At the Sixth Annual Meeting of the Pennsylvania Society to Protect Children from Cruelty in 1883, Reverend Charles E. Ames read from its annual report, arguing that “the statistics contained therein” offered a “half-told” story. Speaking to a prominent philanthropic audience sympathetic to the work of child saving, Ames made a calculation, finding that it cost about two dollars to “rescue one child from the evil influences of intemperance and cruel neglect,” and that the total annual expenditures of the Society amounted to less than “the cost to the community of trying one murderer.” Through this comparison, Ames provided a speculative calculus, a grammar,^[1]Erica R. Edwards, reading Hortense Spillers, defines a grammar as organizing and communicating knowledge, a “‘dynamic of naming and valuation,’ or a treasure chest loaded with the inventions … Continue reading that made it “impossible to overestimate the importance of the work of rescuing children from lives of misery and neglect.”^[2]Pennsylvania Society to Protect Children from Cruelty, “Sixth Annual Report of the Pennsylvania Society to Protect Children from Cruelty: Covering the Operations for the Year 1882,” … Continue reading Exemplifying an early paradigm of risk prediction, Ames tapped into the public’s emotions while relying on the unique ability of administrative data to both stand in for capital, and violate philosophical principles of the individual, in presupposing an unrealized, predetermined future based on others’ actions.

In the child welfare, or family policing, system today, data collection, analysis, and risk-prediction algorithms play an outsized role in structuring caseworker–client relationships, and practice and policy decisions. These algorithms are most frequently deployed in the referral and investigation process, condensing and synthesizing administrative data—data produced from government services—across a variety of departments. Although marketed as an aide for reporters and investigators tasked with making high-stakes decisions in a limited amount of time, risk-prediction algorithms frequently come under public scrutiny for being opaque, or “black boxes,” and making discriminatory decisions due to their data disproportionately containing information on Black, lower class, and generally minoritized communities. In practice, disproportionate representation in datasets produced by discriminatory policies and practices creates a feedback loop that reproduces the same discrimination through new, purportedly objective, measures. These information systems’ foundational truth is that administrative data is capable of representing people and events.^[3]Ezekiel Dixon-Román, “Toward a Hauntology on Data: On the Sociopolitical Forces of Data Assemblages,” Research in Education 98, no. 1 (2017): 44–58, https://doi.org/10.1177/0034523717723387; … Continue reading

But, as Reverend Ames demonstrates, predicting risk and using data to represent a person, event, or place in lieu of actual presence and actions is not a new phenomenon. Critical analysis of data-driven structures today has frequently privileged the contemporary context of risk prediction, resulting in a computationally specific diagnosis that reifies the algorithm’s potential through calls for accountability and transparency, but which fails to attend to the foundational contradictions of administrative data. As J. Khadijah Abdurahman argues, “[d]emands for transparency…elide holistic analysis of how governance is inextricably linked to policing and fundamentally desires to ‘produce [the] truth of the social world.’”^[4]J. Khadijah Abdurahman, “Calculating the Souls of Black Folk: Predictive Analytics in the New York City Administration for Children’s Services,” Columbia Journal of Race and Law Forum 11, no. 4 … Continue reading Further, due to the racializing and classed nature of surveillance, enforcement, and child abuse and neglect definitions, advocating for representative, non-disproportionate datasets threatens to produce an interpretation of results that fails to acknowledge that the outcome variables these algorithms predict on will only ever be “found” in racialized and differentialized people and spaces. In effect, this threatens to verify discrimination as empirically validated and justified through a self-referential equation. As R. Joshua Scannell argues with predictive policing, “Crime does not exist prior to policing. Policing produces crime…Policing does not have a ‘racist history.’ Policing makes race and is inextricable from it.”^[5]R. Joshua Scannell, “This Is Not Minority Report: Predictive Policing And Population Racism,” in Captivating Technology: Race, Carceral Technoscience, and Liberatory Imagination in Everyday Life, … Continue reading

