“A fundamental human right”

Access to credit, which Nobel Peace Prize winner Muhammad Yunus pronounced as “a fundamental human right,” is inequitably distributed in the United States. ^{Footnote 1} Roughly 16% of Black, as well as Latinx adults, do not have a bank account, compared with 7% of White adults. Among Black adults with bank accounts, 29% are reliant on alternative, and often more predatory, financial services, such as auto title, payday, and non-bank loan debt and/or check cashing services, consistent with being underbanked. The figure for underbanked Latinx adults is similar, at 25% and falls to 11% for White adults. ^{Footnote 2} For American Indians and Alaskan Natives (AIAN) living in and around the 326 independent jurisdictions in Indian Country, the problem of un- and underbanking has long been understood as even more extreme, but efforts to quantify its scope have been stymied by consistent undersampling of AIAN communities in national surveys (Brown, Cookson, and Heimer, Reference Brown, Cookson and Heimer2019; Akee and Jorgensen, Reference Akee and Jorgensen2014). ^{Footnote 3} American Indians living on reservations are some of the most economically marginalized US communities (Akee and Taylor, Reference Akee and Taylor2014), with lower average credit limits (Dimitrova-Grajzl et al., Reference Dimitrova-Grajzl, Grajzl, Guse and Todd2015), higher mortgage interest rates (Feir and Cattaneo, Reference Feir and Cattaneo2020), and less access to financial markets (Wellhausen, Reference Wellhausen2017; Brown, Cookson, and Heimer, Reference Brown, Cookson and Heimer2019; Anderson and Parker, Reference Anderson and Parker2008).

In a first-of-its-kind survey, the authors collaborated with one American Indian nation, Nation “A,” to measure its tribal members’ capital access. ^{Footnote 4} Nation A, a federally recognized tribe with more than 10,000 members, is among some of the most economically marginalized communities. Extrapolating from the survey, an extraordinary 33% of Nation A tribal members do not have a bank account, and 50% of those with bank accounts are underbanked. ^{Footnote 5} Even in the context of the strong prepandemic economy, 31% of Nation A survey respondents reported that they could not come up with $400 in case of an emergency, whether through savings or (informal or formal) borrowing, and a further 16% were unsure. ^{Footnote 6} Such high percentages in our survey of Nation A were not unexpected, as its reservation is a “banking desert” without a local provider of formal financial services and has been that way for longer than tribal elders can remember. The closest retail bank branch is about ten miles away on roads that are difficult to drive in winter.

Nation A leaders have long deliberated on how to improve capital access. As the reservation’s second biggest town is not covered by cell service, internet-enabled solutions remain unrealistic. ^{Footnote 7} The Nation A government runs a well-received small-dollar loan program, but expansion is not thought viable. Rather, discussions have centered on attracting a physical branch of a preexisting retail bank. Indeed, physical branches remain important throughout the United States; in 2017, 84% of Americans visited branches, and almost all did more than access the ATM (Merry, Reference Merry2018). It was therefore a turning point when Nation A’s tribal government successfully negotiated the entry of a commercial bank branch, and it was all the more notable as the branch is owned by American Indian Nation “B.”

Nearby Nation B’s urban reservation helps to make its casino and hotel very profitable, although, like many American Indian nations, it is diversifying away from gaming given gaming’s uncertain future. One of Nation B’s key ventures is Bank [X]. ^{Footnote 8} Bank [X] is unique in pursuing expansion in Indian Country; to this end, it has become certified as a Community Development Financial Institution (CDFI), a federal program that allows it to reorient from profit-maximization toward commercial viability. Bank [X] proposed to open its first Indian Country investment on Nation A’s reservation.

For the branch to be commercially viable, as Bank [X] requires, and to improve access to capital on the reservation, as Nation A desires, Nation A’s tribal members must support the branch and, most importantly, become its customers. Previously un- and underbanked customers would derive direct, material benefits, and more customers overall would both improve the branch’s commercial viability and its potential for contributing to community development. These outcomes are consistent with the mission of the Federal Reserve System, the external institution with a specific mission to mitigate financial exclusion in Indian Country. The Federal Reserve is further interested in proof-of-concept of a Native-owned bank successfully expanding via investment in other American Indian nations. Thus Bank [X], Nation A, and the Federal Reserve System shared the goal of bolstering tribal members’ buy-in to the bank branch before its groundbreaking. Individual-level public opinion—for example, general trust in banks, or support for Bank [X] in particular—can affect economic development even in highly impoverished settings when opinions are translated into consumer choices (Rudra and Tobin, Reference Rudra and Tobin2017). Thus how to promote potential community buy-in for the bank is important.

