2

Assessing the Landscape: The Measurement and Modeling of Structural Racism

EXPERIMENTAL DESIGN

René D. Flores (workshop planning committee member and Neubauer Family assistant professor of sociology at the University of Chicago) explored the use of experimental methods to identify the most common racial stereotypes in the United States, as well as to better understand how the intersection of race, class, and gender shapes these stereotypes.

Flores explained that stereotypes are assumptions about a group’s behavioral traits and capacities (Bobo et al., 2012) that reveal how race is made. As the “building blocks of racial difference,” stereotypes illuminate the changing nature of ethnoracial boundaries (Lamont, 2009) and serve as a means of explaining social processes such as discrimination, harassment, and intergroup relations (Fiske and Neuberg, 1990; Jussim et al., 1996). However, he observed that much of the existing research on stereotypes has examined stereotype content deductively—typically using closed-ended formats (Bobo and Kluegel, 1991; Schachter, 2021). While some inductive open-ended exercises exist, they are typically collected on convenience samples (Karlins et al, 1969), and issues of external validity may arise. Furthermore, some stereotype content research suggests that the intersection between two stigmatized categories (e.g., race, class, and/or gender) could actually decrease stigma (Pedulla, 2014) and that when gender and ethnicity are combined, novel stereotypes could be created (Ghavami and Peplau, 2013). Flores described how he and his colleague, Michael Gaddis (University of California, Los Angeles), are approaching stereotype content research, with a fully inductive methodology for revealing existing stereotypes; a nationally representative sample of U.S. non-Hispanic White (White) adults, who play an important role in the formation of U.S. ethnoracial boundaries; and a framework to identify intersectional effects.

Flores echoed Margaret Hicken’s (workshop planning committee member and research associate professor in the Institute for Social Research at the University of Michigan) assertion about the value of incorporating more theory into studies of structural racism. Currently, the social science field

is guided primarily by two models for understanding social stereotypes. First, the stereotype content model suggests that competence and warmth are the two key dimensions of stereotype-making. In this model, high-status, powerful groups are perceived as competent but perhaps as having little warmth (Fiske et al., 2002). Second, the racial position model suggests that the key dimensions of stereotype-making relate to groups perceived as inferior or superior and foreign or native (Zou and Cheryan, 2017). To determine which model has greater relevance for racial stereotypes, Flores and Gaddis conducted interviews of 200 White adults, using Amazon’s crowdsourcing platform MTurk. In an open-ended fashion, they asked participants to name one of the most common stereotypes of White people, Hispanic people, Black people, Asian people, men, women, lower-class people, and upper-class people. From these interviews, approximately 116 traits were identified for each group. Flores and Gaddis used this information to populate 44 different Wiki surveys on all possible combinations of race, gender, and class. Using the online survey platform Prolific, 1,450 White survey respondents voted 89 times for a total of 128,946 votes on 664 unique stereotypes. As people voted on the prevalence of common stereotypes (or added additional stereotypes), a ranked order of the most popular stereotypes was produced. Finally, a new survey experiment was designed for a nationally representative sample of 2,500 White adults using YouGov, in which each adult was randomly assigned to one of the 44 combinations of gender, class, and race, and determined their conformance to the 250 most popular stereotypes identified in the prior survey.

Flores remarked that using the stereotype content model, one would expect to find the most common existing racial stereotypes to include degrees of competence (e.g., confidence, independence, competitiveness, organization, and intelligence) and warmth (e.g., tolerance, friendliness, and sincerity). However, Flores and Gaddis found that such stereotypes were not very popular: mentions of competence and warmth accounted for about 10 percent of the stereotypes identified by the nationally representative sample of White people. Flores displayed the top 30 stereotypes for each ethnoracial group as defined by the survey participants. For example, they perceived other White people as high-status, privileged, rich, powerful, and native; Black people as monolingual, poor, welfare-dependent, and low status; Asian people as foreign, career-oriented, and intelligent; and Hispanic people as foreign, poor, and undocumented. The participants revealed consistently positive stereotypes about Asian people, which contradicts the premise of the stereotype content model that “competent” people are perceived to lack “warmth.” Flores and Gaddis concluded that focusing only on the dimensions of competence and warmth would not account for the full social reality of racial stereotypes; the racial position model better accounts for the patterns identified in the data they collected about the most

popular stereotypes (e.g., participants viewed White people as American and superior, Black people as American and inferior, Asian people as foreign and superior, and Hispanic people as foreign and inferior).

Flores and Gaddis also used this experiment to consider how the intersection of race, class, and gender shape stereotype content, and observed that racial stereotypes overlap with class and gender stereotypes (Figure 2-1). For example, the participants’ stereotypes about White people often tended to overlap with stereotypes about upper-class people and stereotypes about men, and stereotypes about Black and Hispanic people were more likely to overlap with stereotypes about lower-class people. Questioning how the composition of stereotypes changes with different combinations of race, class, and gender, Flores and Gaddis discovered a different distribution of stereotypes for lower-class White people. Flores explained that a difference of the same magnitude in the distribution of stereotypes was not apparent for lower-class Black, Hispanic, and Asian people and that there was little effect of intersectionality was shown relative to the other groups regardless of the race, class, and gender combination.

In closing, Flores discussed the benefits of the racial position model, which offered significant explanatory power to understand the racial stereo-

types content model. Additionally, although Flores and Gaddis’s research revealed significant intersectional effects on stereotype content, the direction and magnitude of these effects depend on White participants’ initial impressions of each ethnoracial group’s socioeconomic status. Flores emphasized the value of using experimental design to observe the specific configurations of particular social situations, so as to better understand both the stereotypes that might be produced and their social consequences.

In response to a question from David Takeuchi (workshop planning committee member and professor and associate dean for faculty excellence in the University of Washington School of Social Work) about the method used for the experimental design survey, Flores explained that the experimental design survey was the final part of the data collection process. He noted that the reason for randomly assigning each participant to one of the 44 experimental conditions (which were derived from all possible combinations of race, class, and gender) was to reduce concerns about ocial desirability. This approach allowed Flores and Gaddis to estimate the experimental effects on individually reported stereotypes as a result of being assigned to each of the different conditions.

Hedwig (Hedy) Lee (workshop planning committee chair and professor of sociology at Duke University) asked about the future of experimental design. Flores replied that although experimental design has long been used for psychology research, sociologists, political scientists, and economists are beginning to build on that literature and engage more often with experimental methods—vast and inexpensive datasets are available online (e.g., via Prolific), and many questions can be explored with randomization. He cautioned that issues related to data quality, access, and analysis and replication standards, as well as registration standards and data training, might arise in experimental design, many of which could be addressed with more resources and a better infrastructure.

