Abstract
This article studies the interrelation of the epistemic dimension of the State with the operation of social science embedded in institutional spaces of government and its effects on the process of production of social-science knowledge, conditioning its decisions. The reported study is based on the process of poverty measurement as conducted by the Chilean government. In this process, poverty appears as an object created and used not only as a scientific object of academic inquiry, but simultaneously as an epistemic object of government constituted by the State, and used for regulatory interventions. The research developed shows that there are many points of ambiguity in the decisions required for the construction of poverty, in which purely scientific or technical criteria are insufficient for decision-making and are supplemented by tactical or strategic criteria of government.
Introduction
This study of the process of poverty measurement in Chile contributes to the discussion of two interrelated subjects: the epistemic and performative dimension of the State; and the operation of social science when embedded in institutional spaces of government. In particular, it studies the effects that such an interrelation has on the process of production of social-science knowledge.
My direct object of study is poverty as it appears within the State as a cognitive object that orients the design of practices and programs of intervention as well as decision-making processes, in different areas and levels of the public system. In the Chilean case, this corresponds to poverty measured through the Socioeconomic Characterization Survey (Casen), which has been applied since 1985 and has been a central reference for every Chilean government since the dictatorial regime of Augusto Pinochet.
For more than 20 years, the poverty measured by this instrument has focused the national debate and become the basis for multiple decisions. The data generated on poverty have been considered a reliable expression of reality to use for discussion and decision-making about the social situation of the country. This has occurred not only within the State administration but also in the public sphere, where the results of Casen repeatedly appear as an indisputable record of the poverty situation in Chile.
This epistemic object – poverty measured by Casen – is formed fundamentally in a producing network articulated within the State and in which social science takes part, in a very central way, through the participation of a group of experts. The aim of the empirical research reported here is to describe some distinctive aspects of this particular social-science activity which takes place in that network, and to explore the derivations.
In the modern State, one axis of its constitution and action is given by its cognitive equipment, which has the relevant peculiarity of being provided largely by social science. The operation of this complex equipment is further characterized by its performative power, by its ability to constitute social reality. These features of the State have been theorized by Pierre Bourdieu (1989, 1991, 2012), although he did not explore the specific devices involved and their close collaboration with the social sciences, which is our focus here. Regarding this issue, Desrosières (2008a, 2008b) has made a significant contribution by studying the role of statistical procedures and ‘statistical reason’, which are interconnected with the evolution of the State. Michel Foucault’s (2000, 2006, 2007) notion of governmentality, in turn, provides us with guidelines that give additional clues to the understanding of such operations.
Prominent matters on which these cognitive devices of government operate are the various phenomena considered as ‘social problems’ – crime, substance abuse, domestic violence, etc. – poverty being one that has an older tradition as a government object. Studying the operation of these cognitive devices on poverty production, as a cognitive object, provides insights for a better understanding of the epistemic dimension of the State. The configuration and peculiar form of modern State action are made possible by this epistemic dimension, which is added to the State’s legal and administrative power.
My central question concerns the derivations that this merging of social science with governmentality produce both in the operation of social science and in the content of its products. Second is the question of how this epistemic dimension of the State operates. In the pages that follow, my route is: (1) synthesize some ideas of Bourdieu, Desrosières and Foucault on State action, government of conducts and performativity, which provide a theoretical orientation for this work; (2) present our object of study – the measurement of poverty in Chile – and the procedure followed in our investigation; (3) describe key aspects of the production of poverty as an object of government, and identify points of ambiguity or arbitrariness in which the impact of government concerns on the cognitive and practical configuration of poverty is clearly manifested; and (4) highlight some conclusions about the relevance of this for the understanding and study of the connection between social science and the State.
Poverty and the epistemic dimension of the State: Theoretical elements
An extremely important aspect of the epistemic dimension of the State is its cognitive-performative potentiality, its special ability to institute social reality. This way of considering the State is deployed in Bourdieu’s work Sur l’Etat, published posthumously in 2012, which extends and integrates what appears in some of his other works (Bourdieu, 1989, 1991). The State, says Bourdieu (2012: 291), is constituent of social order. Remarkably, for instance, it configures and ratifies the temporal ordering of our lives: the calendar and the hour distribution bear its stamp. The State produces and canonizes classifications; produces and sustains identities. Diverse administrative practices, associated with statistical procedures, engender realities – like epistemic poverty – that resist the tests challenging them, providing a common reference for public and institutional debate and decision-making (Desrosières, 2008a). The State makes society ‘legible’ so that it can be acted on (Scott, 1998). Desrosières has shown the joint evolution between the State and statistical technology – descriptive statistics and probability calculus. Such statistical operations and arguments have been associated with practices organizing society from the beginning, with an inescapable tension between the scientific rhetoric of truth and the rhetoric of political decision. The interaction of these logics, with its associated tensions, Desrosières (2008a: 117) says, is a ‘field of research for a political sociology of science’. Such is the focus of this article, centered on the derivations of those tensions for the social-scientific work involved.
The stability of ontological trust in the State is based in large part on the fact that the State, along with being a fundamental agent in the constitution of our shared reality, molds us as percipient subjects: it incorporates in us the experience of that world as evident. This is what Bourdieu (2012: 292) calls ‘doxic experience of the world’, doxa being the belief that is not perceived as belief. These capabilities developed by the network of elements appearing as the State, with its historical accumulation of power, enable the point of view of the State to appear not to be a point of view. It becomes ‘the point of view of the points of view’ (Bourdieu, 2012: 53), one that prevails in the struggle between points of view in the social world. The way in which individuals become aware of and discuss ‘social problems’ such as poverty is largely defined from that point of view. This role of the State varies in intensity with the society and the historical period. In the case of France, a fundamental object of study for Bourdieu, and Chile, which is ours, the State has been and remains very central in society. On the other hand, the decisive role played by mass media cannot be overlooked; it provides a much more varied doxa than the State, but also propagates its social constructions.
