Published on September 30th, 2019
Written by Diego Cerna Aragon, CMS Graduate Student, Research Assistant, GMTaC Lab
SISFOH interface. Photo by author.
How do humans make sense of things? How does this process differ from the ways algorithms make sense of social reality? During 2016 and 2017 I was part of a team of researchers1 that studied how an algorithmic system for socioeconomic classification implemented by the Peruvian government worked in geographical areas where the level of state capacity2 has been historically low. The main focus of the research was to explore the inner workings of this system at a local level – especially the routines of bureaucrats working in municipalities. In this post, I will discuss how people – bureaucrats and citizens – interacted with an impersonal and opaque algorithm, whose inner workings exceeded their comprehension yet still required their actions in order to function.
Introducing the SISFOH
The SISFOH (Sistema de Focalización de Hogares, or House Targeting System) is an algorithmic system that assigns each Peruvian household a socioeconomic classification (extremely poor, poor, and no poor). This system is used to determine who can access to social welfare programs that provide benefits such as subsidized healthcare and scholarships in higher education, among others. The system was born as an effort by the Ministry of Economy and Finances (MEF) – the most influential governmental office in Peru since the neoliberal restructuring in the nineties – to control the expenditure of the public budget in social services.
The Ministry’s objective was (and still is) to avoid what they consider an inefficient use of public resources. In technocratic jargon this is called filtraciones (leakages), which means people who may not need state assistance yet still obtain it, provoking the public budget to “leak.” This metaphor of leakages3 functions within a public management paradigm in which the public expenditure in social services is seen as something that should be efficiently streamlined – even if the process of streamlining itself requires a considerable amount of public spending. This is in clear contraposition with alternative approaches that would consider these services as basic and universal (Remy, 2016). Today, SISFOH is executed by the Ministry of Development and Social Inclusion (MIDIS). This ministry is in charge of streamlining the expenditures of almost every social welfare program in Peru.
Crafting and using a black box
The early history of the SISFOH at MIDIS is illustrative of how algorithms incorporate human judgment in what is usually considered purely “objective” information outputs. Between 2011 and 2013, during the project’s transfer from one ministry to another, a couple of things changed. The then newly elected government of Ollanta Humala tried to infuse the welfare programs with a more “inclusive spirit.” While Humala’s government did not renounce the neoliberal dogma of streamlining social spending, it favored a perspective that not only avoided over-coverage (i.e. excessive leakages), but also prevented under-coverage (i.e. precluding people who needed social benefits from receiving them). With this political criterion in mind, the newly appointed officials at MIDIS convened a team of experts in social sciences and public policy technocrats to reform the algorithmic system that they received from MEF.4
Also included in this system are Peru’s municipal bureaucrats, who are in charge of operations at a local level and who interface directly with potential beneficiaries of social service programs. Since there is a sense of distrust and suspicion towards these bureaucrats – given their closeness with the local populations – the process of how the algorithmic system produces citizens’ socioeconomic classifications has been intentionally kept opaque, as a “black box” (Callon and Latour, 1981; Latour, 1999). Technocrats call this practice crear candados (creating locks). For ministerial technocrats, “creating locks” is a strategic component of the SISFOH because they believe that this practice prevents other actors – municipal bureaucrats and potential beneficiaries – from trying to game or manipulate the system (e.g. practices of clientelism, where bureaucrats help applicants to get a certain socioeconomic classification to access welfare programs in exchange for political support for the incumbent mayor). In other words, state technocrats believe that opacity is necessary for an efficient distribution of public economic resources.
These two sides of the SISFOH reveal political and economic dimensions of an algorithm: much more than a mathematical equation, it can mean different things at different times for different people. Whereas for experts and technocrats the algorithm is a kind of collaborative craftwork that integrates human judgment and political interests, for municipal bureaucrats it is a black-boxed tool that they deploy without knowing how exactly works.
Municipal bureaucrat work space in a rural area of the Peruvian southern highlands. Photo by author.
