Understanding Contestability on the Margins: Implications for the Design of Algorithmic Decision-making in Public Services

3.1.1 Chennai Metropolitan Area.

The Chennai Metropolitan Area is made up of multiple administrative districts, each divided into blocks. We focused on one block within one of the districts. This block has a population of approximately 211,000 people, and approximately 80% live in rural areas. Three percent of this population is made up of those belonging to Scheduled Tribes, or Indigenous people of India who are recognized as a historically marginalized group and are thus afforded rights to representation in government, education, and other domains [107]. To conduct the study, our access was facilitated by an NGO operating in the area for the past eight years, focused on community development and access to housing in Irula communities, a Scheduled Tribe in Tamil Nadu. Thus, our understanding of the work of availing housing was situated in Irula communities’ experiences. Connection to land and natural resources is an important aspect of Irula and other tribal communities’ identity, culture, and lifeways [107]. Yet tribes have been subject to significant displacement and lack of access to land over the period of British colonization, through caste-based hierarchies in land ownership, and on account of development projects undertaken by the independent Indian state [107]. Scheduled Tribes thus face high rates of poverty and landlessness [107], making access to government schemes for land and home ownership important.

In studying access to government housing schemes, we found that the application process took years. The most salient aspect at the time of our study was applying for a free patta, or land deed, that is required before being able to construct a house at the site. One can apply for a patta where one has been living, or in a different area identified by community members or government officials. There were multiple stakeholders involved in this process. One was the applicants themselves who were seeking a patta. Another was coordinators and a lawyer who worked with the NGO and supported applicants with pattas and other required documents. Another was the appointed district administration and local elected leaders who process and approve patta applications. Within appointed district administration, this includes the district collector at the highest level, the revenue divisional officer, tehsildar, and revenue inspector at the block level, and the village administrative officer at the village level. Within the elected Gram Panchayat, or village council, patta applications must also be approved by the panchayat leader.

3.1.2 Boston Metropolitan Area.

The Boston Metropolitan Area is composed of dozens of cities and towns, of which we focused on two cities. Combined, they have a population of around 120,000 people. The cities’ demographics reflect significant immigrant communities and racial diversity, composed of 25% to 44% immigrant populations, and 5% to 8% Black, 4% to 10% Asian American/Pacific Islander, and 12% to 66% Hispanic populations. In the past several years, due to gentrification, rising cost of homeownership, and lack of enough investment in subsidized housing, lower-income households have been increasingly subject to housing instability and displacement. Further, those who are chronically homeless are unable to access the permanent supportive housing needed to exit homelessness. These issues disproportionately affect marginalized communities such as undocumented immigrants who are more vulnerable to tenant rights violations. They also affect Black, Indigenous, and Latinx communities who are subject to higher rates of poverty, housing discrimination, and incarceration in Massachusetts [2, 18].

In this context, we studied the processes of applying for public housing and rental vouchers, as well as for priority status for such services. We also looked at cases where people already had public housing or a voucher and whose benefits were being terminated by the housing authority. Finally, we also looked at cases where people were applying to inclusionary housing (a percentage of private developments that is required to be affordable) or using a rental voucher to rent on the private market. Stakeholders in this context include applicants who are seeking or trying to maintain housing assistance. They also include case managers who support people in applying for assistance, appealing denials or terminations, and/or accessing other supportive services that contribute to housing stability. Lawyers who are part of legal advocacy organizations can also become involved if people contact them when choosing to appeal adverse outcomes such as application denials or terminations. Finally, a housing authority for a given city processes applications, maintains public housing properties, and presides over appeals. Workers supporting these functions include tenant selectors who maintain waitlists for housing and process applications, higher-level workers who conduct criminal records look-ups, and hearing officers who preside over appeals.

3.1.3 Motivation for Context and Comparison.

We chose to focus on land and housing because they are high-stakes decisions. Receiving these benefits can be a matter of safety, physical and mental health, and stability. This is a case where it is especially important to design for contestability, as a way to ensure due process and that these services can actually meet their intended goals, rather than reifying historical exclusions. We acknowledge that there are significant differences in each context, including the specific benefits and application processes studied. However, in both cases, because the services we focused on are intended to serve marginalized populations, we are able to look at how accessing these services brings people in contact with the state, requiring them to navigate significant power differentials and control over distribution of resources.