Michel Foucault (2006, 2007) contributes additional elements to the study of such a cognitive-performative dimension of the State. In his final seminars, with his approach to governmentality, he subordinates the disciplinary component, which previously was central in his work, to the modern regulatory mechanisms for conducting conducts. Foucault called ‘governmentality’ a form of power that rests on individuals’ freedom of action, on their autonomy, and which seeks to regulate their action at a distance, in a probabilistic manner, influencing them in a cognitive-performative way, changing their perceived reality and leaving them free to do whatever they decide, acting only on their environment (Cohen, 2011: 47; Foucault, 2006). The ‘governmentalization’ of the State, in modern society, changes its operations. It changes the very constitution of the State. Its epistemic dimension develops enormously. Social sciences become a constitutive and constituent part of the State, become part of its equipment. A significant part of this equipment is the State system of measurement and statistical construction, which has, consequently, a dual character of scientific tool and government tool that marks it constituently. In spite of this, the usual approach is to consider such dimensions separately, without recognizing their interconnection (Desrosières, 2008a). This equipment is the basis of a ‘government by instruments’ (Lascoumes & Le Gales, 2004), a ‘government by numbers’ (Desrosières, 2008b), and a ‘governance by indicators’ (Davis et al., 2012; Hammer, 2011). In modern times, this epistemic role of the State, in connection with science, is crucial for the configuration of society (Carroll, 2006; Wagner, 2003). Such State action makes a valuable contribution to the self-description, self-understanding and self-monitoring of society (Wagner, 2012).
Governmentality relies on some core operations: (1) The constitution of objects of government – to regulate diffuse problems of violence, crime, poverty and the like requires shaping them as clearly identifiable and monitorable targets. This involves building well-articulated theoretical and methodological constructs, convincingly connected with experiential elements recognizable as part of the problems addressed. (2) With such objects already identified it becomes possible to develop intervention programs targeting them. (3) In order to justify and legitimize such intervention programs, narratives are elaborated to depict causalities or responsibilities and mechanisms whose operation is assumed. These narrative accounts, mainly based on the social sciences, are used for framing intervention actions and providing motivation and guidance for them.
Poverty is one of the objects that, early on, were typically configured as objects of State intervention. Governmentality requires giving precise shapes to varied and scattered empirical elements related to the common idea of poverty. That is what the State measurement of poverty has tried to do, at least since the end of the 19th century (Desrosières, 2011). In the case of the measurement of poverty through Casen, its outcomes related to governmentality affect various levels of the State administration. Its most direct recipients are upper-level State officials responsible for the design of policies and programs and the evaluation of their results, for whom such data are crucial in their work and decisions. This poverty, configured and measured as an object and translated into data, serves them as a common focus and facilitates their coordination around this matter. Thus, this epistemic poverty serves to guide and coordinate the activities of a large number of State agents.
The State process of poverty measurement is carried out under the positivist notion of assuming the pre-existence of the measured reality. In contrast to this realistic approach, we use the notion that has been developed by authors like Bruno Latour, in several of his works, and Annemarie Mol (2002), conceiving the objects as a result of multiple operations that constitute them. In this approach, as opposed to the notion of well-defined and self-contained objects, these objects are seen as enacted from outside, as inescapably intertwined with networks of agents (or actants), both human and nonhuman.
Latour has made a variety of studies of the associations and networks of human and material elements that shape objects; these appear simultaneously as an assemblage and as the center of an assembly (a gathering) of convened and participant agents that together configure the objects (Latour, 2000, 2001, 2004a, 2004b, 2005; Latour & Woolgar, 1986). In the case of poverty, economists, sociologists, State authorities and various other human participants help to shape its reality as an object of government, jointly with various technical artifacts. In the process, many empirical elements of the households are selected, mobilized and articulated, in the form of written or digital inscriptions, and, through the mechanisms designed and their human spokespersons, they may, at any time, ‘object’ to what is said about them.
In this perspective, poverty measured by Casen, that peculiar poverty which guides State decisions and which is positioned in the medial public sphere, is an assemblage of selected elements, carried out in the framework of governmentality. As such, it does not exist prior to the process of its measurement and scientific construction (Latour, 2001). However, this poverty, despite its artifactual condition, is fully factual, fully real. The operations that constitute it are real and its product operates in the social reality and has real effects. It is a very real and operational assemblage or assembly, as has been extensively argued by Latour (2001, 2010) regarding these hybrids, or ‘factishes’ as he calls them. It is a constructed and conventional, but at the same time solidly real, poverty.
Empirical research on the measurement of poverty in Chile
In Chile the measurement of poverty has been a matter for government attention only since the 1970s, with the onset of neoliberal transformations. Before 1974 there were no official measures of poverty. Between 1940 and 1970 the social policies and services were in widespread benefit; for example, there were subsidies to the prices of basic products and to generalized free primary education; therefore, it was not necessary to identify the neediest groups. The coverage of those policies varied according to the resources available to the State and according to the pressure various social groups were capable of bringing to bear (Denis et al., 2010). With the military dictatorship of Augusto Pinochet (1973–1990) came a significant reduction in social spending, and the government adopted the policy of subsidizing social segments that presented critical needs, which led it to focus on the neediest, understood primarily as the poorest. This compelled the government to seek to identify those sectors that constituted the target population. The main tools for this purpose would be the Map of Extreme Poverty (1974), based on census data, and the CAS, a card for socio-economic classification (1980), applied locally. In 1985 a more sophisticated instrument was added, the Survey of Socioeconomic Characterization (Casen), administered to a large national probabilistic sample.
Despite being one of the countries that extensively applied neoliberal policies early on (Harvey, 2007). Chile is a country where State intervention in society has been and still is fundamental. On the other hand, a strong positive value is attached by the State to the application of scientific and technical knowledge to government affairs. In fact, the participation of experts in relevant roles in government activities has occurred historically from the beginning of the 20th century, under both authoritarian and democratic governments (Silva, 2010).
The period in which the Casen measurement of poverty was made − 1990–2012 − corresponds primarily to the governments of the Concertación para la Democracia (coalition of center-left parties) until 2009. They have made significant use of social-scientific knowledge (economics, political science, sociology, etc.) to a remarkable extent (Aguilera & Fuentes, 2011; Gárate, 2012; Joignant 2011, 2012), and this has been a feature of all governments of the Concertación (Silva, 2010: 193).