QUASI-EXPERIMENTAL APPROACHES

Jamein Cunningham (assistant professor in the Jeb E. Brooks School of Public Policy at Cornell University) provided an overview of the use of quasi-experimental research design for studying structural racism, particularly in relation to police use of lethal force. He noted that police violence is the leading cause of death for young Black males (Edwards et al., 2019), and in 2021, 1,051 Americans were killed by law enforcement officers—a disproportionate number of these Americans were from Black and Indigenous communities. Police violence is a growing concern not only in the United States but also in Europe and Canada.

Cunningham explained that recognizing and challenging structural racism first requires an understanding of the past. For example, the economic

history community has long relied on quasi-experimental research design to study slavery, emancipation, and reconstruction (Conrad and Meyer, 1964; Fogel and Engerman, 1974; Ransom and Sutch, 1977); lynchings (Cook, 2014; Cook, Logan, and Parman, 2018); and segregation (Collins and Margo, 2003; Logan and Parman, 2017). He stressed that modern policing in the United States is related to the nation’s complex history of race and discrimination. Slave patrols in the South were one of the first and most formal forms of American policing; furthermore, militias criminalized the behavior of former slaves after the Civil War, followed by state-sanctioned violence that directly or indirectly involved police (e.g., lynchings and White mob violence) (Chicago Commission on Race Relations, 1922; Lieberson and Silverman, 1965). Essentially, he continued, police departments opted to serve the White population.

To further illuminate the relationship between the past and the present, Cunningham described the positive relationship between the number of historical lynchings and the number of Black people killed by police, as well as the inverse relationship for White people (Williams and Romer, 2020). When testing this model to understand the persistence of this racial violence, Williams and colleagues (2022) found that places where lynchings occurred in the past have higher police violence today, as well as an increased intensity of racial violence more generally.

Cunningham remarked that emerging quasi-experimental and applied econometric scholarship can be used to understand the evolution of the use of lethal force in policing over time, the role of structural factors in determining the use of force, and the impact of policy interventions over time. He explained that data about police violence became more accessible in the 1950s and 1960s; however, deaths remained undercounted post-1960 (see Cunningham and Gillezeau, 2021; Figure 2-2). Several research studies have explored whether the increase in police killings of civilians in the 1960s could be attributed to structural, legal, or departmental factors. He emphasized that quasi-experimental methods are well-suited to study racial disparities in police violence and to isolate the effect because police violence is not random or conducted in a control environment, and these methods can exploit variation in exposure or treatment.

Reflecting on whether structural factors influence racial disparities in police violence, Cunningham mentioned that segregation has long been presumed to have driven the increase in use of lethal force in the 1960s. When Cox and colleagues (2022) tested the causal impact of segregation on victimization by using the index of dissimilarity and exploiting the historical layout of railroads, they found that segregation strongly predicts racial disparity in homicides; however, they were unable to link structural factors associated with poverty, inequality, and segregation to police killings of civilians in particular.

To analyze legal factors that could influence racial disparities in police violence, Cunningham and colleagues (2021) exploited the variation between when/where police could not collectively bargain and when/where they first could. The researchers found that police killings of civilians began to increase 4–5 years after officers could collectively bargain (e.g., in the event of a killing of a civilian, police unions pay for and facilitate legal representation, meet with their member in advance of making a report, potentially facilitate a “huddling” of officers, and implement procedural protections during interrogation). Furthermore, they found that the introduction of bargaining rights for police increased the use of lethal force against non-White Americans by more than 70 percent, an increase that is not related to segregation or poverty.

Cunningham related that, to explore departmental factors that could influence racial disparities in police violence, Cox and colleagues (2021) exploited the variation in employment discrimination litigation against police departments in the 1970s and 1980s and found that racial diversity matters in police departments after a threat of affirmative action litigation, an effect that explains the relative decline in non-White deaths at the hands of police after 1970. Cunningham asserted that this is another example of the unrelated effect of structural factors, such as segregation, poverty, and

inequality, demonstrating that police can change their behavior to reduce police violence.

Cunningham observed that the use of lethal force entered a relatively stable period with an upward trend in 1980, including a surge for non-White populations after 2000 and for White populations after 2008. This surge can be partially explained by militarization (Masera, 2021), force size and engagement (Goel et al., 2016), polarization and racial bias, and increases in violence and the drug trade (Holz et al., 2019). Cunningham and Stuart (2022) exploited variation in time and intensity of the Great Recession but found that the post-2008 surge in lethal force could not be explained by structural factors such as the related labor market outcomes. Even though the causes of this surge are not fully explainable, he continued, strong causal evidence exists for several mitigating factors: new technology (e.g., the efficacy of body cameras; see Williams et al., 2021), increased oversight and human resources (e.g., procedural justice training [Owens et al., 2018], reporting of incidents on paper and through third parties [Alpert and Macdonald, 2001; Ba et al., 2021], force diversity and peer effects [Ba et al., 2021], and use of past allegations [Rozema and Schanzenbach, 2019]), and new judicial structure (e.g., increased district attorney independence from law enforcement; see Stashko and Garro, 2021).

Cunningham underscored that there has never been a time in the United States when Black people have not been the primary target of state-sanctioned violence. Current use of lethal force is driven primarily by governmental and departmental decisions rather than by “intractable” structural factors; thus, he emphasized that institutions matter and expressed optimism that the public’s perception has begun to shift—political reform now seems feasible. He cautioned researchers against focusing too much on new research designs and losing sight of their research questions; context is important when thinking about the best way to answer these questions. In closing, he indicated that quasi-experimental approaches are valuable in that they allow for estimates of causal parameters that policy makers can understand; however, they are weak in that many factors have to align that lie outside of the researchers’ control (e.g., data availability, data structure, discontinuities).

QUANTITATIVE HISTORICAL DATA

Amy Kate Bailey (associate professor in the Department of Sociology at the University of Illinois Chicago) discussed the value of incorporating quantitative historical data in models of contemporary health inequities. She explained that because “remnants of the past persist” in people’s bodies and in communities, past experiences help to explain current disparities. For example, the social determinants of health paradigm includes

specific features of the built environment and the local community, which are fundamentally influenced by historical processes, and life-course and longitudinal approaches embrace links between the past and the present. Furthermore, epigenetic evidence demonstrates that human bodies determine which genes to express based on individual life circumstances and experiences, and genetic inheritances can be passed to future generations. The persistence of culture and inherited trauma also connect directly to historical measures, and biobehavioral responses demonstrate how historical patterns affect current outcomes.