3.2.1 Chennai Metropolitan Area.

Naveena collected data in July 2023. She conducted three interviews with the NGO’s coordinators and legal advocate, and six interviews with district and village administration. She also visited three tribal villages, where she conducted group interviews with three to ten people each. Participants are summarized in Table 1. Many of these meetings were largely facilitated by the coordinators working for the partner NGO, who were from Irula communities themselves and had been working with the NGO for three to eight years. Naveena reached out to the district collector directly, and contacted a panchayat leader through a chain of referrals from the partner NGO.

Interview questions with applicants focused on experiences with applying for pattas and other benefits and deciding on next steps when responding to adverse outcomes. Questions asked here included "What was the application process like? What were the most challenging or surprising parts?", "What was your initial reaction when you learned the decision? How did you decide what to do next?", and "Can you walk me through your appeals process? What was the most challenging part?" Interviews with the coordinators and lawyer focused on how they supported people in applying and following up on applications and adverse decisions. Questions included "When do you decide to appeal, reapply, etc.?", "How do you support applicants through this process?", and "How do you decide what you need for a successful appeal, reapplication, etc.?" Interviews with government officials focused on experiences with processing and providing decisions on patta applications. Questions included "What are the range of reasons for denials?" and "How do applicants respond when they get denied? What interactions do you have with them after denial?" Interviews were between 30 and 60 minutes, and took place in person. The interviews were in Tamil, Telugu, and English; Naveena speaks Telugu and English and relied on an NGO worker to translate for interactions in Tamil. We did not record interviews as we found that participants were more comfortable. Instead, we took extensive handwritten notes, taking the time to note down quotes if they were particularly illustrative, and wrote them up in further detail soon after the interaction.

Applicants were composed of 18 women, and two men. Pattas and homes are provided in women’s names, and they were important stakeholders in the process. Additionally, we visited villages around noon, when women were present, often doing unpaid care work, while men were elsewhere for paid work. With respect to caste location, other than the applicants and coordinators who were from Irula communities, all other participants we spoke with were dominant caste. We note that there are multiple terms that can be used to describe tribal communities in India; in this paper, we use the terms that applicants used to refer to their community, which was either Irula or Scheduled Tribe people, sometimes shortened to ST.

3.2.2 Boston Metropolitan Area.

Naveena and Sohini collected data between April and June 2023. We conducted 15 interviews with the various stakeholders: six applicants, two case managers, three lawyers, and four housing authority staff across two cities. Participants are summarized in Table 2. Of the six participants using housing assistance, there were three Latina, one Black, and two White participants. Participants consisted of five women and one man. All other participants were White, except for one case manager who was Latina and one housing authority staff whom we do not have demographic data for. We recruited participants through snowball sampling, starting with lawyers and case managers, who were then able to help recruit people who had experience applying for housing assistance. We also reached out to people using housing assistance by flyering near a housing authority office and affordable housing. We also contacted housing authority staff directly.

Interview questions were similar to above. We focused on experience with the application and appeals processes (if they had gone through it), the forms of labor involved in decisions, denials, and appeals, the nature and extent of interactions between stakeholders, and pain points in the application and appeals process. Interviews were between 60 and 90 minutes and took place largely virtually via phone or Zoom, with a few taking place in person. All interviews took place in English, except for one which took place in Spanish, using a human translator through a phone-based service.

4.1.1 Enabling Access to Inaccessible Public Services.

In both the Indian and US contexts, intermediaries helped overcome many of the structural inequities in navigating public services. In the Indian context, applying for a patta first required several documents and IDs which families needed to apply for if they did not already have them. After gathering these materials and applying for a patta, families may also need utilities such as electricity and water to be set up by the local government. The coordinators had gathered a rich understanding of the barriers families faced in applying for pattas and relevant documents, demonstrating the epistemic humility required for accompaniment. It was challenging for families to prioritize the tedious work of applying when the need was not immediately obvious (given that the final outcome of a home was further downstream). Applying required families to take time off from work to coordinate transportation, travel long distances from one’s village to government offices, and pay to make copies of their documents, resulting in them losing income. Further, government officials may claim that applications were never received, so multiple follow-ups might be needed to ensure they received attention (as we will describe in more depth below). Acknowledging such challenges allowed the coordinators to share in the labor required to interface with public services. Coordinators helped put together applications to reduce the burden on families of learning about the opaque process themselves. Coordinators would also tell families when their presence was required in person at government offices, taking on the more persistent (and tedious) work of following up on applications themselves.