The empirical research for this paper is based on both primary and secondary sources. We obtained and reviewed all the documentation used for the Casen measurement process – manuals, instructions, questionnaires, bidding conditions, minutes, memos, internal evaluations, etc. – and interviewed experts and personnel directly responsible for the design and management of the survey, from the State, university departments and the Economic Commission for Latin America and the Caribbean (ECLAC), including some participants in the beginning of Casen. We conducted 22 interviews and reviewed more than 70 documents related to the process. A controversy generated in 2012, during our research, provided us with additional information, coming especially from four forums that we recorded and analyzed, from acts of a parliamentary committee investigating the matter, and from more than 200 press articles.
Scientific production of poverty as an object of government in Chile
The first Casen survey was conducted in 1985, but it was not intended to measure poverty. As stated by the then director of Odeplan (Office of National Planning), the State organism in charge of the survey, the Casen, under the military regime, was developed and used, in its 1985 and 1987 versions, as ‘a measuring instrument of the redistributive effect of social spending’, to see if State spending was reaching those it was intended to reach, that is to say, the low-income population (interview, 20 May 2013; cf. also Haindl & Weber, 1986; Haindl et al., 1989). In none of the published reports of these first measurements did a description of the poverty situation appear, not even in their table of contents. Poverty as an epistemic object of government had not yet been formed. 1 The focus of these works and the use of the survey was to guide the targeting of social spending. This is a central idea that the military government assumed and the Concertación Government would continue, with only rhetorical adjustments.
Measuring poverty using Casen officially began in 1990, under the first government of the Concertación. This was done in a network coordinated by Odeplan and integrated by a university academic center of the Department of Economics, University of Chile, and the Economic Commission for Latin America and the Caribbean, a United Nations organism, which acquired a key role in the calculations. In 1990, after some discussion, the procedure, according to the appreciation of the participants, ‘was absolutely validated’; ‘it was completely legitimized as a method for measuring poverty in Chile’ (interview with expert in charge at the time, 1 July 2013). In fact, from then until 2010 there would be no significant controversy in the field, only some specific criticisms that did not gain momentum and dissolved spontaneously. Throughout this period, there was not even, as noted by the same expert, any questioning by the government, demanding changes in the methodological procedure or in the definition of the poverty line, or for adjustments of income, or other matters. That year sealed the procedure that would be followed for the next 20 years.
This unquestioned procedure appears as a clear manifestation of an expert type of management, apparently oblivious to political interests or other kind of external interests. Those involved would not hesitate to endorse what Zelizer (2005) calls the ‘model of separate worlds’: we are here in the field of science, which does not mix with the world of politics, despite its proximity. That was the vision that measurement agents proclaimed; that was what the participant experts emphasized. In contrast, my review of the measurement process and my analysis of what happened in those 20 years differ markedly from that perception and appreciation. The analysis allowed us to identify numerous moments of decision in the process of constructing epistemic poverty that the scientific and technical reasons are not enough to resolve, and in which the resolution derives ultimately, in a contingent way, from the interweaving of the technical procedure with the concerns of government, so that normative or strategic government considerations are incorporated in the decision. I have called points of arbitrariness these decision points in which scientific or technical criteria are not sufficient to make a decision.
Table 1 lists 11 such particularly crucial and problematic points. For each one, an estimate of the effects the decision would have on the measurement of poverty is registered, based on the range of possible alternatives that, reasonably, could also have been chosen.
Main points of arbitrariness and estimated effects of the decisions made in them.
I briefly review these main points of arbitrariness, delving deeper into some of them to exemplify with greater detail what happens. In each, I consider the interplay between technical scientific criteria and criteria arising from the use of this measurement for government purposes. To this, I add an estimate of the effects on poverty data provoked by what was decided in each case. For reasons of space, I have left two points undeveloped.
Concept of poverty
The choice of one or another concept of poverty is a fundamental decision that affects the measurement. The concepts of poverty refer to the levels of minimum acceptable well-being, or to capabilities the lack of which limit or prevent proper integration and participation in society (Altimir, 1979; Feres & Mancero, 2001; Haughton & Khandker, 2009; Sen, 2000). There is a range of conceptions that guide this measurement throughout the world. The concept of monetary poverty followed by Chile considers poverty to be the lack of sufficient income to purchase basic goods and services. Multidimensional concepts pay attention not only to the economic dimension but also to other aspects such as employment, adequate housing, access to basic services, protection from aggression against persons and property, and living in conditions of dignity and freedom from humiliation (Alkire, 2007; Feres & Villatoro, 2012). There are also relative concepts, as used by the European Union (which defines poverty as an income level below 60% of the median), and subjective concepts, based on assessments by the subjects themselves of their level of satisfaction.
Poverty rates vary enormously as a consequence of the concept chosen. To illustrate this point, when comparing calculations for 2006, poverty in Chile would be 13.7% according to Casen, 25.5% according to the measure of relative poverty in the EU, and 60.9% according to a multidimensional poverty measure. If what is measured is extreme poverty, the World Bank measure yields a rate of 1.5%; a measure of basic housing needs gives 1.9%; the same measure, excluding household equipment, 3.5%; the Casen measure gives 3.2%; and a multidimensional measure, 9.9% (Denis et al., 2010; Mideplan, 2009).
The choice of one or another conception of poverty has been associated with the options for intervention on the problem of poverty. In fact, the original 1985 Casen was constructed to achieve a more effective targeting of government social spending and only later came to be used to measure poverty (interviews with the economist in charge of the project, 26 April 2013, and the director of Odeplan at that time, 20 May 2013). Poverty defined in terms of income generates accurate and easily attainable targets for monetary intervention. It is a poverty for which easy and rapid improvements through policies based on various monetary subsidies can be achieved. Indeed, delivery of these subsidies has become customary since the early 1990s, and their distribution, especially just before measurement of poverty, has had a significant impact on it. There is an elective affinity between such interventions and a monetary definition of poverty, which has contributed to tilt the preference toward it. For the last survey, I counted 13 different subsidies (support for the newborn, social protection bonus, social allowance, winter bonus, and so on) that had been included in the calculation of income.
