INTRODUCTION
Lethal and unaccountable police violence against Black civilians is one of the defining political issues of the twenty-first century in the United States. The past decade has witnessed repeated outbreaks of large-scale social protest following the killing of unarmed Black civilians by police officers. The 2014 Ferguson uprising propelled prior social media activism using the hashtag #BlackLivesMatter into a mass protest movement holding street demonstrations throughout the nation. The recurrence of social protest following continued incidents of police violence against Black civilians since 2014 has rendered Black Lives Matter (BLM) a leading proponent of civil rights, racial justice, and police reform. More recently, the eruption of protest following the police killing of George Floyd in May 2020 stands as the largest episode of social protest in both the catalogue of the BLM movement and the longer history of Black resistance against dehumanization and state violence in the U.S. (Lebron Reference Lebron2020).
The BLM movement has reinvigorated interest in political science in studying social protest (APSR Editors 2020), with one long-standing line of inquiry being assessment of the success of protest in exerting desired effects on public opinion (Lee Reference Lee2002; Mazumder Reference Mazumder2018; Wasow Reference Wasow2020). Applied to the BLM movement and its focus on systemic and unaccountable police violence against Black Americans, this study asks whether or not instances of large-scale social protest against police violence shift public attitudes toward law enforcement and elevate awareness of racial injustice? According to prominent reports, public confidence in the police notably dropped following mass protest over the police killings of Eric Garner and Michael Brown in 2014 (Drake Reference Drake2014; Jones Reference Jones2015). Such reports, however, are descriptive in nature and are limited by relying on snapshots of public opinion provided by national surveys conducted several months (or even years) before and after the 2014 protests, leaving it open to question whether or not such protest events actually cause immediate or sustained shifts in public opinion, how large the effects are, and among whom attitudes change.
Theories of activated opinion suggest that minority-led protest can serve as a grassroots “bottom-up” factor that mobilizes liberal shifts in public opinion on racial issues (Lee Reference Lee2002). Complementing this is work on “focusing events” (Birkland Reference Birkland1998), which argues that sudden, unexpected, and visible events causing harm to a specific subpopulation can push event-relevant issues to the top of the public agenda and provoke shifts in public opinion. Together, these frameworks suggest that instances of social protest against the police, such as the 2020 Floyd protests, should exert widespread effects on public opinion. This expectation is supported by evidence that minority-led protest can shape news agendas and framing (Wasow Reference Wasow2020); lead to liberal shifts in voting on minority-relevant policies by members of Congress (Gillion Reference Gillion2012) and white voters (Enos, Kaufman, and Sands Reference Enos, Kaufman and Sands2019; Wasow Reference Wasow2020); instigate persisting changes in whites’ partisanship, prejudice toward African Americans, and support for affirmative action (Mazumder Reference Mazumder2018); and influence the political attitudes of coethnic bystanders (Branton et al. Reference Branton, Martinez-Ebers, Carey and Matsubayashi2015; Wallace, Zepeda-Millán, and Jones-Correa Reference Wallace, Zepeda-Millán and Jones-Correa2014). Adding to this, the fact that recent instances of lethal police violence against unarmed Black civilians are recorded, available for public viewing, and display visible use of excessive force, may add to their capacity to generate ubiquitous shifts in public opinion.
However, existing literature also suggests that social protest following recent police killings of Black civilians may have negligible or limited effects on public opinion. Research on political socialization contends that learned attitudes toward social groups are deeply ingrained, durable, and highly resistant to persuasion (Krosnick and Petty Reference Krosnick, Petty, Petty and Krosnick1995; Sears Reference Sears1993). Indeed, most political attitudes, particularly groupcentric ones, show an impressive amount of aggregate stability over time, particularly if elites do not change their public positions on group-related issues (Zaller Reference Zaller1992). This research is relevant to the present study because of the two groups involved: Black Americans and the police, with the former representing a long-standing affectively charged attitude object (Lodge and Taber Reference Lodge and Taber2005) and the latter becoming increasingly charged in the wake of the 2014 Ferguson uprising and evolution of BLM into a mass protest movement (Horowitz and Livingston Reference Horowitz and Livingston2016). As such, there are several reasons to expect that racial and partisan orientations have become increasingly important in structuring views toward the police. First, racial attitudes have played an integral part of contemporary partisan sorting and polarization (Tesler Reference Tesler2016), with attitudes toward BLM and the police playing a central role in the process. Second, while events such as the deaths of Eric Garner, Michael Brown, and George Floyd involve the use of excessive force against unarmed civilians by the police, we have witnessed the emergence of counterframes about these events focusing on victim resistance to police orders, “bad apple” narratives that belie claims of systemic racism in law enforcement, and the emergence of “Blue Lives Matter” counterprotests (Banks Reference Banks2018). The presence of these counternarratives and protests suggests that attitudes toward BLM and the police have become racialized and partisan issues (especially among whites), where individuals’ racial attitudes and political orientations structure their perception of episodes of Black protest following instances of police violence against Black civilians.