Coordinators were also able to leverage their knowledge and relationships with government officials to support applicants in avoiding denials. Coordinators worked to build relationships with new government officials as they entered office, which helped them understand how much they could be relied upon. C22 explained how he makes a point to introduce himself and explain the work that he does for Scheduled Tribe communities. Leveraging these connections, coordinators were able to learn about the intricacies of application processes from government officials, like what to do in exceptional cases such as applying for a patta for land that was already owned by others who had since passed away. In comparison, we heard from applicants who used to go individually to the revenue division officer to apply for pattas but were not able to learn anything about the proper procedures even after several attempts, indicating the persistent opacity of public services for marginalized communities.

In the US context, accompaniment was similarly essential for navigating options for housing assistance amidst social and economic constraints. B2 noted how finding resources was emotionally and mentally taxing when experiencing housing instability: “A lot of people are burdened or suffering from trauma or overwhelmed, you know, they don’t know what to do. And so then what happens?” Case managers could be extremely supportive in these cases by doing the open-ended work of helping navigate options to eventually find stable housing, and ensuring clients are prepared for each step of application processes. This was especially important as there was generally a 10-day limit to respond to communications from the housing authority. Such timelines help the housing authority close cases and ensure valuable housing units do not stay empty, but made it challenging for applicants who may not know they need to keep track of communications and be ready to respond. Again, open channels of communication between housing authorities and local NGOs were helpful in ensuring case managers could set clients up for success.

Similar to the Indian context, we saw the extent to which case managers uniquely enabled actions that clients may not be in a position to take otherwise. For example, B1 was denied priority status by the housing authority and successfully appealed the decision. Much of her narrative did not mention any of the reasoning behind the denial or basis for the appeal, despite having a case manager who was able to explain what was happening. When asked if she knew in the end why she had been denied, she said "No, you know, I, I still don’t understand actually why we were denied." Rather, she shared much more about the difficult emotional experience of being pushed out of her current housing, such as her surprise and fear upon finding out about the complaints tenants had about her young son. Ultimately, the fact that she had a case manager who was not only able to understand her denial, but also guide her through the appeals process, contributed significantly to her ability to contest.

Applicants could certainly learn about the application or appeals process by going to the housing authority directly, demonstrating how the work of accompaniment could also happen within public services. Given the smaller power differential between applicants and housing authority staff compared to the Indian context, applicants seemed to have relatively more success. However, there were sometimes barriers here as well, and case managers were also helpful for equitable access to knowledge. For example, B8 shared how the housing authority staff screening one client for state public housing was asking for information that, according to the state regulations, was not actually required—this type of double-checking could be very impactful for undocumented immigrants who may not be able to produce certain documentation and would otherwise receive a denial. Ultimately, it seemed that the level of intensive support needed was not always possible for the housing authority to provide, leading to some appeals happening simply to “...sit there and explain the process exactly of what is needed [for an application]...,” according to B15.

4.1.2 Overcoming Apprehensions around Contestation.

The work of accompaniment also entailed solidarity and support in addressing apprehensions in interfacing with public services, which could be a barrier to contestation. In the Indian context, many women shared how they were hesitant to meet individually with government officials to advocate for their requested pattas, affirming prior work [43]. A few women in one village shared the sentiment that “...unlike people who have studied, we can’t speak well or courageously.” We see this as a sign of the power differential between government officials and Irula communities, where government officials can choose whom they listen to or not, particularly if they are dominant caste. The coordinators addressed community members’ apprehensions by physically accompanying applicants to meetings. They also relied on collective strategies such as letters to government officials from multiple families seeking pattas or other benefits, when needed. However, there was still fear of retaliation given the power that officials had over people’s lives. For example, residents in one village explained how a panchayat leader deprioritized them for government housing as retaliation for not supporting him in local elections.