The unintended consequences of the desire for efficiency
Given that neither the municipal bureaucrats nor potential beneficiaries know how the Peruvian state’s classifications are calculated, they come up with their own explanations, making inferences based on their practical experience. For example, the bureaucrats observed that people with registered properties are less likely to be considered “poor enough” by the system, even if these properties are just empty plots of land. Some refer to these deductive analyses as “folk theories”: intuitive, non-authoritative explanations that people share informally, and use to interpret and intervene in their social realities (Gelman and Legare, 2011; Eslami et al., 2016).
Such analyses are not limited to an abstract or symbolic realm, but guide the concrete actions of the people who ideate them. Here is another example: bureaucrats in another urban municipality noticed that the socioeconomic classification of applicants cannot vary immediately but it may change after a period of around three to four weeks. In an effort to help citizens in precarious situations, the bureaucrats suggested that applicants who did not obtain the classification they required could return to the municipality and reapply after that period of time. As a consequence, people applied multiple times trying to get the classification “extremely poor” or “poor,” which qualified them for the social welfare program they needed. The side effect of this “gaming of the system” was that the municipal office of the SISFOH – which already had human resources and budget limitations – became overwhelmed with applications, slowing down the whole process for all applicants in the district.
In this case, the opacity that was intended to keep the algorithmic system running efficiently backfired. Not providing the municipal bureaucrats with detailed information about the algorithm’s calculation of socioeconomic classifications led the bureaucrats to adopt theories that they intuit in their daily routines, producing unintended consequences for the system.
Visiting applicants for household registrations in an urban area in the south of Peru. Photo by author.
Seeing like an algorithm, feeling like a human
The interactions between the bureaucrats and applicants are interesting moments in which one can observe the concrete consequences of algorithmic opacity. In one district, during a household registration–the phase of the application process in which bureaucrats visit applicants’ houses to verify their challenging living conditions –one of the applicants expressed her unconformity. As the bureaucrat warned her that she may not get the classification that she needed to receive social benefits, she replied, “You are only completing a form, you are not seeing reality.” As we left the applicant’s house, the bureaucrat told me privately, “The government should take into account what we see in our visits.”
This desire to convey what is witnessed during visits is, of course, impossible. The form that bureaucrats complete during household visits is determined by the data that the algorithm relies on to calculate the socioeconomic classification (such as the number of members of the household, the materials of which the house is made, etc.). There is no room for personal details or notations. Even if a bureaucrat considers the person they are evaluating to be “poor” or “extremely poor” using their own criteria, the algorithm may “see” things differently based on the data inputs that the bureaucrats themselves collect on their forms. Municipal bureaucrats in charge of other parts of the social benefits process also expressed a similar sense of alienation. In the same municipality one of the digitadores (a data entry worker in charge of transferring data from forms to the system), was asked about how the algorithmic system handles the applications and told me, disgruntled, “They do it like a computer, there are no feelings,” as she continued feeding the system with the data collected over the previous days.
As some have argued, the simplification of complexity is a “dangerous necessity” (Law and Bijker, 1992) because otherwise complexity would be impossible to handle. Thus the state, in order to intervene in and shape social reality, simplifies the complexity of a population by turning it into a large-scale data set that can be processed expediently. Nonetheless, this effort to apply a more “scientific” method to distribute public resources seems to produce a feeling of alienation among Peruvians who enter into contact with the system. Paradoxically, technocrats’ attempt to offer an algorithmic system that they consider a more fair and efficient way of distributing economic resources may actually undermine the legitimacy of their knowledge and methods.
In this post I have tried to give a quick overview of how the algorithm for socioeconomic classification of the SISFOH is present at different levels in Peru (at ministries, municipalities, and among citizens), with emphasis in geographical areas where state capacity has been historically low and during a very specific period of time (2016-2017). By no means is this offered as a complete and nuanced description of how the SISFOH works in all its complexity. Nonetheless, qualitative incursions such as the research that originated this text could give us clues of where and how to interrogate and analyze state algorithmic systems.
1 The other members of the team were Luis Garcia Ayala and Felix Puemape. Both of them are currently PhD students in Political Science at Temple University. The research has been published in the book Acá no hay ventanillas. La burocracia de la calle en los programas sociales and in the research report Posibilidades y limitaciones de la eficiencia de la gestión de la focalización de hogares en la región Arequipa: un estudio de su implementación desde los burócratas locales.