Bailey described her current collaborative research endeavor, which investigates how local histories of racial violence relate to contemporary pregnancy outcomes (e.g., racial inequities in preterm births, low birth weight, and infant mortality). She noted that racial disparities in pregnancy outcomes can first be recognized on a spatial/community level. For example, Black women living in areas of racial isolation have worse pregnancy outcomes than other women (Kramer et al., 2010), and Black infant mortality profiles in states with expanded Medicaid eligibility were better than those in states that did not expand Medicaid eligibility (Bhatt and Beck-Sague, 2018). Additionally, racial economic inequality (Howell et al., 2016; Kothari et al., 2016; Ncube et al., 2016; Siddiqi et al., 2016), local racial climate (Chae et al., 2018; Orchard and Price, 2017), and the 2016 election and immigration raids (Gemmill et al., 2019; Novak, Geronimus, and Martinez-Cardoso, 2017) help to explain racial disparities in pregnancy outcomes. Racial disparities in pregnancy outcomes can also be recognized on an individual level, she continued. For instance, personal experiences with racism, such as living in a racist community, can increase the likelihood of adverse pregnancy outcomes (Bower et al., 2018).

Bailey explained that these current “patterns of injustice” that affect pregnancy outcomes connect directly to historical racial violence. Local histories of racial violence include activities of the Ku Klux Klan in the 1950s (Cunningham and Phillips, 2007; Owens et al., 2015), burnings of Black churches in the late 20th century (McAdam et al., 2013), the development of Southern “segregationist academies” (Porter et al., 2014), hate crimes reporting (King et al., 2009), Black prison admissions (Jacobs et al., 2012), use of the death penalty (Jacobs et al., 2005), White-on-Black homicide (Messner et al., 2005), and corporal punishment in public schools (Ward et al., 2021). She emphasized that enslavement in 1860 is specifically linked both to contemporary conservative political attitudes (Acharya et al., 2016) that affect policy and to racial economic disparities (O’Connell, 2012).

Thus, Bailey asserted that because history and historical contexts operate as a fundamental cause of disease (Figure 2-3), including historical indicators in analyses is essential.

She underscored that much of the quantitative historical data needed to conduct such analyses already exist—for example, Census data (aggregated data at the state and county levels, and individual data), elections data for state and national offices (down to the county level), Census data of religious bodies, data on legal executions (dating back to the colonial era), data to enable spatial linkages, measures based on vital statistics (aggregated and individual), school enrollment data, and limited administrative record linkages for individuals across the life course. However, she described several challenges in using these data, including an uneven availability of data over space and time, owing to the slow uptake of vital record-keeping across the United States; shifting administrative boundaries that make it difficult to link contemporary measures to the historical context; gaps in historical records that are not randomly distributed (e.g., Census undercounts of marginalized groups); and difficulty in locating and gaining rights to use certain historical data. Bailey highlighted four ways to address these challenges as new historical data continue to be gathered:

1. Additional spatial tools to connect data across time periods and identify spatial relationships within local communities (below the county level);
2. Expanded sources of administratively linked multigenerational records;
3. Better data on local, regional, statewide, and national policies with a focus on their racialized and gendered implications; and
4. Better measures of civil society (e.g., configurations of local business communities).

Bailey emphasized that historical data at both the spatial/community and individual levels have the potential to advance research in health and social equity. For example, she explained that historical data could be used to improve environmental equity. Health researchers already leverage data on historical environmental contaminants and locations of toxic releases

but could incorporate additional data from environmental disasters, epidemics, and disease outbreaks to better understand how exposures vary by community. Record linkages then enable the mapping of residential trajectories and related environmental exposures across the life course, including multigenerational exposures and effects. Historical data could also be used to study how inequality and access to opportunity structures affect social and health disparities. Public policies that shaped access to certain resources, the structure of the local labor market, and local school funding could all be analyzed; furthermore, data on multigenerational program participation (e.g., GI bill benefits were mostly for White men) and occupational trajectories could reveal key insights. Lastly, historical data on the social environment could improve understanding of structural racism—for example, data on civil society organizations that supported White supremacy could affect the context of race, gender, and social class; data on voter behavior and suppression could illuminate political power structures; and data on health care structures (e.g., segregated hospitals) could provide insight into local social dynamics. Vital statistics could also be studied to better understand gender relations and women’s access to power, both of which are critical to child development and survival.

Bailey identified several challenges with expanded access to such historical data. She cautioned researchers about record “survivability”; not everything recorded survived, and not everything of interest was recorded. Researchers would also benefit from considering the power dynamics associated with what was recorded, by whom, and for what purpose, she continued. Knowing what data are available and gaining permission to access them can be difficult, and tradeoffs between time/labor inputs and accuracy are unavoidable. She described current initiatives to strengthen the use of quantitative historical data, including efforts to digitize historical records, prepare digitized records for quantitative analyses, and make existing records more widely available. For example, the Library of Congress, Zooniverse, the Mellon Foundation, and Humanities without Walls are all engaged in efforts to expand digital access to historical data.

Bailey reiterated that communities and individuals are shaped by their past and that of their ancestors; therefore, researchers could use multiple methodological approaches and forms of data to identify the root causes of social and health inequities. She stressed that accounting for contemporary structural racism without considering the historical legacy that created it is an impossible task.

DATA FOR UNDERSTUDIED POPULATIONS

Desi Small-Rodriguez (assistant professor of sociology and American Indian studies at the University of California, Los Angeles) explained that

many populations—at times demographically categorized as “something else”—remain understudied. These populations are often described as difficult to reach and count, hidden, vulnerable, underrepresented, and underserved. However, she emphasized the value of changing these perspectives on understudied populations and the way that they are researched. She asserted that understudied populations are not hidden; rather, they have been historically marginalized, under-resourced, systematically excluded, and deliberately erased. For example, the contemporary racial/ethnic category “American Indian” is a colonizer’s term used to identify Indigenous peoples; however, a panethnic American Indian identity did not exist in the precolonial era (Snipp, 1989). Such “colonial myths” persist as mechanisms of Indigenous erasure, she continued, and the intentional effort to erase Indigenous lands and people extends into the data, obscuring inequalities in health care, housing, and education.

Small-Rodriguez stated that researchers and policy makers are limited in their understanding of Indigenous peoples by population definitions that serve the needs of the nation, not the needs of Indigenous communities. For example, government agencies are guided by the following federal definition of American Indian or Alaska Native (AIAN): “a person having origins in any of the original peoples of North and South America (including Central America) and who maintains tribal affiliation or community attachment” (Office of Management and Budget, 1997). According to the 2020 Census, approximately 9.2 million people in the United States identified as AIAN; however, Small-Rodriguez noted that no documented national life expectancy existed for AIAN until 2021, owing to widespread data erasure. Furthermore, at only 71.8 years, this life expectancy is the lowest of any racial/ethnic group—compared with 81.9 years for Hispanic people, 78.8 years for White people, and 74.8 years for Black people (Arias et al, 2021).