Despite these apprehensions, we found that applicants in the Indian context were ultimately still willing to contest decisions, particularly using collective methods. In contrast, in the US context, apprehensions manifested more as an unwillingness to contest at all. People facing language barriers, immigrants, and those with cognitive impairments may view appeals as risky or not understand the value of going through with them. B1, who immigrated to the US from El Salvador, explained why she did not want to go through with an appeal in her case: “It’s not what I do, like to stir up trouble... because I’m an immigrant... if they can help me, great. But if not, I respect that... I’m very grateful that I’m here and I’m being allowed to live here.” Again, we found that accompaniment was key to addressing discomfort here. B7, a case manager, shared how in addition to communicating in the client’s language, she built trust and pulled in social support:

“Once we build this trusting relationship with clients, they tend to believe you. [We] explain to them in detail using lay language... what are the options, because if we don’t appeal, we don’t have a house... And the other thing... is that there is someone that the client trusts at home or a friend... it’s the case that the friend understands better the big picture than the client. And it’s the friend that also works for us in a sense of explaining to the client, you know, what the benefits might be if you appeal.”

We note, however, that being convinced in one instance may not actually change clients’ perspective on contestation more broadly. When asked if she would go through with this process next time, B1 insisted that she would not, because “I don’t wanna bother anybody... I’m just very grateful that I’m here [in the US].”

We see how in the Indian context, hesitancy came from the possibility of not being heard, but applicants had the conviction to act and ensure their access to benefits. In the US, especially for immigrant populations, apprehension stemmed from not wanting to create “trouble” and demonstrating gratefulness. This could be traced to the differences in relationship to the state—imaginaries of opportunity in a new place may shape notions of risk differently compared to histories of displacement and the desire to preserve community. Further, while the participants in the US context had the right to appeal, in India, we see the possibility of more retaliation due to the power differential between government officials and Scheduled Tribe communities, which had implications for methods of contestation and capacity-building efforts.

4.1.3 Challenges in Connecting to Intermediaries.

Intermediaries were important in enabling access and action, but a significant difference across contexts was how intermediaries were able to get in touch with applicants who need support. In the Indian context, the coordinators had been working for the past year with specific Scheduled Tribe villages, going door to door to survey which family members had which documents or benefits, and whether the family had a patta, a permanent house, and various amenities and utilities. Given the rural setting and distinct boundary of villages, coordinators were able to be systematic in understanding where support was needed. Coordinators were from these communities themselves, which helped establish trust and long-term relationships.

In contrast, participants in the US context noted that there were challenges in connecting to applicants when they most needed support with accessing resources or appealing denials, possibly due to population size and heterogeneity. For denials from the housing authority, information about the right to appeal and about legal services were shared along with denial and termination letters in an attempt to strengthen connections, but this did not necessarily mean uptake. To strengthen timely outreach to people in precarious situations, case managers used strategies like reaching out through school liaisons, or reaching out to people who are currently involved in housing court. Lawyers we talked to also shared that they relied on applicants being referred to them by other legal agencies, and were experimenting with pop-up clinics to further the reach and type of support they offered.

4.2.1 Holding Decision-making Processes Accountable.

As a senior leader at the partner NGO told us, “There is no such thing as just applying for a patta and getting it.” Officials asking for bribes to even acknowledge an application was rampant, especially if families tried to apply for documents or pattas by themselves. We also heard from participants that revenue division officers, who receive numerous applications, would simply forget about applications if they were not prompted. There were strategies that the coordinators used to make it difficult for government officers to claim they never received an application. As mentioned in the previous subsection, the coordinators suggested that families make copies of applications before submitting so that they could produce that copy later. The coordinators were also careful to date the application according to the day it was submitted. During any follow up meetings, coordinators would ask the government officer to sign the application copy on the date of the follow up, so as to have proof that the coordinator had followed up. Coordinators might have multiple follow up meetings every week or two. However, there might be cases where government officials are persistently unresponsive. In this case, coordinators could use a Right to Information request [1] to understand where the application was, and ensured applicants were aware of this right as well. As one NGO worker indicated, accompaniment required innovation: “The NGO has changed how follow ups are done. The coordinators know down to what table the application is on.”

Another strategy for ensuring applications were accounted for was advocating with higher-level officials. In some cases, this could make it more difficult for lower-level officials or politicians to ignore or dispute the application. Instead of submitting an application to a lower-level official directly, one could go on Mondays to submit through the district collector’s office, a 45 minute bus ride from the block center. Mondays were dedicated to receiving petitions and grievances and all block-level officers, namely the revenue division officer, block development officer, and tehsildar for each block, were required to be available. In one observation, a coordinator handed an application directly to the collector and deputy collector, who then called the revenue division officer up and asked him to oversee the application. The revenue division officer affirmed to the coordinator with a nod that he would take care of it, after which the coordinator left the hall. This spatial and physical hierarchy creates clarity that the order comes from above.

