacity' though are usually seen to fall outside the definition. The UN Committee Against Torture has further indicated that acts carried out by 'non-government' entities without consent or acquiescence of government generally fall outside the definition of torture. However, in 'weak states', where state authority is wholly lacking, non-state actors can be seen as 'acting in an official capacity' and their action can therefore be seen to count as torture. 52 Immediately, we can see that there are important questions as to when and where a state becomes weak, as well as what counts as acting in a private capacity. For our purposes, at least one key point is important to note: what or who counts as a public officialat least in contexts with some form of functioning stateis treated as selfevident by the human rights jurisprudence. Yet, as a vast amount of social science has also shown, the structure and meanings of public authority are historically and contextually specific. 53 Public authority can be embedded in very different political, economic, and cultural processes that create different forms of legitimacy, different boundaries between the private and the public, and different relationships with the deployment of violence amongst other things. What counts as a public official can therefore never be taken for granted, either normatively or empirically. Public authority is not a yes/no binary distinction, but a spectrum, with ebbs and flows of intensity and meaning over both time and space. The production and interpretation of indicators around the involvement of public officials therefore need to take into account the slippery nature of public authority.
In our three case studies, public authority had very different structures and textures in each place. In Bangladesh, there is a dense relationship between the formal structures of the state, political parties, and criminal gangs. 54 In some cases, actors might be involved in all three sets of activities. In Kenya, public authority is caught up in wider political networks, linked to political parties, ethnic affiliations, and local government structures built on 'chiefly authority'. 55 The result is an often blurred line between an action done in public and private capacities. In Nepal, political parties, development agencies, and ethnicallybased networks play out across a terrain where the state itself is often unable or unwilling to provide basic resources. 56 In all three countries, the poor are often forced to negotiate across multiple levels and with different actors in order to gain access to resources and other ostensibly public goods such as housing, water, electricity, and security.
The complex structure of public authority in all three sites can be seen in the relationships that our survey teams needed to negotiate in order to carry out the research. Rather than a straightforward 'formal permission' from a government department, the research would not have been possible without the implicit or explicit acceptance of local squatter organisations, political parties, chiefs, or criminal gangs. In Kenya, the chief, whose office was outside the informal settlement could say yes or no to us carrying out the survey, but could not control the actual conduct of the interviews, which was supported by a local paralegal organisation. In Nepal, the squatter associations and various communitybased formal and informal organisations were an essential part of the survey process, and the work could not have been carried out without their permission. In Bangladesh, the cultural prestige and political capital of Dhaka University based researchers were enough to secure us access to the slum.
The implicit assumption of much human rights practice is that torture occurs in places of detention, with key public officials being prison officers, police officers, soldiers, and security officials. Empirically speaking, however, there are at least two problems with this approach. First of all, singling out police or prison officers does not capture all possible public officials. According to the human rights definition, any public official could be involvedfrom teachers to medical staff. To focus on police is to a priori decide who carries out torture, and therefore, problematically limit our findings. Second, what and who counts as the police is far from straightforward. In Kenya, alongside the Kenyan Police, private security firms, nyumba kumi (community based policing) and mungiki (ethnically based militia) have all sought to enforce particular visions of law and order. In Bangladesh, the Police, the semi-autonomous Rapid Action Battalions, mastaans (criminal gangs), and political parties have all attempted to impose their visions of order. In Nepal, the Police and municipal officials have stood alongside political parties (often Maoist), and private security firms, among others. In all three countries, the de facto 'legitimate' and 'illegitimate' deployment of force in the name of the public good has been dispersed across a range of actors, often loosely tied to the formal state.
Given the complex nature of public authority, the task of developing indicators that would enable us to capture their involvement in acts of torture against the poor becomes particularly complex. What, for example, are the signs that we are going to take as an example of a public official? Even if we take being a police officer as a key indicator of being a public official, how do we or our respondents know when the perpetrator in question is a police officer? What are the indicators of being a police officer? One approach would be to say that a police officer is someone wearing a police uniform. But given the wide variety of potential police officers, not all police uniforms may be recognised. And sometimes police officers do not wear a uniform, or take it off deliberately to change roles. 57 Such practices raise further questions as to whether a perpetrator who has taken off their uniform is acting in an official capacity. We also need to decide whether to take the victim's judgment of whether someone is a police officer at face value, given that we cannot ask the perpetrator themselves. The victim may have all sorts of reasons for either not knowing or not telling whether this is the case.
In practical terms, creating and measuring indicators in a survey is an issue of question design. Asking straightforwardly whether a perpetrator is a public official risks creating invalid and unreliable data, as the respondent might not understand what a public official is, or might have very different understandings from other respondents. The approach that we took in Nairobi and Kathmandu was to provide a list of possible public officials that we developed through working groups and pilot studies. approach though raises questions about whether it is exhaustiveand the 'other' category becomes very important. Such a list can only be constructed on the basis of knowledge about the local context. Hence, we developed the list in collaboration with local research and NGO partners and piloted them before doing the actual survey. In Dhaka, a third approach was taken where the identity of perpetrators emerged only in the survey process itself. We did not start with a list, but asked respondents to describe the perpetrator in their own terms. We could then code this account based on knowledge gained from further qualitative follow up.
