Programs are actor networks
The ontology of the MTO program
Note: This is the final post in a reading series about the Moving to Opportunity evaluation. I describe the evaluation and praise its positive aspects here. I critique its evaluation question here and its methodology here. You don’t need to read all of these posts to understand the current post, but reading the first link will help you understand the program I’m talking about.
Sometimes we talk about programs like they are nonhumans with stable characteristics. They are like pills that, once taken, either work or fail depending on the characteristics of the patients (participants) and the nature of the disease (social problem) with which they are afflicted.
Sometimes we talk about programs like they are people. They can do things. They have functions and vested interests. They adapt to changing conditions and try to survive.
We almost never talk about programs as networks of humans in alliance with nonhumans, rhizomatically connected to each other, performatively re-enacting themselves. This is a fairly new way of looking at social structures like programs.
If we could look at programs in all three ways at once: as nonhuman technology, as human actors, and as networks, we would see them as actor-networks.
Actor-network theory (ANT) is the name for the social science perspective that sees the world this way. ANT has roots in the French sociology of science of Bruno Latour and Michel Callon, the early work of Donna Haraway, and Michel Serre’s philosophy of science. The ontology of ANT is, I argue, the biggest breakthrough in social science theory in the last several decades. Why is this?
For most of the 19th and 20th centuries, social science in general has been grappling with the apparent paradox of structure (of society) and agency (of individuals), mainly due to the influence of thinkers like Giddens, Bourdieu, and Foucault. The paradox haunted all social scientists in some way: how can individuals have agency if social structures so clearly shape life outcomes? In the later 20th century, social scientists split basically three ways on the answer. The traditionalists are apparently still mystified by the paradox, the postmodernists deconstructed the agentive self, and the actor-network theorists rethought the whole problem from the ground up. Along with other ANT thinkers, I argue that, if we are going to break up the idea of the agentive self we should also break up the idea of cyclopean social structures. For example, rather than thinking of “the State” as a thing that just eternally exists and has stable interests, ANT thinkers would say that it makes more sense to see it as a network of players that is constantly reshaping itself to perform statehood. These players include groups of humans, such as the citizens, noncitizens, agencies, and representative bodies that compose the state. The players also include nonhumans, like the land, the borders, and the interstate highways. This new ontology gives us room to resolve the structure-agency paradox by pointing out that some configurations of the State actor-network afford more agency to particular players – including the nonhumans – while constraining the agency of other players. Moreover, it turns out that what we mean by agency is much clearer now as well, since agency is our ability to make things happen in the network.1
I’ve observed that some of the lessons of ANT have trickled into program evaluation theory, but when they arrive they tend to take a more social constructivist shape. ANT thinkers critique mainstream social constructivism on the grounds that it is too idealist (in that it usually ignores materiality) and on the grounds that it is anthropocentric (in that it ignores nonhuman actors). For example, a social constructivist account of the linked mental health and homelessness crisis in California would emphasize that we have socially constructed certain behaviors as “insane” and that this leads to people with mental illness being ostracized from social groups. An ANT perspective would emphasize that there are biopsychosocial processes going on in people’s minds, including as a result of the substances people consume, as well as cyclical effects from the risks of living unsheltered, including the risk of further trauma and brain injury, and point out that the crisis is a matter of shifting networks of medical care, housing, and money. When we focus almost entirely on people and their mental states, as social constructivism does, we miss major facts about how systems work and how they change. Assimilating ANT insights into a generally social constructivist view is a result of not understanding the big differences between these perspectives.2
Fully letting ANT in through the front door of program evaluation theory will mean that we start explicitly paying attention to more of the nonhuman actors on the scene. Money is usually one of those nonhuman actors. So are physical spaces, medicines, transit, phones, policies, rubrics, and measurement instruments. We also need to pay attention to the ways that humans and nonhumans are contingently related to each other as part of the program. Programs are more than people doing actions. They are people and nonhumans assembling networks with emergent behavior. The relations between actors are heterogenous, to use one of John Law’s favorite words, and we can’t assume that the way the program works right here is the way that it works over there. If we engage in rhetorical simplification of the actor-network to a single actor, we need to know that that we are paying some price for this simplification in terms of understanding how the network actually accomplishes things.