These unique circumstances offered an opportunity to evaluate the success of stakeholders ^{Footnote 9} in influencing public opinion regarding the bank, in an environment where results are consequential for individuals’ choices over their personal finances and development outcomes across the target population as a whole. The authors solicited interventions from the bank, Nation A’s tribal government, and the Federal Reserve System to include in the survey, conducted from January through March 2020, prior to the bank’s planned groundbreaking in the summer of 2020. Each independently supplied a cue that invoked their credibility as a source, allowing an exploration of when and how stakeholder credibility matters in shaping public opinion (Pornpitakpan, Reference Pornpitakpan2004).

Our survey revealed very high and quite homogeneous baseline support, further reinforced after cueing the bank’s Native ownership. However, the average treatment effects (ATEs) of the tribal government and Federal Reserve experimental interventions are small in magnitude, often insignificant, and exhibit unintended effects—in some cases reducing support for Bank [X]. Average treatment effects are highly similar for both Nation A and Federal Reserve treatments, allaying the concern that the Federal Reserve’s US-tied identity undermines its credibility among Nation A residents. However, exploratory analyses suggest that pro-banking cues from the tribal government and the Federal Reserve reduced support among lower-income respondents but proved effective among those with more information about Bank [X]’s ownership structure before the survey and those employed by the tribe. Our nuanced results illustrate the need for renewed theory-building around the political economy of credibility at the most micro-level.

In what follows, we discuss the theory related to stakeholder interventions and individual attitude, our real-world setting and ethical considerations driving our research design, emphasizing the scholarly benefits of stakeholder-driven hypothesis testing. We situate survey interventions in cueing theory, discuss survey results, and explore potential heterogeneous effects of the interventions. We conclude with implications for scholars and practitioners concerned with the political economy of development. Our results suggest that using stakeholder cues to generate public backing for development-enhancing projects may even be counterproductive, a caution that is especially relevant for underserved communities with a history of marginalization.

Theory: Stakeholder credibility and individual attitudes

Given broad agreement among stakeholders that Nation A’s residents would benefit from using Bank [X]’s services, what strategies can be used to increase support for the bank? To answer this question we borrow insights from scholarship on cues, or “information that enables people to form evaluations about an attitude object without in-depth knowledge” (Eagly and Chaiken, Reference Eagly and Chaiken1993; Nicholson, Reference Nicholson2011). When elites provide cues, a key prediction is that effectiveness varies with the credibility of the source (Mondak, Reference Mondak1993; Guisinger and Saunders, Reference Guisinger and Saunders2017). Source credibility results from the source’s perceived trustworthiness, expertise, or some combination (Pornpitakpan, Reference Pornpitakpan2004). The basic idea is that individuals are more likely to adopt a position on a given issue—even if they know little about it—if the position is endorsed by an actor that they deem credible, while endorsements from non-credible sources may fail to shift public opinion.

Typically, top-down interventions around financial exclusion teach the respondent why they should support the bank via a financial literacy mechanism. In the lab, interventions that instruct respondents on their material self-interests have been shown to change attitudes (Rho and Tomz, Reference Rho and Tomz2017) In the field, financial literacy programs have been deployed in underserved communities around the world, with meaningful evidence of effectiveness (Goyal and Kumar, Reference Goyal and Kumar2021). However, financial literacy interventions have not solved Nation A’s issue. For example, of our Nation A respondents who reported being unbanked, 37% of them said that they had participated in a formal financial literacy program, and 50% self-reported levels of financial knowledge above 6 on a 10-point scale.

Rather than adopting a teaching-based treatment, we draw on theories of elite cues to create two simple and straightforward endorsement-based treatments: we provide respondents with brief messages from two key stakeholders (Nation A’s government and the Federal Reserve) in support of Bank [X] and/or its mission. It’s important to note that the text of both messages was provided directly by the stakeholders and was not edited or standardized by the authors. In making this choice, we sacrifice perfect comparability across treatments; however, in return, we gain increased external validity by adopting language that each stakeholder would be likely to use in other real-world policy messaging. In addition to our randomized treatments, we conduct observational analyses in which respondents are asked how the identity of a hypothetical bank’s owners would affect their support for its operations; these analyses are designed to gauge the extent to which Bank [X]’s Native ownership may affect its support among Nation A residents.

We argue that, to the extent that the stakeholders are indeed perceived as credible by Nation A’s residents, each of the three stakeholders should derive their credibility from a distinct source: Bank [X]’s credibility stems from the identity of its owners, the Nation A legislature’s credibility stems from its democratic institutions, and the Federal Reserve’s credibility stems from its technocratic expertise on financial issues. In the sections below, we describe how we attempt to map these distinct forms of credibility to our treatments and observational analyses.