The judgment over whether a public official was involved in an act of violence was only made possible with in-depth qualitative work. A formal definition of public authority failed to capture the everyday life of many people living in poverty. It was only through qualitative follow-up that the ambiguity, slipperiness and complexity of public authority came into view. An initial analysis of the Bangladesh survey, for example, showed very little reporting of violence by public officials. Respondents reported violence from neighbours and criminal gangs, but not from the policewho seemed to be involved directly in only 2% of incidents of violence. 58 However, we later carried out qualitative follow-up through ethnographic interviews with self-reported victims, and public officials came into view. This was partly because we did not have a clear enough picture in advance of what public authority looked like in the slum. It was only in the qualitative follow-up that the complexity of the relationship between police, political parties, gangs, landlords, and neighbours was revealed. The other reason for the relative absence of the police in the quantitative part of the survey is that people were reluctant to talk about police violence in the relatively initial short survey interviews. It was only in the longer qualitative follow-up interviews that it was possible to build trust and create a space where people would talk about the issue.
The Kenyan survey provides an interesting comparison. In the quantitative survey, 41% of respondents reported incidents of violence, with the police involved as perpetrators in 26% of these. 59 It appeared as if police violence was everywhere, and that Kenya was effectively a police state. However, in the qualitative follow-up study, a slightly different picture began to emerge, where police violence was both seen as a threat and a form of care. In many cases, police officers were invited into the intimate lives of residents and were asked to help them with their sons, together with their neighbours and fellow police officers. In one case from the survey site in Nairobi, a police officer had been transferred. In some of our informants' view, this led to allowing corrupt police officers and criminal gangs the freedom to carry out violence and extortion. 60 While there are differences, the relationship with the police is not one in which the police and residents are worlds apart in all three sites. In many cases, they share intimate worlds for better and for worse.
Across the surveys, a further issue that emerged in the qualitative follow-up was that violence by public officials was not isolated singular events. Quantitative questions tended to produce a picture of discrete events, as is the nature of victimisation surveys. Yet, the qualitative follow-up revealed the ways in which violent incidents were part of long histories of violence and embedded in dense and ongoing social relationships. In this process, the involvement of public officials could not be isolated as an event taking place in a snapshot of time, but rather was a process that flowed across time and space, changing in significance and intensity, and moving between public and private. In one case in Dhaka, an incident that began as a conflict between neighbours, developed over time to a conflict that involved the police, the local political parties, and neighbours, as well as the two antagonists.

Translating Numbers in Context
In translating between incidents of violence, quantifiable forms of measurement, and human rights norms in a valid, useful, and inclusive way, neither validity nor usefulness should be taken for granted. The most important questions to ask are valid and useful to whom? In this final section, we suggest that one possible way to facilitate this translation is to ask these questions head on. This means thinking carefully about who the research is carried out with, and, in particular, reflecting on the ways in which local stakeholders can participate in the research. Relative inclusivity can help produce better data at multiple levels. One of the key advantages of greater local stakeholder involvement is that it helps work towards making the translation between incidents, numbers, and norms locally meaningful. Indicators can be developed, rolled out and analysed in a contextually specific way, capturing the ways in which torture is inflicted on the ground.
The precise implications of participatory approaches and the extent to which they lead to meaningful ownership and transformation amongst victims and survivors is much debated. 61 There are important challenges in making research participatory, whilst also keeping it robust, inclusive, and useful in human rights terms. For example, local research capacities might be limited or local research interests might fit awkwardly with the goals and aspirations of the international human rights movement. Who counts as a 'stakeholder' and what counts as 'local participation' can be defined in different ways to include domestic human rights organisations and the communities where the survey was carried out. Sometimes, the interest of these groups can exist in tension. However, at a minimum, paying attention to the relative participatory dimensions of research helps address important questions about the development of local capacities to define human rights issues.
We do not necessarily want to hold up our surveys as models to be followed by others, as the surveys were not fully participatory in any of the field sites. For example, the original idea and focus for the surveys were set in line with the requirements of British academic funding, stipulating the need to work in certain low-income countries. In the three surveys, we also had different levels of local involvement at the points of design, implementation, and analysis. And in the three surveys, local involvement meant very different things. Nevertheless, we do think that the approaches we took provide a way of reflecting on the issues.