In the end, the police are a well-known and widely trusted institution of local government, at least among white Americans (Pew Research Center 2019). As such, attitudes toward the police may be highly stable over time and resistant to change. Added to this possible attitude inertia is evidence that views toward the police—especially among whites—may be increasingly subject to racial and partisan orientations. Recent scholarship demonstrates that a significant amount of the observed racial divide in Americans’ reactions to police killing of Black civilians derives from anti-Black prejudice among whites (Jefferson, Neuner, and Pasek Reference Jefferson, Neuner and Pasek2020). This finding is complemented by a historical study of the 1965 Watts uprising, where white residents in Los Angeles who harbored prejudice toward Black Americans were more likely to express negative views toward the uprising and endorse punitive measures against participants (Jeffries and Ransford Reference Jeffries and Ransford1969). With respect to partisanship, past research finds that Americans identifying with the Republican Party, as well as those residing in heavily Republican-voting states, are more likely to oppose the BLM movement (Updegrove et al. Reference Updegrove, Cooper, Orrick and Piquero2020). With these findings in mind, one distinct possibility is that episodes of social protest in response to police violence against Black civilians fail to exert a ubiquitous opinion-mobilizing effect among the American public; instead, such events only facilitate attitude change among those already sympathetic to the plight of Black Americans (e.g., low-prejudice and politically liberal Americans). Critically, among racially prejudiced and politically conservative Americans, such events may either exert no effect on views toward the police and awareness of racial injustice or trigger a reactionary shift in opinion comprised of elevated support for the police and repudiation of discrimination against Black Americans.
We subject these competing expectations to an empirical test using the case of the police killing of George Floyd in May 2020. Several features of the Floyd protests render it unique in the universe of protest events (e.g., rapid mobilization, unprecedented scale and media coverage, and international spread). However, the Floyd protests possessed important shared characteristics with other episodes of protest in the catalogue of the BLM movement and the annals of twentieth-century Black uprising against police violence that scholars use to classify protest events (McAdam et al. Reference McAdam, McCarthy, Olzak and Soule2021; Nam Reference Nam2006), such as the inciting incident (e.g., police violence against a Black civilian, acquittal of perpetrating officers), target of protest (e.g., the police), and purpose of protest (e.g., achieve justice/accountability, address systemic racism and police violence, promote reform). Additionally, from the vantage point of the concept of “most likely” cases (Gerring and Cojocaru Reference Gerring and Cojocaru2016), several of the attributes of the Floyd protests that make it unique (e.g., scale and media coverage) also arguably render it more likely than other episodes of protest to exert the broad effects on public opinion suggested by theories of activated public opinion and focusing events. If we fail to observe broad changes in public opinion following protest of the scale of the Floyd protests, the most likely case framework suggests that contemporary social protest against police violence—and especially instances of smaller scale than the Floyd protests—may overall fail to exert broad opinion-mobilizing effects and particularly fail to shift the attitudes of racially prejudiced and politically conservative Americans. Finally, our use of the Floyd protests to study the effect of social protest is consistent with the growing case-driven literature in political science using unique or extreme events to gain insight about the effect of broad categories of events, such as the Deepwater Horizon oil spill to study environmental disasters (Bishop Reference Bishop2014), September 11 to study terrorism (Huddy et al. Reference Huddy, Feldman, Taber and Lahav2005), the 2008 Financial Crisis to study economic recessions (Margalit Reference Margalit2013), the Syrian refugee crisis to study human migration (Hangartner et al. Reference Hangartner, Dinas, Marbach, Matakos and Xefteris2019), and COVID-19 to study public-health crises (Warshaw, Vavreck, and Baxter-King Reference Warshaw, Vavreck and Baxter-King2020). Focusing specifically on social protest, over half a dozen articles concentrate on a single unique protest event—the 2006 Immigration Rallies (e.g., Barreto et al. Reference Barreto, Manzano, Ramírez and Rim2009; Branton et al. Reference Branton, Martinez-Ebers, Carey and Matsubayashi2015; Wallace, Zepeda-Millán, and Jones-Correa Reference Wallace, Zepeda-Millán and Jones-Correa2014)—and notable other works use extreme episodes of ethnic uprising (Enos, Kaufman, and Sands Reference Enos, Kaufman and Sands2019; Hager, Krakowski, and Schaub Reference Hager, Krakowski and Schaub2019).