In the Indian context, responsiveness could not be taken for granted, and the burden of following up was almost entirely on applicants. In the US context, the housing authority was more communicative about application status, but we found that there were similar challenges in ensuring information reached applicants. Overall, applicants for public housing or vouchers could be on the waitlist for years given the scarcity of housing resources. The housing authority periodically purged the waitlist or asked for more information once someone got to the top of the list, and those who did not reply in time were removed. One issue then was when participants were surprised to find out they had missed a communication and had been removed. B6 shared that she regularly followed up on her application for about a decade, but on one such visit, she found out that her name had been removed and staff only told her, “It’s not here. Reapply.”

From the housing authority’s perspective, they send all mail by certified mail, which requires a signature upon delivery and ostensibly offered some security. Phone calls were seen as unreliable due to changing phone numbers and weak ties to identity. Yet staff also admitted to issues of communication, highlighting how certified mail was a sort of bureaucratic checkmark as against more persistent accompaniment. B14 noted that the response rate to pre-denial letters, where applicants were given a chance to contest a possible denial based on their criminal record, was low, speculating whether people even received the letters. B12 noted how she tried to tailor communications to applicants when possible. For example, if the address on file was a homeless shelter, where people did not check mail regularly, she waited a couple of extra days before sending a denial or withdrawal letter. If she knew that someone does not speak English based on personal interactions, she would also send a translated letter, using Google Translate. However, this discretion was up to staff, and dependent on what they knew about applicants. Accompaniment was helpful in advocating for extensions if deadlines were missed. Participants could ask for extensions directly, but as B7 shared, the specific positionality of intermediaries could be important for pushback: “Once there is a lawyer, [the housing authority] is kind of flexible. Lawyers know how to write a letter and say, you know, this is the reason, maybe client doesn’t speak the language. And [the housing authority] didn’t do the job properly.”

4.2.2 Strategizing to Align with Stakeholder Incentives.

Upon actually being notified of an adverse outcome, intermediaries strategized with applicants about how to best approach contestation, particularly given the incentives and needs of other stakeholders in the decision-making process hold. In the Indian context, we found that other stakeholders within a community may disagree with allocation of particular pieces of land for pattas—here, accompaniment meant swaying local stakeholders’ decisions. For example, the panchayat leader is required to sign off on patta applications, but they may decide that they want that land to be used for other community features like a wedding hall or panchayat office. There may also be caste politics at play. According to C24, the legal advocate for the NGO, most panchayat leaders in the district were dominant caste, and there was an underlying passivity to supporting community development precisely because it would allow Scheduled Tribe communities to thrive: “They [dominant caste leaders] don’t like when [Scheduled Tribe] people are able to construct a house and a permanent life. They don’t prefer that because they need their labor, for example, for cleaning.” In cases where there is resistance from the panchayat leader, it may be possible for coordinators or their project manager to speak with them and advocate for the community. Given that panchayat leaders are elected, communities could also threaten to not re-elect them as a way to sway their decision.

Disputes might also arise between communities seeking land. A group of families in one village was applying for pattas and the land they sought in particular had been sold to people external to their community. According to one of the coordinators, the families felt strongly that the land should be redistributed rather than them finding land outside of their community. According to C23, as a very small minority within the district, “ST people won’t be willing to mix communities.” However, there was not much will either from government officials or the landowners in redistributing the land, and the applicants were offered land elsewhere. Families were insistent on finding land in their area, which they felt was tied to the preservation of their community: “Without land, our children could only have a home if they married outside of this area,” as one woman shared. C22 planned on submitting a new application entirely, with the aim that this will prompt the newly appointed revenue division officer to close the old application. The officer could then try to convince the landowners that the land should go to the applicants, and to wait for upcoming government schemes that would prioritize land for them.

In the Indian context, there were multiple inroads into convincing other stakeholders regarding allocation of land, depending on their identity and position. In the US context, appeals may seem like a more structured process, with a hearing officer available to consider both the applicant’s and housing authority’s reasoning and review the original decision. However, similar to the Indian context, intermediaries account for social norms and incentives of the housing authority, and construct arguments and coach clients to work with those incentives in the appeals process.