Research design and implementation considerations

The authors collaborated on the first non-Census scientific survey of adult (18+) Nation A tribal members, funded by the Federal Reserve and executed by the authors together with enumerators from Nation A’s tribal college. The authors participated in their institution’s IRB approval process, the appropriate processes at the Federal Reserve System, and the Tribal approval process to ensure that the research was both ethical and valuable. Bank [X] to endorsed the survey and the capacity of the authors to conduct it. Bank [X] did so while also making clear that it is not part of the research team; it has no privileged access to data; and its investment is in no way conditional on the survey. ^{Footnote 21} In several rounds of in-person testimony, legislators carefully reviewed the survey with the authors. ^{Footnote 22} As was their right, legislators required the authors to cut voting and ideology questions, as well as questions viewed as simplistic or disrespectful. The authors also committed to anonymizing public-facing research (hence Nation “A”). ^{Footnote 23}

Figure 1 summarizes the survey flow. ^{Footnote 24} Dotted boxes indicate the points at which the survey introduced different pieces of information, alongside an information check where appropriate. The survey begins with blocks of questions on demographics; financial knowledge; use of financial services and credit; and respondents’ baseline support for and willingness to become a customer of a local bank. These blocks are followed by our main questions of interest, in an observational form and then a randomized experimental design. The last block is specific to the needs of our stakeholders, including questions about respondents’ priorities regarding financial services and their preferred means of accessing a local bank branch. ^{Footnote 25}

Figure 1. Survey flow.

The survey was intended to run from January through May 2020, which would be approximately one month before Bank [X]’s scheduled groundbreaking. After that time, Nation A would no longer be in a credibly prelocal bank context. We of course stopped the survey abruptly in March, consistent with public health priorities around the arrival of the COVID-19 pandemic. ^{Footnote 26} Nonetheless, we collected 982 responses from the target population, adult (18+) Nation A’s legally recognized members. ^{Footnote 27} For its part, Bank [X] delayed its scheduled groundbreaking until it was prudent to hold a socially distanced event.

To conduct the survey, we collaborated with Nation A’s tribal college to hire and train ten enumerators, who were all female students without prior experience. ^{Footnote 28} Enumerators facilitated the survey on tablet computers via the offline Qualtrics app (Bush and Prather, Reference Bush and Prather2019). Although originally offered as an incentive to collect a given number of survey responses, due to the pandemic halt all enumerators were gifted their tablets regardless of whether they collected at least 100 responses. Enumerators set up stations in high-foot traffic areas on the Nation A reservation, including the casino lobby, which is a typical space used for community events; the health clinic; senior centers; government offices; and the main tribally owned enterprise during shift breaks. ^{Footnote 29} Enumerators also leveraged their personal connections, for instance to the main on-reservation bar and restaurant; disability care services; and drug and alcohol rehabilitation services.

We instructed enumerators to use convenience sampling, rather than selecting potential respondents randomly or randomly within demographic strata, for three reasons. First, the legislature required as a condition of approving the project that as many people from their community as possible participate in the survey. Second, our enumerators helped us settle on a $10 gift card to the only on-reservation grocery store as an effective form of compensation for survey respondents, which had the added benefit of keeping funds in the local economy. Our enumerators assured us that, in this small and low-income community, news of the incentive would travel fast. We therefore saw it as a high risk that randomly denying some tribal members the opportunity to receive a gift card would generate unpredictable confounders via resentment or other mechanisms. Third, methodology aside, the authors believe that implementing randomization—thereby forcing enumerators to prevent fellow tribal members from having their voices heard in the first non-Census survey—was simply inappropriate. However, even enumerators’ implicit sampling strategies were stymied by the survey’s abrupt stop after three months (of a planned five).

To determine whether or not our convenience sample is reasonably representative, we compare it to population averages in Nation A’s official records and its US state (Appendix B.3). The key observational imbalance is our oversample of women, which may be a function of the composition of our (female) enumerators’ social networks (Schroedel et al., Reference Schroedel, Berg, Dietrich and Rodriguez2020). Each of two treatment groups and the third control group are generally well balanced on observables, indicative of successful randomization. Nonetheless, we report results with and without a battery of controls (listed in Appendix B.1).

The work was conducted to ensure that the tribe and bank gained from the research given the ethical importance of reciprocity in Indigenous research (Feir and Hancock, Reference Feir and Hancock2016; Drawson, Toombs and Mushquash, Reference Drawson, Toombs and Mushquash2017; Hayward et al., Reference Hayward, Sjoblom, Sinclair and Cidro2021). Thus the authors delivered a community report to the tribe, the tribal college, and Bank [X] answering questions related to their research interests. The community also directly benefited from the hiring and training of the enumerators, and the provision of technology through the enumerators keeping their tablets. The authors were told by partners at the tribal college the report they delivered would be used as teaching material.