Our analysis of the Floyd protests adds to existing literature on twentieth-century minority protest (Gillion Reference Gillion2012; Lee Reference Lee2002; Mazumder Reference Mazumder2018; Wasow Reference Wasow2020) and growing scholarship on public support for the BLM movement (Arora and Stout Reference Arora and Stout2019; Bonilla and Tillery Reference Bonilla and Tillery2020; Updegrove et al. Reference Updegrove, Cooper, Orrick and Piquero2020). Prior research has explored Americans’ reactions to real (Boudreau, Mackenzie, and Simmons Reference Boudreau, Mackenzie and Simmons2019) and hypothetical (Jefferson, Neuner, and Pasek Reference Jefferson, Neuner and Pasek2020; Porter, Wood, and Cohen Reference Porter, Wood and Cohen2018) police killing of Black civilians; however, this work focuses on differences in attitudes caused by information provided in survey experiments and focuses on reactions to the killings themselves rather than social protest in response to killings. Moreover, research analyzing the causal effect of a prominent episode of social protest—the 1992 Los Angeles uprising—focuses on changes in white voting behavior, not public opinion, and focuses on an event and outcome within a single urban area (Enos, Kaufman, and Sands Reference Enos, Kaufman and Sands2019). In short, what is missing from the literature is a study of the effect of social protest of police violence against Black civilians that focuses specifically on public attitudes toward the police and perceived anti-Black discrimination, is national in scale, and enables the estimation of the causal effect of the event on real-time public opinion. This study provides such a test using data and an analytic strategy uniquely suited for the task.
DATA AND METHODS
One challenge in analyzing the effect of episodes of social protest, like the 2020 George Floyd protests, is having sufficient survey data immediately before and after events occur. To meet this challenge, we use the Nationscape survey (NS) conducted by the Democracy Fund and UCLA (Tausanovitch and Vavreck Reference Tausanovitch and Vavreck2020). The NS is a large-scale weekly survey (N = 6,250 per week) that began in July 2019 and is weighted to reflect the national adult population (Tausanovitch et al. Reference Tausanovitch, Vavreck, Reny, Hayes and Rudkin2019). Because the NS was in the field daily, averaging about N = 900 respondents per day, we can precisely estimate fluctuations in attitudes as a function of discrete events. We use the first 60 waves of the NS (July 2019 to September 2020), rendering a total sample of N = 378,507. We analyze two outcome variables in the NS: (1) favorability toward the police and (2) perceptions of discrimination against Black Americans in the U.S. Each variable is measured using four- and five-point Likert-type scales and recoded so that “4” indicates more unfavorable attitudes toward the police and “5” perceptions of greater levels of discrimination against Black Americans. Appendix A provides information about question wording and variable measurement.