Updating the basic food basket
The poverty line is determined on the basis of a food basket that satisfies basic energy and nutritional needs. Such a basket is associated with the consumption patterns of the population and with characteristics of the food supply, aspects that change over time. The basic food basket used by Casen has been the Fourth Household Budget Survey, conducted by the National Bureau of Statistics (INE) in 1987–1988 in Santiago, with a sample of 5076 households. From 1990 to 2012, the calculation of the poverty line was based only on that 1987–1988 survey, although in this period there have been two new surveys of family budgets – the fifth, made in 1996–1997, with a sample also from Santiago (8358 households), and the sixth, made in 2006–2007, with an urban sample of 10,026 households (Alonzo & Mancero, 2011: 8; Schkolnick, n.d.). Therefore, the original basket has not been updated in 20 years. One after another, governments have refused to make the update, despite its having been recommended by the ECLAC experts in charge, according to whom there was no impediment to making such calculations (interview 17 July 2013). Other experts, in advisory committees, have also subscribed to this recommendation. From the perspective of government authorities, the drawback is that it would have driven up the poverty rate by a considerable amount. It is politically inconvenient to announce such unexpected increases in poverty and difficult to explain them to the population.
One researcher made the calculation of poverty by taking data from the Fifth Family Budget Survey, 1996–1997. He found that, with this calculation, the poverty and indigence rates are doubled (see Table 2). In 2006, the total magnitude of poverty, with the update of the basket, could have reached the high figure of 29.0%, but it was left at 13.7%, without the update, due to concern on the part of the authorities about the negative effect this result would have on the public, given that, with respect to the previous measurement communicated three years before (18.7%), poverty had so greatly increased. In the face of this risk, the authorities of various governments have chosen ‘to do nothing’ and ‘not to alter the method of calculation’. It is a ‘not to do’ that certainly makes a big difference.
Poverty and indigence according to the Casen, with the original and revised food basket (in %).
Source: Mideplan (2007); Larraín (2008: 117).
The Orshansky coefficient
In the Casen, the basic food basket is used for the calculation specifically of extreme poverty or indigence; the poverty line, meanwhile, is calculated by multiplying the level of income corresponding to the line of indigence by 2. This 2 is the value of the Orshansky coefficient. In institutional documents and public communication, this coefficient is used and presented as if it were almost a universal constant, and in the last 20 years it has been subjected neither to public scrutiny nor to discussion. The origin of the coefficient has been completely obscured, and experts, who know its meaning well, have not questioned it.
Despite the presentation of the Orshansky coefficient as a constant to be applied without hesitation, this coefficient is the outcome of an empirical calculation. It is the inverse of the ratio between food expenditure and total expenditure, for the reference population (Haughton & Khandker, 2009), which in the Chilean case has corresponded to the second income quintile. Since this ratio comes from a calculation using empirical data on household expenditure, if these data change over time, the ratio should vary too. The figures from household budget surveys show that the proportion between food expenditure and total expenditure has shown a steady decline, in such a way that now the coefficient, according to one of the experts interviewed (1 July 2013), could be near 2.9, which would increase the rate of poverty significantly. In spite of this, the experts and authorities that have handled the calculations have kept the value of this coefficient equal to 2, and this figure has been consistently applied, invoking the power, difficult for laymen to challenge, of scientific formulas. If the coefficient had been updated, the rate of poverty would be greater, which creates a serious reluctance to update it. Experts who know both the variation in this empirical feature and its inconvenient effect of increasing poverty have chosen not to rock the boat, and the increase has passed unnoticed over the years.
Differential adjustment for food and non-food components
The indigence or extreme poverty threshold is determined according to the basic food basket; the poverty line, meanwhile, includes the non-food component as well. The determination of which non-food goods or services to include as basic is highly controversial. A solution adopted in Chile has been simply to estimate their plausible proportion, without specifying them, which is the function that fulfills the Orshansky coefficient. Year on year, the income needed to acquire the basic food basket is adjusted according to changes in food prices. The non-food component is adjusted, through the use of the coefficient, in the same proportion. This was not a matter for discussion during the 1990s and much of the 2000s, until, in 2007 and 2008, food prices underwent a considerable increase in Latin America, which was not accompanied by a similar increase in the prices of the rest of goods and services. In the case of Chile, between 2006 and 2009, while the consumer price index (CPI) for food rose by 32.4%, the rest of the goods made just 6.4%. Between 2009 and 2011, the variations were 13.1 % and 2.9%, respectively (ECLAC, 2012a: 4). In response to this, in 2009 ECLAC decided to differentiate the adjustment of the non-food component, applying to it the variation of the specific CPI for these particular goods and not the variation relating to foods. The effect was that, while the indigence line was not affected, poverty was reduced by 3.6 percentage points in 2009 and by 4.0 points in 2011, in comparison to the results of the Casen, which continued to apply the previous form of adjustment (ECLAC, 2012a).
The incoming right-wing government of Sebastian Piñera, which had to define the final calculations of the 2009 Casen, decided to maintain the adjustment of all goods according to the CPI for food and not to subscribe to ECLAC’s differentiated way of calculating. In support of this decision, the government cited the technical reason of maintaining comparability. Nevertheless, technically it was possible to preserve comparability, adjusting the non-food component for previous years. In fact, ECLAC had already prepared the corresponding calculations to do this (interview 17 July 2013). However, the underlying political reason was that preserving the calculation method meant that the preceding center-left-wing government of Michele Bachelet ended with a rise in poverty of 1.4 percentage points, whereas the ECLAC calculation would have lowered poverty by 2.2 points. The variations were numerically small and barely statistically significant, but they were very significant in political communication, and as such were used by the government and the pro-government political leaders to proclaim the failure of the Bachelet government in the battle against poverty. A main critic of this decision at that time, and subsequently, was Andrés Velasco, President Bachelet’s Minister of Finance, who proved to be very aware of the political effects of these statistics. According to Velasco (2012: 20), ‘the whole [political] debate Chile has had in the last three years would be different’ if the ECLAC calculation had been followed, a calculation that, according to him and other economists, was ‘the correct one’. The counter-criticism, exposed by former Mideplan Minister of the Piñera government, Felipe Kast (2012), was that such an adjustment would have involved reducing poverty by bureaucratic definitions: ‘3.6% of poor people would cease to be poor for purely administrative reasons’. Behind both claims to the technical correctness of the supported decisions appears a concern for the political resonance of the alternatives at stake. In such a way, concerns about sociopolitical contingency infiltrate measurement decisions.