Extension: Exploring unintended treatment effects

We look into the possibility of heterogeneous treatment effects in a post hoc, exploratory exercise, in the spirit of theory-building and to identify patterns of potential use to practitioners. We focus on the “difference(s) in changes” of the Likely Bank [X] customer outcome, because it is of special importance to economic development outcomes. We also focus on those receiving the Nation A intervention, as heterogeneous effects would be of special importance to Nation A policymakers in understanding the limits of the government’s source credibility with its own population.

We employ the following process. First, we want to know which of our pretreatment covariates (Appendix Table B.1) are the best predictors of how participants responded to treatment; we will focus our heterogeneity analysis on these important variables. To answer this question, we first regress the outcome variable on a set of covariates for the treatment and control groups, and we predict the counterfactual outcomes for each treatment group (Appendix Table B.7). We then use these predicted counterfactual outcomes to generate estimated individual-level treatment effects (Appendix Figure A.2). We then form an indicator variable of whether each individual’s predicted treatment effect was in the unintended direction.

Now that we have estimated individual-level treatment effects—and dichotomized them by expected (positive) vs. unexpected (negative) directionality—we want to know which individual-level covariates best explain them. To do so, we use lasso regression, which minimizes the sum of squared errors while constraining the sum of all estimated coefficients below some threshold, and in the process retains the strongest predictors of the indicator variable while shrinking the rest of the coefficients towards zero (Tishbirani, Reference Tishbirani1996). This data-driven process leaves us with the most relevant covariates to our outcome of interest; we chose this route, selecting variables according to their predictive power instead of their theoretical motivation, in the interest of contributing to future theory-building and providing policy-relevant information to policymakers.

Once we’ve identified these meaningful respondent characteristics, we directly test for heterogeneous treatment effects. To do so, for each (binary) characteristic, we first use SURs to estimate the treatment effect separately for groups with and without a given characteristic. Formally, for each characteristic $c$ :

(2) $${Y_i} = {\delta ^{c = 0}}{D_i} + {\varepsilon _i}$$

(3) $${Y_i} = {\delta ^{c = 1}}{D_i} + {\varepsilon _i}$$

We then perform chi-squared tests to assess the difference between treatment effects for the subgroups with and without a given characteristic ( ${\delta ^{c = 1}} - {\delta ^{c = 0}}$ ). Figure 5 reports the results of this procedure for each respondent characteristic and for both treatments, along with robust 95% confidence intervals. When estimates are positive, individuals with those characteristics were especially enthusiastic in response to Nation A’s intervention. In other words, compared to their initial interest in becoming a customer of Bank [X], individuals with these characteristics were the most likely to become more interested upon learning of their government’s support. In contrast, when estimates are negative, the Nation A intervention was so unhelpful as to be counterproductive. In other words, compared to their initial interest in becoming a customer of Bank [X], individuals with these characteristics were the most likely to report less interest upon learning of their government’s support.

Figure 5. Which respondent characteristics conditioned the effects of treatment on their likelihood of becoming a customer of Bank [X]? Positive values indicate that the treatment had a more positive effect among individuals with the given characteristic; negative values indicate that the treatment had a more negative effect for individuals with the given characteristic.

Overall, Figure 5 suggests that ATEs masked heterogeneity across groups of respondents that merits further attention. ^{Footnote 36} First, particularly for the Nation A treatment, individuals with stronger ties to the community appeared to respond more positively. Those employed by the tribe and those who were surveyed at the casino (a community gathering place, and also the location of the survey roll-out event) seem to have more positive treatment effects, though the latter effect is not statistically significant. Likewise, respondents who paid close attention to tribal affairs would almost certainly have knowledge of Bank [X]’s owners, so the strongest predictor of unintended treatment effects (to the left of the dashed line) suggests weak community involvement. If stakeholders expect that individuals connected to the community are also effective in persuading other community members, then focusing interventions on them could be one way to maximize impact.

Given the policy goal that Bank [X]’s entry would mitigate financial exclusion, and the context of very high baseline buy-in, some of the characteristics associated with more negative responses to the treatments are normatively concerning. In particular, the most negative treatment effects are found for respondents who are young, low-income (particularly very low-income), and who lacked knowledge of and experience with consumer finance (those who lacked bank accounts, and who did not show interest in a free credit report when offered). Our overall interpretation of this exercise is that the Nation A and Federal Reserve interventions were damaging especially to respondents in groups that they were most interested in targeting. As the interventions relied on a credibility mechanism, a broader takeaway is that stakeholders’ ex ante evaluations of their credibility—and not only its average, but also its distribution in the target population—is a key input to deciding whether intervention in itself is on net a productive choice.