Small-Rodriguez suggested that focusing on “peoples” instead of on “populations” would enable movement beyond this type of data erasure. The definition of Indigenous peoplehood—“interlocking concepts of sacred history, ceremonial cycles, language, and ancestral homelands” (Corntassel, 2003)—extends beyond a simple racial category. People data, then, refers to the sovereign nations, which include 574 federally recognized tribes, 326 reservations, and 56 million acres of trust land. To illustrate the complexity of the contemporary Indigenous experience, she explained that blood has become a “sociopolitical and pseudobiological construct of collective identity,” and “racial logics continue to distort kinship systems” (see Rodriguez-Lonebear, 2021). She reiterated that the way in which Indigenous peoples have been racialized and erased by federal systems affects data availability and data dependency, and this “assimilative effort to erase” occurs through this “mechanism of blood.”

Small-Rodriguez observed that determining who counts as AIAN is further complicated by several different boundaries within the population (Liebler, 2018). For instance, the U.S. Census data exclude one-third of the population of the Northern Cheyenne Nation. Indigenous peoples are thus working to change this narrative of data erasure, stressing both that they are still here and that they are data experts. She emphasized that Indigenous peoples consider data as sacred to sovereignty, and without sovereignty, equity and justice cannot be realized. Indigenous data sovereignty is “the right of Indigenous peoples and nations to govern the collection, ownership, and application of their own data.”¹ Indigenous peoples have thus begun to reclaim data sovereignty by decolonizing data, moving from data dependency to control of data by Indigenous peoples and for Indigenous people in all aspects, including collection, analysis, reporting, and storage.

MACHINE LEARNING

Ziad Obermeyer (Blue Cross of California distinguished associate professor of health policy and management at the University of California, Berkeley, School of Public Health) highlighted the danger of algorithms that reproduce and scale up bias in health care, criminal justice, finance, and education. However, some of these dangers can be minimized, he explained, and algorithms could play an important role in fighting bias. To demonstrate the potential for effective algorithm use in the health care space, Obermeyer presented recent research conducted with his colleagues (Pierson et al., 2021) on leveraging algorithms to reduce unexplained pain disparities in underserved populations.

Obermeyer indicated that pain is distributed unequally throughout society and concentrated in the most disadvantaged people—survey evidence reveals that non-White patients experience approximately twice the amount of pain as White patients. In their research, Pierson and colleagues (2021) focused specifically on disparities in knee pain. They noted that the cause for these disparities is far more complex than Black people simply having a higher incidence of osteoarthritis in the knee—Black, lower-income, and lower-education patients reported more knee pain, despite the similarity in the degree of osteoarthritis revealed in the X-rays of their knees and those of White patients. This raised an important question: if the knee is not causing the pain, what is?

Reflecting on various possible explanations for this pain gap from existing scholarship, Obermeyer stated that when experiencing similar stimuli, people with higher levels of stress often have more pain than people with

___________________

¹ See https://nni.arizona.edu/programs-projects/policy-analysis-research/indigenous-data-sovereignty-and-governance

lower levels of stress. For example, anxiety and depression can manifest as pain, and, in some communities, demands for attention toward other aspects of life might decrease one’s ability to cope with pain. Furthermore, he noted that the medical system provides less access to pain management therapies for some communities than for others.

Obermeyer described the typical process used to address pain: a patient describes his or her pain to a doctor, who then orders an X-ray. The doctor reviews the X-ray’s appearance and “grades” it to determine disease severity based on certain criteria, most of which were developed in 1957. These grading scales were developed on one population, homogeneous in race and sex, and may not generalize to the populations seeking care today, Obermeyer asserted. As a result, when human radiologists determine that a knee is not diseased, they might overlook the real causes of knee pain in disadvantaged groups. Although measuring disease severity might seem like the perfect task for an algorithm, he continued, algorithms are usually trained to match human performance (i.e., the radiologist’s review of the X-ray for disease severity), which is not the desired outcome in this case, because something would still be missing.

Therefore, Obermeyer advocated for the algorithm to be trained differently—to listen to the patient (instead of learning from the radiologist) and predict the degree of pain that will be reported based on a given X-ray. He mentioned that data that link X-rays to radiologists’ interpretations are abundant, but data on the link between X-rays and patient pain experience are much sparser. However, once such data are available, measuring disease severity could become a straightforward machine-learning problem. Thus, an algorithm could help confirm that if the pain is predictable from the X-ray of the knee, the pain is coming from the knee, and if it is not, other causes could be explored. Pierson and colleagues (2021) found that the algorithm explains nearly half of the pain gap between Black and White patients, which is far more than what radiologists account for in their interpretations. Obermeyer pointed out that decision-making currently depends on radiologists’ perspectives. Guidelines for receiving a life-transforming knee replacement, for example, are based on both the severity of knee pain and the severity of knee disease; by inserting the algorithm’s prediction of severity to make this decision instead of relying on the radiologist’s assessment, the fraction of Black people eligible for knee replacement doubled.

Obermeyer underscored that algorithms have the potential to perform better than humans instead of reproducing their errors and biases. To achieve this, he continued, algorithms in the field of medicine should learn from nature (i.e., patient experiences and health outcomes). However, because data on patient outcomes and experiences are siloed, and the infrastructure to connect good researchers with good data does not currently

exist, researchers are working toward a solution to move the field forward. For example, Nightingale Open Science² works with health systems, companies, and governments to build data infrastructure and curate datasets around unsolved medical problems, with a focus on high-priority problems and populations—these deidentified datasets are then accessible to nonprofit researchers on a cloud platform for free.

REFLECTIONS AND DISCUSSION

Serving as the first discussant for this session, Takeuchi commended the speakers for their efforts to include new designs, methods, and populations in the study of structural racism, and expressed his support for the development of theories for structural racism that align with methods to best uncover drivers of health outcomes. He cautioned that without such theoretical frameworks, structural racism research could become a cottage industry.

Reflecting on recent research endeavors, Takeuchi noted that of all of the projects funded by the National Institutes of Health between 2018 and 2022, 41,784 focused on health disparities. However, less than two percent of those projects mentioned racism, and only about 25 projects included a theory of racism. Less than 20 of these projects used historical, machine learning, or experimental approaches, suggesting that these methods are significantly underused in the study of systemic racism and health. Furthermore, Takeuchi reported, only a small number of the funded projects on racism focused on Indigenous peoples, Asian Americans, Native Hawaiians, and Pacific Islanders. He also reflected specifically on an article published in 1970 that said that Black, Indigenous, and other people of color were less likely than White people to seek care for mental health issues, and when they did, they received lower-quality care and experienced poor health outcomes. Despite significant improvements in mental health care and increases in the number of people insured, the same pattern of findings persists in 2022. He asserted that combining new theories and new methods with conventional approaches could begin to address these issues more systematically.