Inclusion of non-monetary government benefits
In the Casen calculation of household incomes, subsidies or social benefits in kind, such as lunches, breakfasts and other foods delivered by the supplementary feeding programs of the National Board of Student Aid and Scholarships (JUNAEB), are not included (Ihnen, 1988; Larrain, 2008: 145). In addition, there is public-sector spending, differentially distributed, in education and health, which can be considered an indirect monetary transfer, but that is also excluded (Henoch & Troncoso, 2013). Technically, though, such free goods and services could be assigned a monetary value (Beccaria et al., 1997: 91; Citro & Michaels, 1995).
The decision to exclude these non-monetary benefits stems from the origins of the survey and was one of the issues initially discussed. The declared reasons for exclusion were pragmatic. For instance, it is difficult to assign value to the health service, which is used only sporadically. Education, meanwhile, to the extent that it is a general benefit, contributes little to the differentiation between poor and non-poor: its contribution to income covers a very wide range of the population (interview 26 April 2013).
The inclusion of these benefits in the calculations would increase households’ incomes and reduce poverty. Nevertheless, insofar as they were not included from the start, their absence has not been questioned, even though it has effects on policies. Deciding not to include State benefits that are not direct monetary subsidies reinforces the importance of using monetary measures to reduce poverty; the other benefits, of more complex application, will therefore not be privileged when ‘fighting poverty’. This is an indirect effect of a methodological decision. This exclusion is associated, from the beginning, with the preference given to the use of direct monetary subsidies to reduce (monetary) poverty.
Equivalence scales
The size and composition of households influences their food requirements. In larger households, there are economies of scale, and having children involves a smaller cost. In this regard, the international literature on poverty measurement recommends using equivalence scales (Alonzo & Mancero, 2011). To calculate such equivalence scales, Engel curves and demand equations are used, applying various methods like Rothbarth’s or Prais-Houtthakker’s (Alonzo & Mancero, 2011). The results obtained by the different methods, however, are markedly different, due to conceptual and analytical differences, and there is no consensus on what is the most appropriate method (Haughton & Khandker, 2009: 28).
The application of equivalence scales would leave out of poverty many households whose average per capita income is only slightly below the poverty line, by the mere fact of having more members or including children. This, therefore, could bring down poverty by several percentage points (see Haughton & Khandker, 2009: 88).
If, technically, there are different procedures available and the introduction of equivalence scales is recommended by the experts, why then are they not applied in Chile? The exclusion of equivalence scales at the beginning, in 1990, becomes a reason for not including them later, since it would make retrospective adjustments necessary in order to maintain comparability, and these are difficult to explain to the public. Moreover, the presentation of results referring to ‘equivalent adults’ does not have the clarity and concreteness desirable for communicating with a wide audience. Instead of a pristine statement such as ‘140,000 Chileans have moved out of poverty’, the expression would now be ‘140,000 adult equivalents …’. No politician or government authority would feel comfortable appealing to these abstract statistical entities. These practical reasons affect a substantive decision that contributes to shaping poverty.
Incorporation of new questions in the questionnaire
The determination of household income is technically complex. It is necessary to cover a great variety of situations that generate income and different income types. Over the years, the accumulated experience of repeatedly administering the questionnaire would help to specify new questions or review and adjust the existing ones. However, if these changes are very significant, they affect the comparability of the measurements over time. This creates a tension between two technical objectives: to achieve more reliability and validity vs. to maintain longitudinal comparability.
In the 2000, 2003, 2006 and 2009 surveys, several changes were made in the questionnaire, the most important being those to the 2006 Casen. Along with several other modifications, such as moving toward the center of the questionnaire the questions on income which, until 2003, were located near the end, the most important change was the addition of 20 new direct questions associated with different revenue streams collected by the survey (OS-MDS, 2012: 22). In fact, the specific sources of income, as ECLAC registers them, without considering subsidies, grew from 30 in 2003 to 54 in 2006 (ECLAC, 2012b: 71). Thus, direct questions about matters such as pay for overtime, annual and monthly bonuses, food stamps, gratuities, benefits in kind (food, drink, clothing, childcare or nursery services, etc.) were added.
If we review the results of applying the definitive questionnaire to the total sample of the 2006 survey, according to ECLAC records, the income from private bonuses and benefits rises from CL$ 18,068 million in 2003 to CL$ 95,720 million in 2006. On the other hand, the income in kind from the main occupation rises from CL$ 9,451 million in 2003 to CL$ 31,247 million in 2006. The total increase attributable to new questions, then, would be CL$ 99,447 million, 2 an increase of around 300%, discounting the effect of inflation, and which in turn is reflected in an increase in the aggregate figures of remunerations received by the respective wage-earners.
Before administering that 2006 survey, it was pretested, looking for possible effects of the changes made in the questionnaire. There were no statistically significant differences in the amount of income registered between the old questionnaire and the new (OS-MDS, 2012: 26). However, in the opinion of some experts involved, the majority of the new questions are applicable to relatively small fractions of the population, and therefore the pretest, with a small sample of about 1200 cases, is not sufficient to detect, with statistical significance, such effects, even if they exist (Casas-Cordero, in forum 31 October 2012).
In 2006, there were no political bets placed on the Casen results. The comparison would have been between two governments of the Concertación’s parties (Ricardo Lagos’ and Michelle Bachelet’s governments) and it was not crucial whether the poverty figure rose or not by a few tenths. The situation meant that the new income questions added were not a contentious matter, and there were no disputes.