I looked to investigate these contradictions through a historical analysis of administrative data. The field’s definition of administrative data as data produced from service delivery, and as a product of computational enhancements and policy changes in the 1990s, enabled a political and strategic forgetting of “the archive.” Importantly, this forgetting failed to acknowledge the fact that computational media are simulations of physical precedents.^[6]Friedrich A. Kittler, Gramophone, Film, Typewriter, trans. Geoffrey Winthrop-Young and Michael Wutz (Redwood City, CA: Stanford University Press, 1999), xl; Lev Manovich, Software Takes Command (New … Continue reading What my reading here offers is a demonstration of how administrative data, the information powering data-driven governance today, is unique in its ability to simultaneously provide a valuation and represent children and families’ social and cultural variables, through the cost and labor of services that produced the data collected. To contest the grounding assumptions for how we understand data and documentation today,^[7]Lisa Gitelman, Paper Knowledge: Toward a Media History of Documents (Durham, NC: Duke University Press, 2014), 12. I turned to early child welfare annual report archives from 1877 to 1923 to better understand the precedents and historical trajectory of administrative data.

Societies to Protect Children from Cruelty

The contemporary US public child welfare system grew out of a private, philanthropic community that emerged in the nineteenth century. The organization my research primarily focuses on, Societies to Prevent Cruelty to Children (SPCCs), arose in the 1870s in response to economic downturns, the isolation of poor people in the cities, the spread of organized labor, increased immigration, and concerns around almsgiving to the undeserving poor.^[8]For the emergence of SPCCs, see Linda Gordon, Heroes of Their Own Lives : The Politics and History of Family Violence, Boston, 1880–1960 (Urbana and Chicago, IL: University of Illinois Press, … Continue reading While placement organizations like Charles Loring Brace’s Children’s Aid Society^[9]One of the organizations that created and facilitated the Orphan Train Movement. had existed for decades, and there were legal precedents of adults being prosecuted for child abuse and neglect, no organization conducted active public surveillance and monitoring of children and families. These SPCCs took a new approach to child welfare, taking on investigatory roles, and receiving and providing referrals for placements and services. SPCCs not only resemble how the child welfare system operates today, but in some cases, like what I discovered in the Pennsylvania SPCC archives, are the same organizations conducting case management services.^[10]Following a number of mergers and shifts in responsibilities, in Philadelphia, the SPCC continues to operate as Turning Points for Children. Through 2023, Turning Points was the city’s largest case … Continue reading This makes the child welfare system fairly unique as it has maintained a semiprivate, functionally “neoliberal,” character since its initial emergence.

The SPCC Annual Report Archives

To understand the origins of administrative data in the child welfare system, I went to the annual reports of the Pennsylvania SPCC, one of the oldest in the country. I sought to understand how the Pennsylvania SPCC was utilizing these reports, and the narratives and data contained in them, as a platform, or a technology for mediating and disseminating information to a reader. As a new organization and intervention, the SPCC’s annual reports became a crucial technology where officers could communicate with the public, and define the terms by which child-saving work could be measured.^[11]I found thinking about these annual reports as a predecessor to the dashboard to be a useful metaphor, see Shannon Mattern, “A City Is Not a Computer: Other Urban Intelligences (Princeton, NJ: … Continue reading Further, the importance of the reports was amplified by the fact that SPCCs relied on a philanthropic public to fund its para-police activities, through a perforated donation sheet in their opening pages, and requiring the community to report families to the office.^[12]Founded in 1877, the Pennsylvania SPCC began receiving some public funding in 1885. The Pennsylvania SPCC self-describes as being an “auxiliary to the police force.” Pennsylvania Society to … Continue reading Throughout the years the annual reports detail a range of mechanisms for receiving reports and triggering investigations, including children showing up on their own accord, children being brought to the office, referrals from and collaboration with police, patrolling, and having neighbors, and witnesses contact the SPCC (in person, by letter, or phone) with complaints. The annual report, for the Society, prefigured platform capitalism and its mediation of capital, labor, and surveillance.^[13]Seb Franklin, “Reproduction at the Interface,” Representations 165, no. 1 (Winter 2024): 69, https://doi.org/10.1525/rep.2024.165.3.63.