Opening the general discussion, Frank Edwards (assistant professor of criminal justice at Rutgers University) expressed his interest in learning more about the gap between Census data and tribal data described by Small-Rodriguez. He wondered if the source of the bias has been decomposed (i.e., who was missing from the Census data but captured in the tribal data?) and how contemporary data could be used alongside historical data to understand event exposure and population change over time. Small-Rodriguez suggested the need for increased linkages between tribal data

___________________

² See www.nightingalescience.org

and other existing data, and she encouraged demographers and researchers to support this work, with tribes in control and tribal leaders cogoverning the process. For example, uncovering why younger populations from the Northern Cheyenne population were overrepresented in the Census data as compared with the tribal data is important. She noted that tribal nations are acutely aware that they are undercounted in the Census, which underscores the need to continue to leverage the data that tribes are collecting. Seth Sanders (Ronald Ehrenberg professor of economics at Cornell University) agreed that the tribes are in the best position to generate the data of most interest to their communities. At the same time, he continued, because comparative work is important for science, parallel data collected outside of the tribes would be useful to address certain questions. He considered the use of panel data for this purpose but wondered who would have access to those data. He advocated for the use of models that simultaneously benefit science, give control to sovereign nations, and help to develop the human capital of tribes. Small-Rodriguez emphasized the value of creating a tribal data standard in the United States, which would enable increased opportunities with comparative data. Furthermore, she pointed out that the United States has never conducted a national survey of Indigenous peoples, unlike its peer countries.

Reflecting on the Latin American experience in particular, Flores remarked that the process of defining who is Indigenous is complicated, which has implications for measures of inequality. He asked about the gold standard for the measurement of inequality in Native American populations in the United States. Small-Rodriguez explained that identifying a gold standard is not feasible since each tribe is its own nation, and the heterogeneity within a tribal population can be even greater than that between populations. She said that a community partnership would be beneficial to study the comparisons within and between tribes.

A participant pointed out that community-engaged primary data collection can be costly and time consuming but wondered whether using secondary data, which are easier to access but lack context, removes a researcher too much from the communities of interest. Flores explained that he first develops a research question, collects primary data through several methods (e.g., focus groups and personal interviews), and designs his own experiments, but he uses secondary data to corroborate his observations. Cunningham said that his work relies on secondary data, but he also makes an effort to talk to police officers to engage different perspectives: their opinions might reveal different takeaways that could be placed in historical context. Bailey commented that her work is almost exclusively in the analysis of secondary data, but she personally builds many of those datasets from historical archival data. She suggested collaborating with students, if they are available, to build these datasets as well as to embed contemporary out-

comes within a historical context to help avoid disengagement. Obermeyer added that his inspiration comes from clinical work, including difficult decision-making about patient care in the emergency room. However, that level of engagement does not have to come from direct practice; ethical research includes talking to people and investing time to amass information about institutions to identify quasi-experimental variation. A self-described “data rebuilder,” Small-Rodriguez said that she uses secondary data and identifies opportunities to creatively link existing data for populations that remain understudied. She emphasized that secondary research does indeed help to support community needs.

A participant posed a question about whether the term stereotype has only negative connotations. Flores responded that stereotypes are complex because they are explanations that one group provides about another; they could be positive, negative, or ambiguous. For example, being “stubborn” could be either positive or negative, because the meaning of a stereotype is based on a particular social context. However, he pointed out that some of the methodologies used for stereotype research are not yet designed to capture meaning.

Trevon Logan (workshop planning committee member and Hazel C. Youngberg distinguished professor of economics at The Ohio State University) added that stereotypes about White people have to be “positive,” owing to the structure of racialization, and researchers have a responsibility to interrogate this language. He also highlighted the importance of key language distinctions related to structural racism; for example, mistrust is based on intuition, and distrust is based on experiences. He stressed the value of choosing words carefully when engaging in structural racism research—for example, distrust better frames conversations about Native populations’ feelings toward the dominant population than mistrust. Reflecting on the connection between language and the use of algorithmic approaches to understand pain experiences and socioeconomic outcomes, he observed that interrogating science while using the language of science is difficult for inherently racialized processes.

A participant wondered whether the U.S. Food and Drug Administration could play a regulatory role in ensuring that algorithms eliminate rather than replicate biases. Obermeyer highlighted his work auditing algorithms that are widely used in the health care space and finding a significant amount of racial bias (e.g., algorithms that deprioritize care for sicker Black patients in favor of healthier White patients). He described this as a significant market failure, in that this bias could have been caught before the software was implemented. Thus, he championed the role of regulation; the next step is to determine what to regulate. He mentioned that drug regulation offers a useful precedent for algorithm regulation: one first defines the outcome before obtaining drug approval. Similarly, determin-

ing the information that the algorithm should produce would create the standard by which the algorithm would be held accountable for biases and for accuracy (see Bembeneck et al., 2021). Lee inquired about the potential for machine learning or other approaches to capture more informal processes by which people’s needs are not being addressed—for example, pain patients who are not approved to have magnetic resonance imaging (MRI) and so are not included in the training dataset for the algorithm. Obermeyer responded that an anterior cruciate ligament tear, for example, can only be observed in an MRI; thus, prediction is based on those who have access to an MRI—that is, the algorithm can predict a biological event but only conditional on having the MRI. He cautioned against the practice of infusing disparities to access into the algorithms and emphasized that steps can be taken so that the algorithm can generalize to people who have not had the MRI. He encouraged a careful, detective-like approach to addressing disparities that surface in datasets, which could lead to improving the performance of algorithms for health care and redistributing resources to patients in need. In some settings, however, such as criminal justice, he asserted that no such workarounds exist because the data are too biased, and algorithms might not be the best approach.

MIXED METHODS APPROACHES

Paris “AJ” Adkins-Jackson (assistant professor in the departments of Epidemiology and Sociomedical Sciences in the Mailman School of Public Health at Columbia University) explained that racism assigns value and discriminates (Jones, 2000), and that racism has multiple levels—for example, interpersonal, internalized, institutional, intraorganizational, and extraorganizational (Griffith et al., 2007; Jones et al., 2019)—and multiple dimensions—for example, via residential segregation, access to health care, and civics (Bailey et al., 2017). Structural racism, she continued, occurs across the life course, influencing educational, social, and economic opportunities, as well as leading to immediate (e.g., injury) and cumulative (e.g., multimorbidity) health effects (Glymour and Manly, 2008).