In contrast, in 2012 the comparison was between politically opposed governments. So, one of the high points of the controversy surrounding the 2012 Casen occurred as a result of the inclusion of a new question in the questionnaire, and its effects on income. In the 2011 version of the Casen, the MDS (Ministry of Social Development) added a new question, y11, which asks about labor income received for work, activities or business done by unemployed or inactive people and non-formal family workers, that is, by people to whom questions on labor income are not applied since they are not considered employed in the reference week. The definition of employed or unemployed refers to the previous week (reference week), but earnings are consulted regarding activities in the previous month. This gap allows an unemployed person (last week of the month M) to have labor earnings the previous month (M–1). That is the reason for including y11 given by the MDS in a memo to ECLAC. 3 A central argument was that question y11 responded to the need to better identify the labor earnings of the unemployed and inactive, which, until the 2009 Casen, had been recorded, although in a nonspecific manner, in the residual question ‘other income’ (y17d in the previous survey). Therefore, the MDS concludes, including y11 is an improvement of the questionnaire that does not include any new revenue streams (cf. OS-MDS, 2012), i.e. it improves reliability and maintains comparability with previous measurements.
The ECLAC expert in charge of Casen initially excluded this question from the calculations, but after insistence from State authorities and having received a detailed written argument, allegedly supported by evidence from the survey experience, provided by experts from the Ministry of Social Development, the expert finally accepted its inclusion (Hernando, 2012). In the document with its final decision, ECLAC says: ‘whereas it is difficult to resolve with crystal clarity which is the alternative [include this variable or not in the aggregate household income in 2011] that best meets the purposes of maintaining the comparability of income measurements between the two years, and while there is no doubt that y11 represents a type of labor income in fact received by people in the reference month, it was finally decided to include it as part of the total household income in the last year’ (ECLAC, 2012b: 26, 27).
Nevertheless, while accepting the arguments of MDS, ECLAC makes it clear that the point is debatable (‘it is difficult to resolve with crystal clarity’) and in a footnote adds information that provides a new reason to have doubts. It goes on to say that the variable ‘other income’, which in the 2009 survey had reported revenues of CL$ 2152 million with 69,934 recipients, recorded in the 2011 Casen CL$ 1477 million and 28,602 beneficiaries. That is, it has decreased, as expected, since in this second survey many of these incomes and recipients are incorporated in y11. But, y11, which should benefit from this migration, reported revenues of CL$ 24,861 million and 205,503 recipients, and this is a much bigger growth than expected: 37 times the amount of revenue and five times the number of recipients. ECLAC mentions this and registers it, but does not make an argument of it.
Had question y11 been excluded, as initially proposed by ECLAC, the poverty rate would have been 15.0%, instead of 14.4%, which was the rate finally announced. That 15.0% would have meant that there was no difference from the 2009 figure (15.1%). Indigence, meanwhile, would have been 3.2%, instead of 2.8%, which is not statistically significant compared with the 3.7% of 2009. That is, y11 was important to the discourse proclaiming the victory over poverty announced by government, showing numbers that would warrant the decline in poverty and indigence. The convenience of the changes for the government was obvious enough to many following the debate on Casen and was a reason for the fierce controversy generated around the inclusion of question y11.
The claim that these revenues were already being measured in the previous survey question under ‘other income’ is plausible and is supported by some of the information provided. On the other hand, it is evident that the inclusion of y11 generates an income report far superior to the old question on ‘other income’, which affects comparability. Nonetheless, that is not attributable to the fact that this stream of income was not previously covered – it was, although in other ways, but to a cognitive effect of memory activation. 4 This, however, is a different problem. If the existence of cognitive effects is argued, this judgment criterion should also be applied to the previous surveys, and this would lead to a review of several of them, particularly the 2006 Casen. This uncharted problem made it advisable, for the experts who raised criticisms, not to follow that path. 5
We then come to the following premises: (1) The inclusion of y11 involves a clear improvement in the quality of the measurement of income. (2) Empirical evidence plausibly supports that these revenues were previously registered. (3) Inclusion of the question causes a marked increase in income, thus affecting comparability, but this time for reasons never used before to excluding a Casen question.
What to decide, technically, in this case? Again we see that the technical criterion is not sufficient in itself to decide; it is not enough to incline the decision toward including or excluding the question. Paradoxically, excluding y11 because of the increase of income that it causes, affecting comparability, would imply applying the reason of cognitive effects, which is a new consideration and, if allowed, would create another, even more serious, problem of comparability.
How to decide under these conditions? The arguments and data can be orchestrated in ways supporting both inclusion and exclusion. All the players also say about themselves that they are acting technically. An expert and authority of ECLAC, when questioned by a parliamentary committee on the inclusion of y11, categorically stated that ‘the agency adopted a procedure that was absolutely technical and with the best reading on the information available’. 6 Even so, these experts are forced to recognize, as one of them says, that ‘there is a zone of ambiguity. Ultimately, one is forced, with all the available information, to wager.’ 7 The decision made in the area of ambiguity, however, is unavoidably tainted by the preferences and interests of the agents involved. All the participant experts knew the effects of one decision or another on the results of the measurement and none of them acted under the veil of ignorance. None, moreover, was an isolated decision-maker; they all were members of teams, located institutionally, connected with authorities, groups and organizations, with preferences or political sympathies. This applies to MDS as well as ECLAC experts, and to all external, expert or non-expert, critics. These networks push the decision, albeit imperceptibly, in one direction or another.
Imputation of income answers
From the beginning of the Casen, the Chilean government commissioned ECLAC to do the work of evaluation and correction of measurement errors in the reporting of income collected by the survey. In this regard, a first task performed by ECLAC is the correction for non-response. The non-response is particularly high, in all countries, with respect to self-employment. For the calculation of income in the USA, about 25% is obtained by attribution and not by direct answer (Feres & Villatoro, 2012: 49). In addition, higher non-response rates occur at the extremes of the distribution, making it a major problem in measuring poverty. In the 2011 Casen, 5.4% of the employed people did not report income. Thus, on the basis of their calculations, ECLAC imputed income to 349,312 employed persons, with an average income of CL$ 430,666 (ECLAC, 2012b: 80). It is certainly a significant amount to be attributed, considering that for that year the poverty line was CL$ 72,098. Why, then, are those who do not respond not simply excluded from the calculations? The reason given is that non-respondents have a different distribution from the rest of the sample, and to extract them would impose making sample adjustments and modifications in the expansion factors (ECLAC, 2012b: 35).