The ontology of Moving to Opportunity
How did the final evaluation of the Moving to Opportunity program write about its ontology? On the one hand, this is an unfair question for most evaluation, since ontology is tacitly assumed and rarely discussed. On the other hand, however, ontology is critical for certain decisions in evaluation design. For a randomized controlled trial, we need to assert that the treatment was the same across all the subjects in that arm, or at least that differences in treatment were so small as to be ignorable. Violations of this precondition for the design are usually treated as implementation failures. There is another ontological requirement that the subjects in different treatment conditions don’t somehow cross into another treatment arm. If you are assigned the “placebo” condition in the study, this requirement says that there is a problem if you are accidentally given the medication we are trying to test instead of a sugar pill. As a randomized controlled trial with two treatment arms, the MTO evaluation needed to be able to argue that the program possessed certain characteristics like these.
Here is what the final evaluation says about the program itself:
The experimental group received Section 8 rental assistance certificates or vouchers that they could use only in census tracts with 1990 poverty rates below 10 percent. In each city, a nonprofit organization under contract to the PHA provided mobility counseling to help families locate and lease suitable housing in a low-poverty area. After one year, families were able to use their voucher to relocate without any special MTO-imposed constraints on their moves. Families who stayed in their new low-poverty neighborhoods less than a year did not receive a new voucher. Aside from this requirement, experimental group families were required to abide by all of the regular rules and requirements of the Section 8 certificate and voucher programs, including having a limited amount of time to search for housing and sign a lease before they lost the rights to their subsidy, being required to contribute 30 percent of their adjusted income toward rent (the same rent requirement as in public housing), and prohibitions on rental assistance to households that engage in certain types of criminal activity.
The Section 8 only group received regular Section 8 certificates or vouchers that they could use anywhere; these families received no mobility counseling.
There is a similar, shorter, description of the program in the Executive Summary, as well as some measures of implementation fidelity such as the proportion of participants who accepted the voucher. The above passage is the most detailed description of what the program actually is and what it does.
This is not because MTO was a simple program to implement. Many program staff were involved. Some of them brokered relationships with landlords in low-poverty areas, while others provided the “mobility counseling” mentioned above. In what did this counseling consist? It’s impossible to say from the evaluation. How did the participants react to the intervention? We have outcomes metrics and some general statements about how not all participants stayed in their assigned neighborhoods, but that is all.
Comparing the final evaluation report with the book written by qualitative researchers who separately studied MTO is at times surprising. Let’s consider three wrinkles in the program implementation.
First, we learn from the ethnographic study that some of the MTO sites barely got off the ground because of the housing nonprofits subcontracted by HUD to implement the program:
While nonprofits have a long record of involvement in providing social services, coaching prospective homeowners, and developing and managing low-income rental housing in the United States, few had any experience counseling public housing tenants on a move to “opportunity.” Nor did they have experience coordinating with the public housing agencies, which signaled some resentment that HUD had required them to contract out for any portion of a housing voucher demonstration. Add to that the fact that two of the nonprofits that came forward to be part of the MTO program were fair-housing advocacy organizations with a history of suing local landlords and public housing agencies—the very players whose cooperation was essential in MTO. In Chicago and Los Angeles, these two nonprofits were ineffective and had to be replaced.
If we think about the program as a nonhuman object with stable characteristics, these details are not a big deal. If we think about the program as an actor-network, these details give us pause. The program includes landlords and nonprofits as part of its network. When these actors don’t gel… no program!
Second, consider another bit of “heterogeneity” we learn about from the ethnographic study, the active political campaign in Baltimore City to block program implementation.
After the official HUD announcement of the MTO agreement with the Housing Authority of Baltimore City in March 1994, a small number of county residents joined a new group, the Eastern Political Association (EPA), which decided to openly oppose MTO. EPA’s stance was heavily influenced by media coverage that linked the impending demolition of some of Baltimore’s worst public housing, through the HOPE VI redevelopment program, with the arrival of MTO and the wider sense of decline and vulnerability in eastern Baltimore County communities.
EPA sponsored a number of community meetings with the purpose of distributing information against MTO. Central to this effort was their belief that the small numbers of MTO families—who would be widely dispersed across a large housing market—were not the only public housing residents headed their way. “Many residents see MTO as the first step in a government plan to tear down the city’s housing projects and move the residents to eastern Baltimore County,” reported the Baltimore Sun. Residents saw the claim that MTO would only fund vouchers for 145 families as a “smoke screen” concealing government plans to move in all 18,000 families in public housing.