Our analytic strategy involves leveraging the random timing of the police killing of George Floyd and ensuing nationwide protests and the use of a regression discontinuity in time (RDiT) approach to estimate change in favorability toward the police and perceived anti-Black discrimination just before and after this event. The random timing of these events assuages concerns about “anticipation effects” present with other RDiT designs focusing on planned interventions (Hausman and Rapson Reference Hausman and Rapson2018), as the police killing of Floyd was neither planned nor anticipated and the BLM protests erupted rapidly after Floyd’s death. Regression discontinuity designs (RDD) leverage as-if-random variation around an arbitrary cutoff to estimate local causal effects that correspond well to RCT treatment effects (Wing and Cook Reference Wing and Cook2013). The “running variable” we use is time—the number of days before (which take negative values) and after (which take positive values) the spread of massive protests in the wake of the Floyd killing.
We set the cutpoint (where the running variable = 0) to May 28, the first day after the outbreak of nationwide protests following the killing of Floyd. Although Floyd was killed on May 25, the cellphone video of his killing—and thus public knowledge of the event—didn’t emerge until May 26 when protesters took to the streets in Minneapolis. The following day, May 27, protests spread across the U.S., engendering a spike in media coverage, as is shown in Figure 1. Thus, we expect the full “treatment” of the protests to be initiated and reflected in public opinion data by May 28, which we choose as our cutpoint. Importantly, we find no evidence of an increase in survey response following Floyd’s killing (Figure A.1) and that the NS data is balanced on key demographics on either side of this cutpoint (Table A.1). Together, these checks suggest that any observed effects of the protests are not driven by event-initiated changes in survey response. Following best practices (Cattaneo, Idrobo, and Titiunik Reference Cattaneo, Idrobo and Titiunik2020), we model the running variable using a polynomial of order 1, which is least likely to overfit the data, though we show that our results are robust to other specifications (Figure A.2). We use a triangular kernel that, with a mean-squared-error (MSE) optimal bandwidth, yields a point estimator with optimal properties. Finally, we chose the bandwidth using a standard nonparametric approach that minimizes the MSE of the local polynomial RD point estimator given our choice of polynomial order and kernel.
Figure 1. George Floyd Media Coverage, Social Media Posts, and Search Behavior
Note: Lines indicate rescaled major newspaper and social media trends for “George Floyd” (top) and Google search trends for related keywords (bottom). Major newspaper article counts come from Media Cloud transcripts of the 50 newspapers with the largest circulation in the United States.
RESULTS
We begin in Figure 2 by first plotting daily mean attitudes toward the police (column 1) and perceptions of discrimination against Black Americans (column 2) for our full sample, and separately for White, Black, Latino, and Asian respondents. As can clearly be seen in the plots, the Floyd protests had a substantial effect on public attitudes: within the full sample, the event increased police unfavorability by 0.28 points (p < 0.01), or 27% of a standard deviation, and increased perceptions of discrimination against Black Americans by 0.19 points (p < 0.01), or 16% of a standard deviation. Disaggregating the data by White, Black, Latino, and Asian respondents, shown in rows 2 through 5, reveals similar trends. Full RD results, including estimates with bias corrections, can be found in Appendix Table B.1. While these shifts were rapid and substantively meaningful, the data suggest that attitudes among White Americans shifted back toward their pre-Floyd baseline means over time. Shifts in attitudes among Black, Latino, and Asian Americans, on the other hand, appear more durable in the posttreatment period.
Figure 2. Police Unfavorability and Perceived Discrimination against Black Americans
Note: Points indicate daily average unfavorable attitudes toward police and perceptions that Black Americans face discrimination in the United States. Full estimates in Appendix Table B.1.
The results presented in Figure 2 suggest there was a ubiquitous shift in public attitudes. However, we have yet to assess whether the overall effects mask underlying cleavages between those higher or lower in prejudice or between Republicans and Democrats—two of the strongest cleavages in American politics. Given that we are primarily interested in uncovering possible gaps in opinion by prejudice and partisanship, we estimate ordered probit models on weekly data, predicting each outcome as a function of racial attitudes, partisanship, and a host of controls. In Figure 3, we plot both the coefficient and 95% confidence intervals for indicators of prejudice (rows 1 and 2) and partisanship (row 3), which allows us to assess both whether attitudes are becoming more racialized and partisan as well as estimate whether this polarization in attitudes is driven by movement solely among those lower in prejudice (or strong Democrats), those higher in prejudice (or strong Republicans), or both. We use two indicators of prejudice included in the NS survey: (1) a Black–White favorability differential, which subtracts Black from white favorability Likert scales, and (2) the “generations” item from the well-known racial resentment scale. For partisanship, we use the standard seven-point scale ranging from strong Democrat to strong Republican. See Appendix A for more information on these items.