The reports themselves shifted their strategies of representation over time in response to innovation. Initially centering narratives and reproductions of case notes, they moved to incorporate photographs of children in the mid-1890s. By the latter half of the first decade of the 1900s, reports centered statistical tables, averages, and aggregated information. The Society’s record system included information on the nature of complaints; child’s name, age, and placement; defendants; complainants; witnesses; and case notes,^[14]Zanti et al., “Examining Early Power Dynamics,” 305. where investigations were largely interview based.^[15]Michael B. Kahan, “Jewish Girls’ Street Peddling in Gilded Age Philadelphia: Ethnic Niche, Family Strategy, and Sexual Danger,” Journal of the History of Childhood and Youth 12, no. 3 (Fall … Continue reading The administrative data produced from investigations was then used both for recordkeeping and for reproduction in the annual report, which was distributed to organization members and the broader philanthropic community.

Data as Commodity

The Society was explicit in its identification of administrative data as a valuable commodity. In discussing their data, they routinely infused it with capitalist signifiers and processes of production. Officers described statistical reports as “of our labors” that would serve as a form of repayment “for the time spent in its perusal,” contributors’ money, and labor.^[16]This discourse on data demonstrates Jonathan Beller’s rewriting of the M-C-M’ formula as M-I-M’, arguing that information and commodity are synonymous. Jonathan Beller, The World Computer: … Continue reading The data contained within the annual reports was explicitly curated, produced for, and with a reading donor in mind. In their 1887 Report of the Board of Managers, the president and secretary state that they saturate the report with more brutal cases, irrespective of the fact that those cases are not as representative as their policing of neglect:^[17]Pennsylvania Society to Protect Children from Cruelty, “The Pennsylvania Society to Protect Children from Cruelty Eleventh Annual Report: Covering the Operations for the Year 1887,” … Continue reading

We have reason to believe that the community becomes more excited and moved to sustain our work by the report of a single case of physical cruelty, than by the knowledge of the rescue of a score of children from a life where they undergo more suffering, but of a kind that leaves no marks upon the surface…We wish it were not so…for then it would be much easier to create and maintain a public interest in our work, and a more generous support.

For the Society’s officers, the affective and material stimulus they anticipate will be produced from data on physical punishment is more important than an accurate depiction of the phenomena, even when cases of neglect led to “more suffering, but of a kind that leaves no marks.” Therefore, they are willing to exploit particularly brutal cases of abuse by saturating their reports with case file and photographic reproductions of their observations to prompt more donations. This occurs at the expense of distorting the nature and scope of their work, and through a logic that renders data and capital exchangeable. The overrepresentation of abuse also allows the Society to pathologize the parents of children they deemed to be neglectful and abusive. This move, using words such as “brutal,” “depraved,” and “ancient,” to situate these families and communities as outside of modernity, allowed the SPCC to intervene in ways otherwise unacceptable, removing children from their homes and placing them with other families and institutions. The justification for this intervention was found through the images of child welfare work the Society curated, necessitating removal and facilitating capital accumulation in the name of protection and prevention.^[18]Denise Ferreira Da Silva, “The Racial Calculus: Security and Policy During the COVID-19 Global Pandemic,” in Ordering the Human: The Global Spread of Racial Science, ed. Dorothy E. Roberts, Eram … Continue reading

For the SPCC, administrative data is an output of labor, and a commodity to be consumed by a reader. Labor was transformed into information, and the volume and type of information justified and funded their labor. This information was then packaged in reports and functionally “sold” as part of a call for donations.