Intermediaries, particularly lawyers, helped frame appeals in ways that were legible and acceptable to the housing authority. Applicants might have legal defenses such as reasonable accommodation [3] or protections from the Violence Against Women Act [4], which defend against issues of improper discretion by the housing authority. However, as B10 shared, “They [applicants] haven’t framed it and they don’t know that there are actually some fairly good legal handles that other people can use,” making connections to legal services important. Importantly, he also noted that even if applicants do present information that could support such a legal framing, the burden could be placed on them to go further: “...they may not use terminology that prompts agencies to help. Even though it’s clear that agencies are not supposed to require magic language. You don’t have to say reasonable accommodation.” Without legal representation, clients may end up ceding interpretation of their circumstances to the housing authority, which may not offer inquire further, whether that is due to lack of resources or political will. Additionally, as B11 shared, when applicants appeal a decision, they must consider the housing authority’s perspective and meet a certain standard of evidence that addresses the reason for denial. In another example of ceding interpretation to the housing authority, clients might “...just go there and they say, gee, I’m really sorry, it won’t happen again,” (B11). Whereas a lawyer’s approach might look more like sharing “...our person is in treatment for anxiety and depression, and we will get them another assessment, and talk to their doctor about making changes to their medications... So these things don’t happen in the future, and that will be sufficient to get the housing authority to say, ‘okay’...” (B11).

Intermediaries also coached clients for appeals. According to B11, establishing credibility was especially important in cases where a client is being terminated and their appeal relies on refuting the housing authority’s claims. According to him, “...it’s not a fair presumption, but there may be this presumption that the housing authority has no stake in misrepresenting the facts, and the tenant does, because the tenant’s stake is the loss of their housing.” Building this credibility entails educating clients about the housing authority’s perspective, how to refrain from outbursts and disagreements, and how to answer questions concisely. This was done with the intention of matching how “lawyers and hearing officers are much more in the business of distilling things,” (B11). B13, a hearing officer, acknowledged that legal representation can help keep a hearing on track, but she was also confident that she made decisions based on a legal basis, not interpersonal dynamics. However, B7 also explained how without coaching clients, she foresaw a number of issues for people that apply or appeal themselves, especially the most vulnerable: “What about those families that apply by themselves... the client needs to [make claims] and make sure that she’s convinced of what she’s saying because clients with cognitive impairment don’t understand the question. They need to say yes, [but] they say no... You can prepare as much as you can, but when you go there in person, it’s different.” We also heard from participants that even if there was no possible recourse, having an advocate was a matter of dignity. For B6, she ultimately understood that there was no legal defense for the particular reason her voucher was being terminated but she maintained that an advocate would have helped during an intimidating appeal: “If I wouldn’t have left with the [voucher], I would’ve at least left with my dignity and my self worth.”

4.3.1 Legitimacy through Knowledge and Social Standing.

In both contexts, we see the power of accompaniment by legal advocates who are able to use their knowledge and positionality to bolster applicants’ pushback against unjust decisions. In the Indian context, filing a writ petition in the state’s high court could be used to force a decision from government officials on whether land is actually available. This method, facilitated by the fact that the NGO worked with a lawyer, was used once several follow-ups were deemed ineffective in getting a response. Exemplifying epistemic humility, C24 noted that escalation should only be building on prior community efforts. It is only in this case that “100% they’ll support” legal routes. C24 himself could also directly engage with actors in charge of provisioning resources like pattas and utilities, to use the power of potential court orders to spur action. In one case, families were allotted land in a village, but were still waiting on patta documents and functioning utilities. After the coordinators and their project manager’s follow-ups with the panchayat leader proved ineffective, C24 intervened, after which the panchayat leader immediately set up a water connection. According to C24, he was able to credibly suggest that he could petition the court or collector if the panchayat leader did not act.

In the US context, we saw applicants turn to legal advocates and other actors to support claims against unjust decision-making processes. In one case, B2 shared that her name had come up for a voucher twice, and both times, the housing authority informed her that they had never received a response to their request for updated information. In the second instance, B2, as a volunteer with legal services, began leveraging her connections with housing justice lawyers, as well as bringing in city council members and news channels, to bolster her case that this was unjust treatment. She felt that having legal services on her side was more convincing than her own attempts, demonstrating how accompaniment sometimes offers legitimacy:

“A lot of times, in the communities of color, people don’t value our voice... We can tell you, A) you can’t do this. B) you’re breaking the law. And it doesn’t matter. They say, what are you gonna do about it?... But when you have a voice of someone [lawyers] who holds a higher position or who can cause them some trouble, now they’re going to listen.”