Figure 3. Coefficient and Predicted Value Plots for Prejudice and Partisanship
Note: Ordered probit coefficient (column A) and predicted probability of evaluating police “very unfavorably” (column B) by prejudice as measured by White–Black favorability ratings or racial resentment (generations item) and partisanship with 95% CIs. Ordered probit models are run on each weekly independent cross section and control for education, gender, age, race, household income, partisanship, and ideology, which are held at their means in simulations.
Beginning with the coefficient plots in column A, we find strong evidence of increased racialization and partisan polarization of attitudes. Pretreatment waves indicate that these attitudes were already polarized by racial and partisan orientations, though in all cases the coefficients significantly increase following the eruption of the Floyd protests.
In column B, we plot the probability of reporting a “very unfavorable” view of the police as a function of respondents’ prejudice and partisanship; these figures reveal that almost all of the movement in attitudes is among those lower in prejudice and among strong Democrats. These analyses suggest that the Floyd protests facilitated attitude change primarily among those who were already sympathetic to the BLM movement and failed to exert a meaningful effect on attitudes among those higher in prejudice and political conservatives.Footnote 1 Further, our analysis suggests that the size of the shift in mean attitudes among those lower in prejudice and strong Democrats shrank considerably over the following weeks, suggesting that, absent sustained protest, effects may decay. This said, it is clear that even with this observable decay, mean unfavorability toward the police among low-prejudice and strongly Democratic Americans nonetheless remained higher than pre-Floyd means several months post-Floyd, suggesting a possible durable shift in the intermediate term. We show similar results for perceived discrimination against Black Americans in Appendix Figure B.1. Further, we show in Table B.4 that a difference-in-discontinuity approach yields substantively identical findings to the modeling approach used here.
ROBUSTNESS AND MECHANISM CHECKS
While the results from our RDiT analysis are compelling, we conduct a series of additional checks to bolster our confidence. First, we show in Table B.3 that results remain unchanged when we cluster our standard errors by day or week. Second, we demonstrate in Table C.1 that the Floyd protests had little effect on event-irrelevant attitudes, such as favorability toward Jews, Evangelicals, socialists, whites, and Barack Obama.
In addition to these checks, we performed a series of exploratory analyses intended to offer insight about potential mechanisms linking the Floyd protests to attitude change. First, Appendix Figure C.1 explores shifts in opinion toward the police associated with 15 other police killings of unarmed Black civilians that occurred prior to Floyd and while the NS was in the field but that did not trigger nationwide social protest or significant national news coverage. Figure C.1 reveals little to no changes in attitudes toward the police surrounding these 15 other killings, suggesting the importance of social protest as a mechanism linking incidents of police violence to attitude change. On this point, the effect of social protest may in turn rely on subsequent intervening processes that facilitate changes in mass opinion. Prior research suggests that (a) physical proximity to the location of street protests and (b) media exposure serve as potentially important mechanisms linking social protest to attitude change (Branton et al. Reference Branton, Martinez-Ebers, Carey and Matsubayashi2015; Enos, Kaufman, and Sands Reference Enos, Kaufman and Sands2019; Wallace, Zepeda-Millán, and Jones-Correa Reference Wallace, Zepeda-Millán and Jones-Correa2014; Wasow Reference Wasow2020).