2 Here I am referring to political science’s concept of state capacity: “the capacity of the state actually to penetrate civil society, and to implement logistically political decisions throughout the realm” (Mann, 1984). This concept is particularly relevant in developing countries, such as Peru, given the existence of “brown areas” (O’Donnell, 1993): areas where state capacity is very low or nil.
3 For concrete examples of the deployment of this metaphor, review the work of Peruvian economist Enrique Vásquez Huamán (e.g. Vásquez Huamán, 2006).
4 It is worth mentioning that recent comparative evaluations of social welfare programs in the region have shown that the level of ‘leaking’ in Peru is below the Latin American average (Robles et al., 2015).
Callon, M. and Latour, B. (1981). “Unscrewing the Big Leviathan; or How Actors Macrostructure Reality, and How Sociologists Help Them To Do So?”. In Knorr-Cetina, K. and Cicourel, A. (eds.), Advances in Social Theory and Methodology: Toward an Integration of Micro- and Macro-Sociologies: 277-303. London: Routledge & Kegan Paul,
Cerna, D., Garcia, L., Puemape, F., Sosa, P., Renteria, M., and Rozas, L. (2017). Acá no hay ventanillas. La burocracia de la calle en los programas sociales. Lima: Instituto de Estudios Peruanos
Eslami, M., Karahalios, K., Sandvig, C., Vaccaro, K., Rickman, A., Hamilton, K., and Kirlik, A. (2016). First I “like” it, then I hide it: Folk Theories of Social Feeds. In Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems: 2371-2382. DOI: https://doi.org/10.1145/2858036.2858494
García, L., Cerna, D., and Puemape, F. (2018). Posibilidades y limitaciones de la eficiencia de la gestión de la focalización de hogares en la región Arequipa: un estudio de su implementación desde los burócratas locales. Lima: Consorcio de Investigación Económica y Social. Available at: http://www.cies.org.pe/es/investigaciones/analisis-sectorial/posibilidades-y-limitaciones-de-la-eficiencia-de-la-gestion-de-la
Gelman, S. and Legare, C. (2011). Concepts and Folk Theories. In Annual Review of Anthropology. Vol 40: 379-398. DOI: https://doi.org/10.1146/annurev-anthro-081309-145822
Latour, B. (1999). Pandora’s Hope: Essays on the Reality of Sciences Studies. Cambridge: Harvard University Press
Law, J. and Bijker, W. (1992). “Postscript: Technology, Stability, and Social Theory”. In Law, J. and Bijker, W. (eds.), Shaping Technology/Building Society. Studies in Sociotechnical Change: 290-308. Cambridge: Massachusetts Institute of Technology Press
Mann, M. (1984). “The Autonomous Power of the State: Its Origins, Mechanisms, and Results.” In European Journal of Sociology, vol 25, issue 2: 185-213. DOI: https://doi.org/10.1017/S0003975600004239
O’Donnell, G. (1993). “On the state, democratization and some conceptual problems: A Latin American view with glances at some postcommunist countries”. In World Development, vol. 2, issue 8: 1355-1369. DOI: https://doi.org/10.1016/0305-750X(93)90048-E
Remy, M. (2016). “La política social en el Perú: entre la focalización y los servicios universales”. In Grompone, R. (ed.), Incertidumbres y distancias: El controvertido protagonismo del Estado en el Perú. Lima: Instituto de Estudios Peruanos
Robles, M., Rubio, M., and Stampini, M. (2015). Have Cash Transfers Succeeded in Reaching the Poor in Latin America and the Caribbean?. Inter-American Development Bank. Available at: https://publications.iadb.org/en/publication/17026/have-cash-transfers-succeeded-reaching-poor-latin-america-and-caribbean
Vásquez Huamán, E. (2006). Programas sociales ¿de lucha contra la pobreza?: casos emblemáticos. Lima: Consorcio de Investigación Económica y Social, y Universidad del Pacífico. Available at: https://www.mef.gob.pe/contenidos/pol_econ/documentos/Programas_Sociales_EVasquez.pdf