Adkins-Jackson presented strategies for combatting structural racism using mixed methods. She encouraged researchers to pair methodological tools, such as surveys, observations, biomarkers, ethnologies, social media, interviews, photovoice, archives, and ethnographies, to better understand the complexity of structural racism. The use of mixed methods can be done sequentially (i.e., quantitative before qualitative) or concurrently (i.e., qualitative plus quantitative) to explicate a phenomenon about which there is limited information, to corroborate existing evidence, or to dispute existing evidence. Mixed methods designs can be correlational/causal, experimental, phenomenological, or comparative, or can take the form of case study or

grounded theory (Johnson et al., 2007). She described case studies and comparative analyses as particularly effective approaches to leverage multiple methods for the study of structural racism.

Adkins-Jackson offered two examples of recent effective mixed methods research. First, she highlighted the work of Ashley Gripper (Drexel University), who used grounded theory and the sequential mixed methods of an interview followed by a survey—which is a common approach in the field of psychology—to explicate the experiences of urban farmers and their understanding of environmental justice and health. Second, she described the work of Brittney Butler (Harvard University), who created a comparative study using the concurrent mixed methods approach of collecting survey data, observation data, and interview data to corroborate the impact of anti-Black racism on birthing individuals across different datasets.

Adkins-Jackson cautioned researchers against becoming nonreflexive scientists, who are unaware of how much harm they create by preselecting methods—a decision that immediately introduces bias into a study. To avoid introducing bias, she encouraged researchers to recognize that structural racism is experienced by people; it is not merely a scientific exploration. She also stressed that structural racism researchers take care to avoid practicing racism themselves: both collaborators’ and participants’ contributions to the research have value. Thus, true reflexive scientists practice self-reflexivity (Ford and Arhihenbuwa, 2010; Hardeman and Karbeah, 2020), partner with the community (Leung et al., 2004), read and cite scholars from historically marginalized communities (Onwuegbuzie and Collins, 2007), collaborate with interdisciplinary colleagues, and examine methodological frameworks.

Emphasizing the value of theoretical and methodological frameworks in particular, Adkins-Jackson explained that a theory shapes the relationship between an exposure and an outcome (e.g., critical race theory, intersectionality [Crenshaw et al., 1996], and fundamental cause theory [Phelan and Link, 2013, 2015]). A methodological framework, then, guides the methods and analyses based on the existing research question. Reflecting on the concepts that support perspectives about phenomena, she observed that many people have a positivist perspective—that is, a fixed, objective reality that can be understood through logic and reasoning. However, she posited that if knowledge is cocreated, multiple truths will arise with no fixed destination; therefore, a social constructivist perspective facilitates an understanding of the meaning derived from these multiple truths, as well as a particular phenomenon’s impact on a community, which is the desired measurement in structural racism research.

As an example of using mixed methods to examine the impact of structural racism on a community, she described the work of the late Candice Rice (University of California, San Diego). Rice used concurrently observation, archives, and autoethnography, as well as a case study on the impact

on Black mothers of crime and substance abuse policies enacted after the 1992 Los Angeles uprising, to dispute the idea that poor behaviors create these policies; rather, these policies lead to poor behaviors. Furthermore, the data from this and other mixed methods research revealed that high stress leads to poor health for Black women in particular (see also Adkins-Jackson et al., 2019). The women’s specific narratives illuminated a form of structural gendered racism, pointing to the White supremacist, patriarchal, heteronormative, able-centric system as a cause of poor health for Black women: a system that impacts the larger Black communities for which “Black women are socialized to care” (see Adkins-Jackson et al., 2022; Laster Pirtle and Wright, 2021).

PLACE-BASED APPROACHES

Michelle Johnson-Jennings (professor at the University of Washington and director of the Environmentally-based Health & Land-based Healing Division at the Indigenous Wellness Research Institute) described an Indigenous perspective on place, and she depicted the use of Indigenous land-based healing to improve community health and well-being and to combat structural racism.

Johnson-Jennings remarked that for the Choctaw people in particular, land is central to health and well-being. In order to illuminate this Indigenous connection to place, she compared the health frameworks of Indigenous peoples with those of people who practice Western medicine. The Indigenous framework for health and healing outlines that the ancestors provide instructions for interacting with the land and offer gifts of strength and health; relationships with nature are important to preserving health, and illness arises from an imbalance in these relationships; and healing is space-, place-, and community-oriented. In essence, she continued, Indigenous people believe that they can be healthy only if their land is healthy. This framework contrasts that of Western health, in which ancestors pass on their diseases to the next generation; mind, body, spirit, and nature are disconnected; illness arises from microagents; and land has no place or relationship in healing.

Acknowledging the presence of Indigenous peoples, Johnson-Jennings explained, demands recognition of the objectification, enslavement, marginalization, oppression, and genocide they have experienced. The erasure of “the problem of Indigenous people” has been both systematic and purposeful: historical attacks (e.g., massacres, warfare, illegal sterilization), removals (e.g., illegal healing practices, starvation, land allotment), and assimilation policies (e.g., boarding schools, nutrition experiments, and animal husbandry) are reflected in the structural racism experienced by Indigenous communities today through limited access to health care, self--

determination restrictions, and environmental pollutants on reservations. She provided several examples of these systematic methods to eradicate Indigenous place and being, including body objectification in the 17th and 18th centuries; a movement toward eugenics with feeblemindedness starting in the 19th century; and the development of IQ tests with Indigenous people as the deficit to justify both their removal and their sterilization (see, e.g., Fitzgerald and Ludeman, 1926; Garth, 1931). According to Johnson-Jennings, 25–50 percent of all AIAN children were removed from their homes and placed in foster care or boarding schools in the 1940s, and at least 25 percent of Native women of childbearing age across the United States were sterilized by 1976 (see Lawrence, 2000).

Johnson-Jennings indicated that these systematic, place-based attacks on Indigenous people continued throughout the 20th century. By either removing Indigenous people from their sacred land or confining them to certain places, the culture continued to be erased. Place for Indigenous people thus became defined by vulnerability, discrimination, and a lack of control, and she asserted that these colonial narratives of the past continue to have negative impacts on Native people (e.g., gaps in educational opportunities and issues with law enforcement). The historical trauma of this structural racism within Indigenous communities can be understood as the “cumulative emotional and psychological wounding over the lifespan and across generations, emanating from massive group trauma experiences” (see Brave Heart, 1998, 1999, 2003).

Johnson-Jennings underscored that although Indigenous people are working to recover from this historical trauma, it compounds present trauma in the form of increased risk of disease susceptibility. She stressed that this context is particularly important for researchers who want to better understand Indigenous health issues. For example, racial discrimination can increase pain and can lead to increased rates of smoking among individuals in the AIAN community (Johnson-Jennings et al., 2014). However, despite this history of suffering, she emphasized that Indigenous communities are resilient. This trauma can be transformed by fostering supportive relationships and engaging in the traditional Indigenous land-based practices for innovative healing—that is, “(re)connecting to the land and centering the land in order to conduct healing, or a health intervention” (Johnson-Jennings et al., 2020).