When the dust cleared, the second year of funding that had been appropriated for MTO had been blocked. It was an election year and virtually every candidate in Baltimore County expressed opposition to the program. The program did eventually roll out, but in a limited fashion. The community, particularly of property owners, had become suspicious of the program and its participants.
In the language of ANT, the program (simplified to a node) didn’t fit into the actor-network of Baltimore County in the way that the body rejects a transplanted organ. Where HUD had hoped that the local network would flex to accommodate the program, the network rigidified itself instead. This rigidity was, of course, a performative rigidity, since Baltimore County did not have ways of directly prohibiting the movement of people into its territory in the first place – it is a county and not a gated community. In fact, it turned out that the areas from which many of the program’s opponents had come were actually too poor to be considered as relocation zones for MTO – poor residents had already been “let in” or the existing residents had become poor.
Third, let’s think about the definition of a “low-poverty neighborhood” used by the program – a neighborhood with 10% poverty rate or less. The MTO program was inspired by an earlier desegregation program following the class-action lawsuit Gautreaux v. Chicago Housing Authority (1966). The court-ordered remedy relocated more than 7,000 Black families using vouchers and had had notable positive effects on participants who relocated to suburbs. However, in the 1970’s and 1980’s the Federal government began to opposed race-conscious policies and it was common sense that such initiatives, even if they were funded, could be shot down in the courts later. This led to a strategic shift for desegregation advocates:
As insiders recall, the meeting at HUD began with Polikoff proposing a single-site replication of Gautreaux in Baltimore, but HUD staff believed there might be room to extend the demonstration further, on the condition that income and not race be used to target “opportunity” areas to which families would relocate… HUD relied on income make-up alone as a metric of neighborhood “opportunity.” The legally controversial, race-focused objectives of Gautreaux were replaced with the neighborhood poverty rate threshold.
The focus of the program had shifted from race to class, with class being rounded down to poverty rates. This operationalization led the program down an interesting path: it was designed to increased “opportunity” for participants, particularly in terms of educational and career outcomes, but didn’t necessarily relocate participants to areas with better educational and job opportunities. Most families ended up staying in the same school districts in which they lived originally, despite moving to low-poverty-rate neighborhoods. Employment rates or job openings were not tracked by the evaluation at the neighborhood level, so it is unclear whether participants who moved to low-poverty neighborhoods actually moved to areas with more local jobs. Indeed, since many participants moved from urban areas, which are often zoned for mixed-use, to suburban areas, which are often zoned solely for housing, it is quite possible that they moved to areas with fewer jobs – the design of the study doesn’t tell us this. We do know that, in the end, the program didn’t decrease unemployment.
In other words, regardless of the attempts of the program’s planner to socially construct areas with low poverty rates as “opportunity zones”, a material reality of educational and employment opportunities persisted outside this construction. Nonhumans like zoning laws, physical spaces, and labor markets are part of the program’s actor network and exert influence.
ANT ontology in evaluation
My contention is not merely that ANT is a good “perspective” to adopt, but that programs are indeed actor-networks. Using the MTO study as an example, I am arguing that actor-network theory’s ontology is simply a correct description. Programs are not just stable technologies or actors – programs are assemblages of humans and nonhumans, both of which exert power in networks. Programs are made up of both material connections (such as housing, neighborhoods, and money) as well as conceptual connections (like goals, values, and theories of change). The MTO program shows us how taking a view of programs as stable technologies leaves a lot of very important information outside the frame.
This is my final post in the reading series about the Moving to Opportunity evaluation. If you are interested in another analysis of an evaluation, let me know in the comments or via a direct message. You can also suggest that I look at specific evaluations in your field or area of interest.
Summarizing such a huge amount of theory in such a simple paragraph is an obvious disservice to smarter people than myself, but I’d rather get my hands dirty and try to sum up than just tell you to go read Michel Callon’s classic essay about scallops or the one I just read about ANT and real estate zoning laws in Switzerland. I guess this is my subscriber pitch?
Some ANT theorists mock social constructivism outright. That’s not my intention here – I prefer to think of social constructivists as people who correctly understand part of the picture but need some help understanding the nonhuman world. Many of us who were initially trained in the social sciences probably thought we didn’t need to understand the nonhuman world to do our jobs – it turns out we do need to.