In Nepal, the research was carried out through a Nepal-based research organisation with a long and established track record of carrying out robust policy-relevant research. The initial focus on the squatter settlements came from working with Nepal-based human rights organisations who suggested that this group was likely to be victim of state violence due to attempts at resettlement, but that this was also where the biggest gaps in their knowledge were, precisely because of this human rights organisations had very little presence on the ground. As described above, we therefore had to negotiate access through various squatter organisations. These organisations were somewhat cynical about 'yet another survey', and implied to us that there was little they could learn, as they already lived it as a day-to-day experience. They were also worried about the ends to which the survey might be put, given the context of possible forced resettlement. They were much more interested in a full census. Rather than human rights concerns, they were interested in participating in the survey if we could employ their members as enumerators, and if they could use the maps and GPS system for their own work. One thing this work did was provide a careful warning against thinking of community buy-in as ever settled. Whilst we had negotiated access with what we thought were the relevant squatter organisations, partway along the research, other organisations came forward, claiming to represent some of the squatters.
In comparison, the research in Kenya was carried out in close collaboration with a local human rights NGO -IMLUwhose executive director has been first author on the subsequent report. 62 While the research was perhaps less methodologically innovative and had to be carried out with the help of consultants, it has been integrated into the work of IMLU and has travelled back through IMLU and their local partners to the research site. While the translation between human rights norms and empirical data was far from perfect, it has been much easier to translate the research results back into human rights practice. For instance, in October 2016, a feedback workshop was organised in the primary field site with the participation of local authorities and chiefs, police, CBOs and other NGOs along with victims of police violence. In June 2016, the research was presented by IMLU to the human rights and NGO community in Kenya, and in August 2016, IMLU participated in a briefing for the UN Committee Against Torture on protection of witnesses and victims of torture and ill-treatment where the results of the research were presented. The research was also used in a shadow report presented by IMLU to the UN Committee Against Torture, and, according to the executive director, in the strategic planning of IMLU.
In Bangladesh, the research was carried out by a university department with little direct formal relationship with human rights practice. This was partly because the world of human rights practice is seen, by some in Bangladesh, as compromised by political affiliation. Afterwards, it has proven relatively more difficult to translate the findings into something that can be used by human rights organisations in Bangladesh, although human rights practitioners are aware of the work and will hopefully use it as part of their own activities. This is the case, even if the research in Dhaka probably was the most robust in social science terms.
While the Kenya experience by no way represented a seamless translation, it is indicative of one possible way of mediating the possible tensions between robust, inclusive, and useful data. It involves engaging with domestic human rights activists and paralegals while producing the analysis rather than presenting results to them after the fact. IMLU had very little research experience, and it was often difficult to produce robust 62 Kiama and others (n 29). analysis according to social science criteria. However, the results did become more useful than would have been the case if we had not engaged with the human rights world along the way. Hence, in Kenya, collaboration provided the opportunity to make the research useful as well as robust. Bangladesh provides an example of the limitations of this strategy. Not working with the human rights world was not just a choice of convenience. The research might very well have been compromised from the outset had they been included.

Conclusion
How can we make torture count? And count in a way that is robust, useful, and inclusive? In this article, we have sought answers to these questions based on three surveys carried out in Nairobi, Dhaka, and Kathmandu. Rather than systematically introducing the results from the surveys, we have discussed the methodological challenges in counting torture and ill-treatment in ways that are robust, inclusive, and useful. Counting torture in robust, inclusive, and useful ways means making the difficult translation between experiences of violence, quantified forms of social science measurement, and human rights concerns. For all the difficulties involved in this process, producing numbers about torture can potentially help us begin to grasp the prevalence of torture in a way that is not possible through case-by-case analysis. Numbers alone though also risk excluding many experiences of torture. A formalistic approach to understanding the involvement of public officials, for example, can miss the often chimeric quality of the state in the lives of the poor.
Counting human rights violations can never be a simple process of abstraction. Any attempt to produce numbers on incidents of torture requires context-specific judgments about what to count, where, how, and why. Further, numbers never stand alone. By themselves, they can tell us very little about the prevalence and nature of torture. They only become meaningful once their production and interpretation is embedded in local contexts and can be translated back in terms of human rights norms. Through the careful combination of deductive and inductive techniques, quantitative and qualitative analysis, coupled with constant reflections, we can begin to produce indicators that reveal otherwise ignored violations, whilst creating robust data that can be used as part of human rights work.
Human rights indicators need to be embedded in the specific contexts within which human rights abuses are carried out if they are to begin to capture the variety of these violations. Embedding indicators in local specifics might imply, at first glance, that it is harder to compare across different places. 63 But it is also important to note that human rights practice does not require that violations take exactly the same form in all places and all times. Just because torture in Bangladesh takes a different shape than it does in Kenya, it does not mean we cannot talk about torture in both places. Human rights, as with counting, always require judgements about specifics as much as generalities. Only by heeding these insights can we hope to keep human rights at the centre of our inquiry in ways that allow for useful, robust, and inclusive documentation.