Johnson-Jennings also accentuated the value of decolonizing research by recovering subjugated knowledges and documenting social injustice. Doing so creates a voice for the silenced and challenges racism, colonialism, and oppression (Smith, 2021). Most importantly, this decolonizing of health research helps to create a place where Indigenous peoples can survive and thrive, “guided by their ancestral knowledges and practices centered upon the lands” (see Johnson-Jennings et al., 2019, 2020).

NOVEL APPROACHES TO SURVEY DATA

Courtney Boen (assistant professor of sociology at the University of Pennsylvania) provided an overview of how longitudinal social surveys can be used to explore the structural, institutional, and relational processes that produce and maintain racialized inequities in health and other outcomes, with particular attention to how better data infrastructure and methodological approaches could center equity and justice. She noted that these types of surveys (e.g., the National Longitudinal Study of Adolescent to Adult Health; National Social Life, Health, and Aging Project; Fragile Families and Child Wellbeing Study) engage a life-course perspective, which considers health and aging as lifelong processes, incorporates the timing and duration of exposures, evaluates how the past shapes the future, and links lives to better illuminate the emergence of inequities across the life span.

Boen pointed out, however, that the use of social surveys to explore racialized inequities is not without theoretical, epistemological, and methodological challenges. First, race is typically operationalized in survey research as a static, individual-level trait rather than understood as a proxy for complex, dynamic, relational processes of historical and contemporary racialization and racism. Reflecting on the ideas of sociologist Tukufu Zuberi, she explained that racial categories continue to be used to justify structural racism and White supremacy, but much survey research still includes a variable for race without consideration for what that measure represents. Second, she observed that survey research often promotes methodological individualism, which emphasizes the study of individuals or groups who experience oppression and discrimination rather than the study of the racist systems that oppress, discriminate, and create inequities. Third, although conventional regression estimators are a key aspect of survey research, they can be misaligned with relational theories of race and racism.

Boen remarked that the first step in improving the use of survey research to understand the drivers of racialized health inequities is to have better data to capture the structural and institutional processes that produce these inequities, as well as better data infrastructure. She encouraged researchers to recognize that individual-level measures are imperfect and imprecise proxies for complex systems of social relations (e.g., race as a function of historical and contemporary processes of racialization and racism; and socioeconomic status as the product of exploitation, theft, and extraction). This awareness can shift the research focus from the individuals who experience harm to the systems that create the harm. However, she underscored that such a shift also demands data linkages to illuminate those processes that maintain structural racism. These data linkages could be enabled with increased funding and publishing incentives to share policy, institutional, and contextual data as well as increased priorities to reduce

administrative barriers. For example, Boen described current research that links county-level data on Immigration and Customs Enforcement with state immigration policy data and geocoded Health and Retirement Study data to better understand how changes in county-level immigration enforcement and state law affect individual- and group-level changes in health risk across adulthood (Boen et al., forthcoming). Legal violence against immigrants, Boen and colleagues argue, has implications for the production of racialized inequities in health: increases in county-level immigration enforcement led to increased health risks among foreign-born Hispanic adults in particular. In a related study, state-level immigration policy data were linked to data from the National Agricultural Workers Survey and revealed that restrictive state-level immigration policies shape racialized legal status inequities and health care use among U.S. agricultural workers. Non-White workers experienced the most significant barriers to health care (e.g., high cost, lack of information, lack of access to transportation, fear of legal status being discovered) after these policies were implemented (Schut and Boen, forthcoming). Boen summarized that both studies highlight how data linkages can be used effectively to explore the institutional drivers of population health gaps.

Boen discussed another common approach used to measure structural racism that focuses solely on markers of disparities and discrimination in particular institutional domains, which can conceal other forms and consequences of racial violence, racist social control, and structural racism. Thus, she encouraged researchers to also consider how overall levels of exploitation, violence, exclusion, and social control can reflect the structural racism that affects social inequities. Several systems (e.g., legal, health care, and welfare) emerging from historical and contemporary racism attempt to maintain the racial order; Boen asserted that researchers’ methods to measure structural racism should reflect this reality.

Boen explained that a second step in improving the use of survey research for understanding the drivers of racialized inequities is to develop methods that better align with critical, dynamic, and relational theories of race and racism. She noted that social survey research relies heavily on conventional regression estimators and focuses on identifying the causal effect of discrimination; however, this approach is limited. A question arises about the possibility of separating “nonrace” variables from “race” (i.e., proxies for processes of racialization and structural and institutional racism)—a separation she portrayed as untenable. This challenge is compounded with longitudinal data, for which conventional regression assumes an absence of time-varying relationships among the variables that is inconsistent with dynamic theories of race and racialization. In an attempt to overcome the limitations of conventional regression in identifying the life-course social exposures of racialized inequities, Graetz and colleagues (2022) suggest

that the parametric G-formula be used to better theorize and model mediators as connected to a system of historical and contemporary racism and racialization—that is, instead of separating variables by “nonrace” and “race,” everything is treated as a mediator (see also Aislinn Bohren et al., 2022).

In closing, Boen remarked that identifying the dynamic processes that enable the production of racialized inequities in health and other outcomes across the life course and across generations has the potential to advance science, intervention, and policy; social surveys are well-suited for this work but only if the data and methods match the theory. She asserted that data and methods that ignore structural and institutional processes that generate inequities are insufficient, and better data linkages and expansion beyond approaches that separate “race” from “nonrace” are critical. She emphasized that causal inference research using survey data does not have to be used to identify marginal effects that inform well-defined interventions (Schwartz et al., 2016). The next step, she continued, is to incorporate methods for modeling the historical and dynamic life-course processes that maintain racist social systems and to build data infrastructure that captures the complex processes that create health inequities; these actions could support more holistic ideas about how racialized inequities in health and other outcomes have been produced and move society closer to achieving equity and justice.

NOVEL APPROACHES TO ADMINISTRATIVE AND CROWD-SOURCED DATA

Frank Edwards (assistant professor of criminal justice at Rutgers University) described his interest in using descriptive methods and novel data to quantify the social distribution of state violence in both historical and contemporary ways. He explained that novel data not only present new opportunities to measure state violence but also reveal the limits of official statistics. Currently, although public data on the activities of violent U.S. state institutions for the purposes of research and evaluation are often sparse and of low quality because they rely on self-reporting by law enforcement agencies, internal data holdings for the purposes of surveillance and control are significant. He stressed, however, that this data opacity is a feature rather than a flaw of public systems that administer state violence. For example, states collect information to enhance their operations; state agencies that engage in violence collect data to target people and increase power. He cautioned researchers to be mindful of these types of data-generating processes.