We present in Figure 4 results from subgroup RDiT analyses that explore the conditioning role of residential proximity to the Floyd protests, as well as self-reported attention to politics and media use, on attitude shifts toward the police (details in Appendix A). We find little evidence that living near the location of the protests (Panel A) played a noteworthy role in conditioning attitude change, as the RDiT estimates are nearly identical regardless of the amount of protest activity (e.g., “No Protests” vs. “More than 2 Protests”) in respondents’ county of residence. However, Figure 4 provides suggestive evidence that attention to politics and media consumption served as potential mechanisms generating attitude change from the Floyd protests, as we fail to observe statistically significant RDiT estimates among respondents who did not pay attention to politics or reported no media consumption whatsoever. In contrast, we only find statistically significant RDiT estimates among respondents who reported interest in politics and active consumption of news and social media. Further, in Figure 4 Panel B, we show evidence that partisan media consumption matters (Kilgo and Mourão Reference Kilgo and Mourão2019): attitude shifts among those who consume primarily liberal media (i.e., MSNBC but not Fox) is of greater magnitude and durability than those who consume primarily conservative media (Fox but not MSNBC). When combined with the lack of effects observed for the 15 police killings that did not generate large-scale protest (Figure C.1), our findings overall suggest a causal process where an instigating event (i.e., police killing) leads to protest activity and media coverage, which then affects public opinion.
Figure 4. Proximity to Protests and Exposure to News and Social Media
Note: RDiT estimates with 95% CIs for unfavorability toward police via county-level exposure to protest, attention to politics, or self-reported media consumption (Panel A) and predicted probability of evaluating police “very unfavorably” by partisan media usage (Panel B). Ordered probit models are run on each weekly independent cross section and control for prejudice, education, gender, age, race, household income, partisanship, and ideology, which are held at their means in simulations.
CONCLUSION
The findings in this study are of theoretical and practical importance. Theoretically, they illustrate that theories of activated mass opinion developed in the context of twentieth-century minority-led protest apply in part to an extremely notable episode of twenty-first-century social protest. However, consistent with accounts of the reactionary countermobilization of racially conservative Southern whites in response to Civil Rights-era Black protest (Lee Reference Lee2002; Wasow Reference Wasow2020), as well as recent literature on White backlash (Parker and Barreto Reference Parker and Barreto2013), we find that such effects are not observed among racially prejudiced and politically conservative Americans. Indeed, we find that mass protests over the killing of George Floyd further divided the attitudes of low- and high-prejudice Americans as well as Democrats and Republicans. These findings are of practical importance to the activist and reform community, as they suggest that social protest following tragic incidents of lethal police violence against Black civilians can create a favorable opinion climate for pursuing reforms that are directed at redressing racial bias in policing (e.g., Arora and Stout Reference Arora and Stout2019). However, our findings also suggest that persuading segments of the population predisposed against the cause of protesters may require preemptive frames designed to defuse reactionary counternarratives that activate prejudice and partisanship.
While the findings in this study are based on an analysis of a single and noteworthy episode of social protest against police violence, there are reasons to expect protest events of similar or greater magnitude in the future. For example, many expert observers view the rapidity and scale of the Floyd protests as the product of mounting frustration and years of movement building and, thus, as an amplified version of prior BLM protests (Politico Magazine 2020). Indeed, evaluation of media coverage of BLM protests in response to the killing of Eric Garner, Michael Brown, and Tamir Rice in 2014; Walter Scott and Freddie Gray in 2015; and Alton Stirling and Philando Castile in 2016 suggests a snowballing effect (see Appendix Figure C.2), with each subsequent episode of protest garnering greater amounts of media attention. Absent wide-scale reform of the institutions of policing and criminal justice, it is likely that the US will continue to witness police killings of unarmed Black civilians and the exoneration of those involved. Increasing frustration and exasperation in the American public, and Black Americans in particular, coupled with greater BLM brand recognition, more robust resources and networks, and increasingly sophisticated organizing techniques, suggest continuing and potentially larger protests in the future.
Supplementary Materials
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/S0003055421000460.
DATA AVAILABILITY STATEMENT
Research documentation and data that support the findings of this study are openly available in the APSR Dataverse at https://doi.org/10.7910/DVN/A2XDZP.
ACKNOWLEDGMENTS
The authors thank the participants of the American Politics Workshop at Washington University in St. Louis, Marcel Roman, Hakeem Jefferson, and the anonymous reviewers and editors for their feedback and suggestions.
CONFLICT OF INTEREST
The authors declare no ethical issues or conflicts of interest in this research.
ETHICAL STANDARDS
The authors affirm that this research did not involve human subjects.
Comments
No Comments have been published for this article.