The imputation of income is made considering the attributes of each income recipient who does not report it, and attributing to them the same level of income reported by people with similar characteristics. The characteristics considered are seven: occupational category, region, condition of head of household, sex, educational level, branch of economic activity and occupation (ECLAC, 2012b: 35). The procedure, called ‘conditional means method’, consists of imputing the missing income according to the values found in similar sub-samples formed according to the simultaneous crossing of those seven variables.
As is predictable, this imputation increases the income of the surveyed sample and reduces poverty. In 2009, growth rates generated by these ECLAC calculations were 5.4% in total income, 8.1% in salaried employee income, and 11.5% in self-employed income; in 2011, the respective rates were 2.2%, 6.0% and 7.3% (ECLAC, 2012b). There are, however, other methods of imputation, in addition to the conditional means method, and all give different results. A study using data from Argentina in 2007 showed that, comparing the results from diverse methods, the rate of indigence varied between 10.3% and 12.8% (Medina & Galvan, 2007); that is, the mere choice of imputation method increased or decreased poverty by 2.5 percentage points. The experts who participate and who are part of the government have clear notions about the effects of the choice of methods on poverty figures. Nevertheless, the extent to which this choice impinges on decisions made is something that remains obscure.
Adjusting income comparing with national accounts
Once the imputation of missing income is completed, ECLAC checks the degree of error in the total income registered. For this, a pattern of comparison is needed that conveys the ‘true value’ of the income distribution. The pattern used for comparison is the income registered in the national accounts provided by the Central Bank of Chile. Put simply, the adjustment method consists in adding income to or subtracting income from each category of income, according to the discrepancies observed between the survey and the respective national account, when comparing the global amount registered in each category.
As a product of the ECLAC processing, the resulting average income from the salaries of employees, that in the 2011 Casen is CL$ 369,641 and rises to CL$ 375,300 after correcting for missing responses, finally, after adjustment according to national accounts, reaches CL$ 410,578. The increase, however, is particularly marked in the income from self-employment, which rises from CL$ 279,072 in the answers of respondents to CL$ 286,793 in the correction for no declaration and, finally, to CL$ 589,073, with the adjustment based on national accounts, representing a 111.1% increase over the original figure.
Such an adjustment is made on the assumption that the national accounts are a reliable record of all the income received in the country and its distribution among the different streams considered. This, however, is merely an auxiliary hypothesis and is highly debatable. A particularly important aspect is that the referent constituted by the national accounts is much less univocal than it looks. In fact, for this comparison process, ECLAC experts must not only establish homologations between different income categories of the survey, on the one hand, and national accounts, on the other, but also between different series of existing national accounts. These accounts have undergone several methodological changes since the beginning of the Casen. In 2002 there was a change from base 1986 to base 1996, which had another structure. In 2007 a new series with base 2003 was established. In 2012, the same occurred with another new series using base 2008. These successive changes affect both the definitions of the different income variables and the respective amounts. These changes, says ECLAC (2012b: 7), ‘create obvious difficulties for reconciling the concepts and categories, not only between national accounts and the Casen but also between national accounts series made with different base years and methodological characteristics’. The problem, according to ECLAC experts, would have been especially marked since the change enforced from 2002, which added new categories and removed other classifications. As a result, by 2008, differences between the results of the current series and the earlier were significant. For example, GDP in the 2008 version is 5.2% higher, and the level of final consumption expenditure of households is higher by 8.2% (ECLAC, 2012b: 12). Also, ‘mixed income generated by households for 2008 increased approximately 65% in the new base with respect to the estimate in the previous series’, and the wages of employees fell 2.9% (ECLAC, 2012b: 13). That is, these different values for the same year are due solely to the fact that the calculations are based on national accounts built with different methodologies. While national accounts are taken and treated as an expression of ‘true value’, no doubt this is a slippery and elusive truth.
It follows from the above that full adoption of the new national accounts series would involve having to change the poverty calculations from previous years in order to remain comparable. What ECLAC has done, however, is what their experts call a ‘simple fitting’ of the four series of national accounts, maintaining the original base to protect historical comparability (ECLAC, 2012b: 57). This involves translating the variations in income of the new series to the categories and procedures of the old series, and using these categories for the present-year calculation of poverty. This is an operation of sophisticated scientific plumbing, with many risks of mismatching and error, although they do not appear to get reported.
The alternative to these fittings and adjustments, which cloud the process and make it very difficult to follow and replicate, would have been, with each new series, to recalculate poverty backwards. This would have been methodologically precise and clear. Nevertheless, what is methodologically clear for experts, introduces confusion when presented in the public sphere. For example, in 1998, having presented the figure of poverty at 21.7% as a fact of reality extracted by a rigorous scientific procedure and characterized by its objectivity, how would the government explain to the public, through the press and television, after a few years, that this number must be adjusted by several percentage points, and another few years later, that it is necessary to adjust the figures yet again? It would give the audience the uncomfortable impression that the State authorities were moving poverty data at will, contradicting the common idea of scientific rigor and the realist (positivist) conception accompanying all these processes of State measurement. It would reveal a poverty that did not correspond to its perception in the public and governmental spheres.
Adding further complexity to the process, the adjustment factors actually employed, based on national accounts, have been the object of … adjustments, as stated in the ECLAC report. ‘Special treatment’, for example, has been given to the category of mixed income, ‘because it shows an excessive increase [20.3%], in the context of the evidence on the growth of national product and employment in the period’ (ECLAC, 2012b: 54). Therefore, the report concludes, ECLAC ‘has chosen to apply a special, essentially conservative approach, which is, in the case of self-employed and employers’ earnings, to maintain the adjustment coefficient calculated in 2009, thus giving preeminence, in these categories, to the variation shown by the data from the survey itself’ (ECLAC, 2012b: 56). Without this adjustment to the adjustment factor, the coefficient would be markedly higher than that applied in 2009.