Data as Derivatives

As demonstrated through Reverend Charles E. Ames’ opening remarks, on the flipside of this chain of equivalences is the child and family being investigated. Ames’ formula imposes a numerical value on the child, advocating for a preventative approach that would see the child removed from their home in hopes that this would prevent future, more expensive, criminal behavior. Through Ames’ statement, the Society projects an image of the neglected child as possessing a dormant criminality that justifies preemptive child welfare intervention on the basis of savings to the taxpayer. The value of the child is equivalent to the cost of services they received. Speaking on race, ethics, and value, J. Reid Miller asks, “if what is determined by race is by definition what is not self-determined, how can it ethically reflect the subject at all?”^[19] J. Reid Miller, Stain Removal: Ethics and Race (New York: Oxford University Press, 2017), 59. Here, administrative data, operating similarly to racialization, represents a derivative, a position on the future that offloads risk onto those in precarious situations.^[20]Beller, The World Computer. Much like today, albeit through discourse rather than number, the child becomes constructed as a formula of social, cultural, and economic variables toward understanding probabilities for future, risky behavior.

When looking at this calculus as a whole, without taking away from the particularities of data-driven technologies, we find a chain of equivalences that demonstrates the conceptual foundation of modern governance. Labor and information become each other’s condition of possibility. Because their information was a representation of labor, the information, or administrative data, could be easily quantified. Further, the information itself represented children and families, and could be evaluative of future outcomes. In the case of the administrative data from the annual report Ames was reading, this meant that, financially, it was worth the risk to intervene “preventatively” through the child welfare system to save the criminal justice system money, at the risk of the preventative intervention being unnecessary. For the child and family, this derivative calculus forecloses the possibilities of their future, rendering them risky, traumatized subjects^[21]For research on how discourses of resilience and trauma vacate the subject of agency and obfuscate systemic violence, see Yoosun Park, Rory Crath, and Donna Jeffery, “Disciplining the Risky … Continue reading beholden to one of two paths: family separation or criminality. Combined with the SPCC’s semiprivate nature, this positioned their administrative data as the key tool to attracting investment through emotional data presented to donors and by placing a value on services provided that offered taxpayers a potential savings in the future through an assumption that, at least some, abused and neglected children would inevitably become murderers.

Looking Back to Look Forward

Administrative data has emerged as a key site and calcification of power that facilitates how the state’s surveillance flattens a person’s life to “the case.”^[22]See Virginia Eubanks, Automating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor (New York: St. Martin’s Press, 2019); Abdurahman, “Calculating The Souls of Black Folk”; … Continue reading By looking back, the case of the SPCC troubles the naturalization of computation^[23]Beller, The World Computer, 9–10, 19. and the “newness” of data by challenging the assumptions of an origin point, and the historical and theoretical inheritances of its grounding system.^[24]For recursive epistemologies and methodologies of historicization, see Bedour Alagraa, “The Underlife of the Dialectic: Sylvia Wynter on Autopoeisis and Epistemic Rupture,” Political Theory 51, … Continue reading The process of data construction is laden with power differentials, cultural and historical context, and other interests and affects that manifest within the data.^[25]For theories of data assemblage, see Dixon-Román, “Toward a Hauntology on Data.” Also relevant here is Alexander Galloway’s work on the interface, see Alexander R. Galloway, The Interface … Continue reading As Denise Ferreira da Silva lays out, algorithms have inputs, and these inputs, or data, are explicitly and preemptively constructed with the form and purpose of matching the algorithm,^[26]Denise Ferreira da Silva, foreword to All Incomplete, by Stefano Harney and Fred Moten (Minor Compositions, 2021), 5. shattering the assumption that data is, or can ever be, raw material. In the context of the SPCC, we see how labor, the readership, and financial concerns inform the data disseminated and interpreted rather than vice versa. These ground assumptions of data are a crucial site of contestation as we continue to jockey and battle the information asymmetry^[27][Digital] interfaces render people visible in different ways, according to conflicting interests and alignments, see Mateescu and Ticona on “visibility regimes,” Alexandra Mateescu and Julia … Continue reading propelling our platforms and worlds forward.

Footnotes