In both contexts, dedicated resources were key—in the Indian context, the lawyer commits time to the community’s legal needs via the NGO, while in the US context, there were legal services committed to housing and other social justice issues. However, resources for this work were limited. For example, in the US context, B14 noted that the dearth of legal services for housing in the area meant most organizations focused on keeping people in housing, with fewer dedicated to supporting people with denials.

4.3.2 Working with Community on Collective Action.

In both contexts, we also see the importance of strong relationships amongst intermediaries and communities in coordinating collective contestation. In the Indian context, petitions did not always yield results. In these cases, the coordinators, as part of Scheduled Tribe communities, would organize community members for a sit-in at the collector’s office. The hope was to garner media attention, as well as prompt higher level officials within the state to ask the district administration to resolve the issue. Notably, actors like the lawyer who were not part of the community would stay out of this process, indicating different forms of credibility in different tactics. Rather, tribal associations might be involved here, as they were practiced in taking an activistic approach to community issues. In one illustrative case, a sit-in helped demonstrate the importance of making an exception to rules around benefits for widows. A women’s husband had disappeared and it was unclear if he had passed away, so there was no death certificate to provide for a widow card. The woman and a number of others in similar circumstances had been repeatedly denied by the collector, until they went as a group to demand the cards. According to C24, this garnered media attention and the collector provided the cards right away.

The relationships coordinators built with government officials could also pay off in unintended ways to escalate issues. In the case of the panchayat leader deprioritizing housing for one village, they found that advocating to government officials was not helpful. In the end, what helped was an image of the village that C22 posted on his WhatsApp status, mentioning how “I work in this area, the houses are old, water gets into the houses when it rains, shouldn’t the government help? Don’t they have responsibility here?” Such posts were not unusual for C22, but because C22 had the welfare officer’s contact, the officer reached out.

Notably, these demonstrations in the Indian context were in support of more favorable decisions in particular instances for particular groups of people. It is an important question as to whether these exceptions then become integrated into policy, or if they remain exceptions, but it was clear to participants that the overall responsiveness of government officials to Scheduled Tribe communities was increasing due to coordinators’ and applicants’ engagement. In the US context, legal services played a key role in forms of advocacy that more formally changed the basis of decisions in housing. For example, B14 explained how housing authorities host public participation processes and legal service organizations are active leaders in them. In one example, advocates were able to shorten the lookback period for criminal records checks, in support of equitable access to housing for justice-involved individuals.

Strong relationships among community organizations and clients were also important for legal case-making. At the time of the study, B9 was building a case against the use of an algorithmic tenant screening tool by landlords, which was denying applicants in cases where there did not some to be reason for it, such as when they had vouchers. According to B9, getting plaintiffs for the case was “not just a one-way street where oh, you’re gonna be a good plaintiff”—this process required coordinating with housing NGOs to reach out to people who had been denied by the tool and actually support them in addressing the denial. This is ultimately what enabled the investigative process of understanding why applicants were denied by the algorithm, testing what kind of applications would receive a positive versus negative outcome, and testing the responsiveness of landlords’ and the screening company’s appeals processes.

5.1.1 Algorithmic Accountability.

Prior work has called for a situated approach to algorithmic accountability premised on the idea that contestation and its effectiveness are shaped by sociopolitical contexts [51, 58], with significant differences across Western and non-Western settings [79]. Our findings point to how apprehensions around contestation do manifest across geographies, though in different ways depending on applicants’ relationship to the state. We find that it is then important to cultivate both the motivation to contest and the safety of acting on that decision, if we seek to enable contestation. We found that in the Indian context, for example, communities felt strongly that they should be able to get pattas in certain areas, but there were still concerns about the power differentials between communities and the state, so collective methods of contestation were useful here. In such a context, it may be important for algorithmic tools to help applicants and intermediaries understand who else has been denied or deprioritized (with information anonymized) and why, so that there is a basis for bringing up concerns within their communities and with decision-makers [6, 62]. In comparison, in the US context, we found that there were clients who did not want to appeal, while there were others who did but faced barriers in terms of legal knowledge and other resources. In this type of setting, enabling contestation may have more to do with raising awareness of rights related to appeals and connecting them to the end goal of achieving better outcomes. This builds on findings in prior work on toolkits for thinking about algorithmic equity [51]; in addition to AI literacy efforts, such toolkits may also need to tie into information about more fundamental rights and legal resources.