This reveals that the benchmark (the national accounts) has some serious reliability problems, even for the ECLAC experts who have championed its use, making it necessary to correct the adjustment using a ‘conservative interpretation’ which assumes that the rate of increase derived from the calculations is ‘excessive’. This adjustment of the adjustment is not based on new calculations but on an interpretation. The adjustment of survey income to national accounts is thus subjected to an economic hermeneutics that allows for some of the calculation results to be ignored. That is to say, the gap between the survey and national accounts determines the adjustment until the experts decide that it does not determine it.
These details highlight the blurred margins of technical decisions. There is no doubt that technical criteria structure the whole process, but continually there are decisions for which the technical or scientific reason does not provide sufficient criteria. What drives the decision to adjust income, on one occasion, according to the national accounts and, the next time, using a ‘conservative estimate’? The reason is that the expert in charge knows economics, and there are figures that do not fit with his knowledge about the economic reality of the country. But, how does he define the precise numbers of the adjustment, how does he choose how conservative to be with them? Another expert in the same situation could reach a very different decision. This is the result, as one of the experts interviewed says, of the expert’s ‘common sense’; a hazy sense, whose operation can be infiltrated by economic prejudice, political interests, fears, or whatever, which usually occurs unreflectively and without much awareness.
In any case, the main problem with the ECLAC procedure for income adjustment is not the chain of debatable conventions adopted, but the lack of transparency in the operations, which has been maintained for many years. During the 2012 controversy, there was agreement to indicate that the ECLAC process appeared to be a black box for the different actors involved. Even the undersecretary of MDS noted that the process of correction and adjustment carried out by ECLAC ‘is a somehow blind process for the ministry as well; there are things that we are not able to replicate’. 8
Conclusions
The present analysis shows poverty, as an epistemic object, taking shape at the intersection of the criteria and procedures of science with the concerns and tactical and strategic criteria of governmentality. Such a combination enhances the social-scientific work, allowing the assemblage of a large operational data production, and invigorates the performative power of the resulting construction. The very complexity of the measurement operation makes it resource intensive. Each measurement using Casen in recent years has involved budgets of over three million dollars, a figure that exceeds by far what any research center in the country could fund. The foundation in the State and the relevance of this measurement for governmentality also assures the circulation of this construct within the State institutional space and the public sphere.
On the other hand, the framework of governmentality leads to the fact that, at the various points of ambiguity, where technical or scientific criteria are insufficient for deciding the concerns and interests of government are employed to prioritize and decide. We have shown some cases where this happens and their effects on the poverty results. From the wide range of alternatives justifiable in technical or scientific terms, the strategic or instrumental criteria of governmentality induce selection of those most suitable for its purposes, according to the interpretation made by State officials or according to the anticipation of such convenience made by the experts themselves. By this means, the tactical and strategic criteria of governmentality enter into the cognitive process of science and hybridize the decision criteria that constitute or enact poverty. The scientific construction of poverty is thereby tinged by the criteria of governmentality. The resulting poverty is fully marked by the contingencies of that process, which can cause a wide range of variation in the cognitive configuration of poverty and thus in the ensuing figures.
We have seen, for example, how the technical argument about the necessity to maintain comparability of the measurement through time is not sufficient for the decision. There are sound reasons for updating the basic food basket and the Orshansky coefficient, and yet this has not been done, markedly affecting comparability, without being impugned. Any alteration of the questionnaire affects comparability, however many modifications have been made since 1990, but only once has this been questioned. That is, the application of the comparability argument follows a course that is not explained solely for technical or scientific reasons.
Thus, the poverty that takes shape is a peculiar artifactual social reality engendered in that union between science and governmentality. This is how social science works in association with governmentality. Its constructions achieve wide circulation and acquire great visibility and performative force, but the cost is that the technical and scientific uncertainty points are consistently resolved under the pragmatic criteria of government, whether these are of a tactical or a strategic character.
The way in which poverty is produced, as a cognitive object within the State, is articulated with the aim of designing and implementing forms of social intervention. The entire complex equipment for measuring poverty is a major cognitive instrument of government. It is a governmental instrument of observation for monitoring social reality, guiding intervention and giving a public account of what has been achieved. It is a piece of cognitive equipment that simultaneously has performative power, shaping social reality. It is part, together with a family of similar instruments, of the increasingly sophisticated epistemic and performative dimension of the State, based on social-scientific knowledge. These measuring devices become ‘obligatory passage points’ (Callon, 1986) for public action, and the networks that give form to State action must inevitably pass through them.
This construction of poverty involves important normative dilemmas. At stake is what is meant by wellbeing of the population and how to define the respective acceptability thresholds. In the way in which Casen was constituted, this normative debate of large political projection took place behind closed doors, between experts and government officials. Never, in 20 years, was it a matter of open debate that allowed a wider representation commensurate with the political relevance of the matter. The presence of experts, with their high qualifications and experience as well as the technical complexities of the measurement were used, and still are used, to display and frame the measurement as a technical matter, to be addressed by experts, thereby closing it to any further and wider discussion. In this way, social science is used to close the normative debate on this epistemic construction, concealing its relevance and political content. Achieving this effect of closure requires keeping invisible the points of ambiguity or arbitrariness, whose normative aspects, then, are not resolved according to due process with adequate political representation. The present article has sought to shed light on precisely these points of ambiguity and the way in which they are decided.
Footnotes
Acknowledgements
A first version of this article was presented at the Annual Meeting of the Society for the Social Studies of Science, in San Diego, California, 9–12 October 2013. I am grateful for the collaboration of Fernando Valenzuela, co-researcher in the project, and the valuable participation of Francisco Salinas and Alba Vásquez in the fieldwork.
Funding
This work was supported by the National Council of Science and Technology of Chile under Fondecyt grant number 1121124.