Edwards emphasized that no official source of data on police violence in the United States exists. He described Fatal Encounters—an unofficial source of data on police violence collected by a single journalist—which

uses searchable news archives to create a streaming dataset that captures all deaths occurring during contact with law enforcement prior to entrance into a correctional institution. These data reveal that 3.2 people were killed per day by police use of force in the United States between 2013 and 2021. The National Vital Statistics System from the Centers for Disease Control and Prevention, however, reported only 1.6 people killed per day during the same time frame (Figure 2-4). Edwards explained that this discrepancy highlights an issue in the data-generating process: underreporting. For instance, if the cause of death was not labeled by the medical examiner as “by law enforcement,” it is not recorded as such.

Edwards noted that this example demonstrates the value of unofficial data sources; however, he encouraged researchers to use novel data cautiously and critically, as unofficial data are not without challenges and biases. For instance, searchable local news data could have temporal biases, especially given that online news was not fully accessible across the nation until 2010. As a second example of the use of novel data, Edwards described research with colleague Sadaf Hashimi on the age-specific risk of police use of force in New Jersey. The data from legally mandated force reports collected by a local news organization, NJ Direct, appear to

show high levels of force against young Black men; however, the data are incomplete because the levels of reporting compliance vary significantly across New Jersey, likely owing to variation in policy and training across the state’s 450 police departments. Thus, he explained that because underreporting leads to missing data, these estimates are likely undercounts and would be considered lower bounds.

Edwards mentioned that administrative data, another type of novel data, are generated routinely for the administration of various state programs and collected to meet organizational needs and for legal compliance; they are not produced with the research community in mind. Looking at administrative data on prenatal substance exposure screening from 2010 through 2019 from the National Child Abuse and Neglect Data System, Edwards and his colleagues found that in Minnesota, for example, American Indian infants were 15 times more likely than White infants to be reported to child protection agencies with infant or prenatal substance exposure. This type of study of administrative data is important, he continued, but despite federal requirements to collect them, these data are missing from at least 20 states. He reiterated that researchers should interpret the data with a critical eye amid issues of nonreporting and spatial bias.

Reflecting on the future, Edwards emphasized that official data collection is likely not the best path forward owing to federalism and the diffusion of oversight (i.e., with more than 10,000 law enforcement agencies in the United States, much data reporting is voluntary), as well as to issues of historical and contemporary racial politics and data asymmetry. He underscored the value of building infrastructure for unofficial data collection, especially to track state violence. He added that because news data are powerful but limited, funders and universities could commit to long-term streaming data collection, which is relatively inexpensive yet invaluable. Administrative data on state violence in particular have several challenges: agency administrative data can be of high quality but are often limited in scope geographically, and agency involvement in approval processes limits the scope of critical research when researchers are not granted access to datasets. He asserted that new processes for managing data access and linking administrative datasets that would be independent of state agencies and administered more impartially could enable critical research on structural racism.

REFLECTIONS AND DISCUSSION

Serving as the second discussant for this session, Flores observed that all of the speakers highlighted the complexity of the study of structural racism, which has medical, social, educational, psychological, and historical dimensions. The systems of racial oppression—for example, environmental pollution and land impacts, immigration policies, and criminal

justice issues—vary across populations. He reiterated that understanding these systems is essential to appreciate how structural racism shapes people’s lives. He also reflected on the modeling challenges described by some of the session’s speakers and agreed that researchers would benefit from recognizing that race is fluid and shaped by the context and history of a place, with different consequences for different individuals. Since some methods can produce biases, he continued, listening to narratives about people’s experiences with discrimination and oppression and using methodologies that incorporate this community knowledge is essential. Flores noted that speakers also stressed the importance of incorporating and linking many types of data to match the complexity of structural racism, although challenges arise here too. For example, because administrative data likely underreport oppression and violence, other sources that complement and interrogate these official sources of data are critical. With all of these complex challenges in mind, he encouraged researchers to reflect on a key question: how can systems of knowledge be created that develop more insight on structural racism?

A participant inquired about barriers to forming interdisciplinary teams for structural racism research, as well as strategies to overcome them. Adkins-Jackson encouraged researchers to think more broadly about using science to achieve justice in the real world. That mindset could lead to the creation of more balanced teams to move forward with collaborative research. Edwards mentioned existing professional barriers to justice-focused work, and he suggested that quantitative researchers would benefit from greater humility and openness toward other paths to knowledge. Johnson-Jennings advocated for the development of research protocols in collaboration with the community of interest, which prioritizes a focus on science for the common good that could impact future generations positively. Boen pointed out that disciplinary norms related to publishing, funding, and career trajectories present challenges for interdisciplinary research; however, the best science happens “on the fringes” and at the intersections of disciplines. She urged researchers to leverage tools from several disciplines to examine complex social problems and to overcome barriers to achieving justice- and equity-oriented science. Another participant wondered about the impacts of these contrasting disciplinary norms, especially when funders assume the role of gatekeepers that give preference to short products that do not have the space to incorporate the voices of communities. Boen replied that although interdisciplinary teams continue to face challenges, funders are now more accepting of the role that these teams play in addressing complex issues such as structural racism. Institutional supports help facilitate these collaborations, she continued, but they can still be difficult, especially when a decision has to be made about where to publish the research. She encouraged researchers to be honest and transparent about expectations for

authorship and funding at the start of any collaboration. Adkins-Jackson emphasized the need to move beyond viewing only publications and career trajectories as the end products of collaboration.

A participant inquired about successful examples of incorporating community voices into research. Johnson-Jennings described her experiences working with the Choctaw Nation on land-based healing to address the historical trauma of structural racism and related health disparities. She said that they avoid “wallowing in the trauma and the drama” of the past, and instead study the ancestors’ archival narratives with a focus on healing and resilience for future generations. She advocated for grassroots initiatives that center on love from the community and its needs, instead of on the trauma, with community members contributing throughout the process.

Logan asked how researchers could think about data specifically as a tool for narrative. Edwards suggested that researchers observe movements on the ground that are striving for justice and consider how data could be used to support those stories. In other words, he continued, researchers could take different forms of knowledge, translate it into different domains, and search in new ways for known concepts. Johnson-Jennings added that data should be investigated in relation to how they have reinforced structural racism in communities—data reframing and renarration are often crucial for this task. Boen mentioned that although quantitative scholars often focus on complex models of causal inference for validation, great power can be found by linking the descriptive analysis of stories to rich, critical theory.

Another participant asked how best to evaluate the reliability of data collected by private entities. Edwards noted that all data analysis demands strong detective work—that is, thinking carefully and critically about data-generating processes as well as data quality (e.g., using reasonable priors for assessment). Adkins-Jackson championed Cunningham’s previously discussed strategy of asking the community of interest and other stakeholders for their opinions on research findings as one effective way to test the validity and reliability of data.