We also bring attention to the temporal and community-oriented nature of building capacity for contestation. We found that there was a range of attitudes towards contestation, and not all applicants were convinced of the importance, even after going through a successful appeals process. This suggests that it may be important to build on efforts towards critical awareness for the general population, as proposed in prior work [47, 79], to also find ways to better connect people to social and legal resources when they receive adverse outcomes. Actors beyond applicants and intermediaries may be important here as well, such as community associations or tenant unions in the context of housing. This is doubly beneficial as we also saw how such strong community relationships are what enable the social process of coordinating collective contestation, such as legal case-making. These relationships could also be important for scrutinizing the use of less obviously visible use of algorithms, such as in decision-support or organizational planning contexts.

5.1.2 Human Review.

Policy documents have stated the importance of human review of algorithmic decisions [72], and HCI research finds that human review (as opposed to algorithmic review) provides the opportunity for dialogue and influence, assuming the interactions are conducted with dignity and compassion [61, 110]. We find that what enables applicants to go through contestation with dignity is non-trivial. Contestation in public services may be at odds with dignity, as some participants pointed out. Appeals can be a scrutinization of character, behavior, and deservingness, or government officials may be unresponsive or retaliatory. Designing a just review process then is a question of not just human versus algorithmic review, but also factors like whether applicants have access to advocates working in solidarity with them.

Designing a just review process may also be a question of what kind of support clients need in framing their appeal. For example, we saw how there was room for public service administrators to more actively and consistently help clients construct an appeal. We also note that the purpose of review was not always for contesting the substance of a decision. Follow-ups or appeals could be to ensure applications were still under consideration, or to better understand application requirements (in cases where this did not happen early on). In a context where algorithms are used, the latter presents a case where a review process should actually support learning goals. This suggests that the decision-making domain and prevalent reasons for denial may matter greatly when designing review processes.

5.1.3 Explanations.

HCI and AI research have focused on numerous types of explanations to make decisions and systems more transparent and enable applicants to achieve different decisions [6, 57, 62, 109]. Our findings confirm that denials are often shaped by power dynamics stacked against marginalized communities, limiting the applicability of counterfactual explanations in the AI literature [48]. This reaffirms prior work advancing methods such as audits [25, 27, 90] and process-centric explanations [109] that enable contestation of the larger decision-making process. Still, our findings show that we can go beyond transparency of decisions and processes at the point of denial as well, since specialized knowledge such as legal defenses is sometimes necessary for appeals. Prior work discusses how there is room to expand the types of information algorithmic tools provide, in order to enable more actions [64]. Our findings emphasize the range of actions that tools could better support. Appeals, for example, can be a matter of reframing information, claiming to be an exception, or providing missing context for existing information. For explanations then, it may make sense to provide a related set of common defenses or helpful contextual information that people may use to build their argument for an appeal. Still, we emphasize that intermediaries also addressed apprehensions around contestation and emotionally supported applicants, indicating how equally important are sociotechnical interventions that provide social support for contestation processes.

Our findings draw attention to the amount of labor put into contestation. Intermediaries and applicants were already doing significant work to understand the intricacies of applying for housing, how decisions get made, and how to effectively appeal, in order to support clients and push back against the housing authority when needed—in cases where clients were not working with intermediaries, much of this work was on them. Our findings suggest then that the relationship between explanations and contestability needs to account for the possibly undue burden that may be placed on people contesting decisions or systems. For example, Kuo et al.’s work on decision-support in homeless services showed that connecting clients to resources without the support of the algorithm required significant advocacy, and that non-use of the algorithmic assessment was discouraged [60]. In the context of our study, we question how to ensure that designing for contestability does not make it harder to advocate for clients or push back against decision-makers, especially given that clients on their own may be more likely to be overlooked by public service administrators. For example, it may be important to investigate whether there are aspects of explanations or justifications, such as level of detail, that contribute to a lower likelihood of contestation. In the case of automated or semi-automated decisions, it may be important to provide guidance for human reviewers on how to evaluate applicants’ appeals and whether algorithmic decisions and explanations should even be taken into account if they are being contested, taking care to avoid increasing the work of appeals.