The widespread attention paid to the murder of George Floyd by Derek Chauvin–with sustained, national protests and indisputable video evidence–led many journalists and political observers to wonder whether, and to what degree, this event sparked a racial reckoning. While evidence for a policy-based reckoning is limited (Jefferson and Ray, Reference Jefferson and Ray2022), some research suggests the events affected liberals’ attitudes towards police and perceptions of discrimination against Black Americans (e.g., Reny and Newman, Reference Reny and Newman2021). We extend this line of research by probing whether and how Floyd’s murder and the ensuing protests changed Americans’ racial attitudes more extensively than previous studies. We first present results from a content analysis of partisan media that demonstrates the proliferation of diverging activism and backlash frames in interpreting the murder and ensuing protests. We then analyze, in turn, three different attitude measures from different data sources (Gallup’s ‘Most Important Problem’ judgments, racial group favorability measures from the Democracy Fund + UCLA’s 2020 Nationscape surveys, and implicit associations from the Project Implicit ‘virtual laboratory’) to provide a comprehensive analysis of event-driven change in Americans’ racial attitudes during 2020.
Weaving these myriad data sources together enables us to extend existing research by examining, side-by-side, three different instantiations of racial attitudes: prioritization of racism and race relations as the country’s most important problem, racial bias manifested in explicit expressions of favorability of Whites relative to Blacks, and racially biased mental associations captured by the Implicit Association Test (IAT). While all constitute forms of racial attitudes, they each offer distinctive opportunities to assess how George Floyd’s murder and the ensuing protests altered how Americans think about and express attitudes about race and race relations. The first two capture consciously articulated evaluations but differ in the reference point (a problem facing the country versus personal favorability ratings). The last two reflect relative group judgments, but they differ in how much they reflect controlled articulated opinions (as captured by explicit favorability judgments) versus unguarded mental associations (as captured by the IAT). We thus offer the most exhaustive assessment to our knowledge regarding racial attitude change, and one that focuses on racial group attitudes, which are commonly seen as more impervious to change than the attitudes (such as evaluations of the police) analyzed in existing work (i.e., Reny and Newman, Reference Reny and Newman2021). Because different racial attitudes have different origins and social and political implications (e.g., Kinder and Ryan, Reference Kinder and Ryan2017), broadly assessing change matters for fully understanding the event’s attitudinal repercussions.
We find clear attitudinal discontinuities, suggesting salient public events like Floyd’s murder can change racial attitudes. Yet, we also find divergence in the direction, degree, and durability (within our study period) of change across measures and subgroups. Floyd’s murder suggested racism was an important national problem to all Americans, but change was especially large and long-lasting for racial liberals: for Blacks compared to Whites, and, among Whites, for Democrats compared to Republicans. Floyd’s murder moved explicit group favorability in a more positive direction towards Blacks relative to Whites, primarily because of White Democrats. Finally, for all Americans, summer 2020 produced more positive mental associations to Blacks, in a particularly strong and long-lasting fashion for Whites relative to Blacks. Racial attitudes, we show, are malleable; however, the direction, degree, and durability of change varies by attitude measure and by subgroup.
1. Can high-profile events change racial attitudes?
Conventional social-science wisdom casts doubt on the potential for George Floyd’s murder and the ensuing protests to shape racial attitudes. Politics have long been said to exist as a ‘sideshow in the great circus of life’ (Dahl, Reference Dahl1961, p. 305). Events, however highly publicized, may simply not penetrate the perceptual screen of ordinary citizens. Even if events do seize attention, there is no guarantee people will connect them to a societal problem requiring attention or that individuals’ racial attitudes, long viewed as archetypal of symbolic politics (Sears et al., Reference Sears, Hensler and Speer1979), will change.
Yet other research suggests such media-charged events can influence attitudes and behaviors. Events can force unknown or ignored issues into public awareness, thereby setting the agenda for policymaking (Lawrence, Reference Lawrence2000; Kingdon, Reference Kingdon2011). They can forge new connections between standing predispositions and public opinion (Pollock, Reference Pollock1994; Kam and Kinder, Reference Kam and Kinder2007) as well as change policy attitudes (Sears and Valentino, Reference Sears and Valentino1997; Thomas et al., Reference Thomas, Krosnick, Shook, Chiang, Krosnick, Chiang and Stark2016). Considerably less work, however, has examined whether events can change predispositions like racial attitudes; below we articulate the theoretical pathways by which they can do so.
Media coverage can serve an agenda-setting function (Iyengar and Kinder, Reference Iyengar and Kinder1987; Boydstun, Reference Boydstun2013; Wasow, Reference Wasow2020). George Floyd’s murder and concomitant protests may have been sufficiently high-profile to convince Americans of a problem needing attention. This motivates an Agenda-Setting Hypothesis: Floyd’s murder and ensuing protests increased the priority Americans set on addressing race and race relations.
Emerging research raises the possibility that Floyd’s murder may have produced a durable, sympathetic change in Americans’ views of Black Americans by providing new information. Publicized instances of police violence such as the beating of Rodney King have changed perceptions of discrimination (Sigelman et al., Reference Sigelman, Welch, Bledsoe and Combs1997; Tuch and Weitzer, Reference Tuch and Weitzer1997), something Reny and Newman (Reference Reny and Newman2021) also uncover for Floyd’s murder. Even scholars who view racial attitudes as difficult to change credit the civil rights movement with shifting prejudice’s expression (Kinder and Sanders, Reference Kinder and Sanders1996; see also Enos et al., Reference Enos, Kaufman and Sands2019). Research also suggests earlier periods of Black Lives Matter (BLM) activism stimulated ‘persuasive arguments about the value of Black lives or the virtue of struggling against racism’ (Sawyer and Gampa, Reference Sawyer and Gampa2018, 1041), changing Americans’ implicit and explicit views of Black Americans. Altogether, this research suggests that Floyd’s murder and the ensuing protests may have pushed Americans to hold more sympathetic, less hostile views of Black Americans, motivating an Event-Driven Progress Hypothesis: Floyd’s murder and ensuing protests produced more racially progressive attitudes.
Still, we note that this perspective ignores how depictions of the same event can, and as we will show in this case, have diverged. The type of attention, and frames applied, can differ in important ways, with implications for individuals’ racial attitudes. Race continues to divide U.S. politics, and racial activism often meets backlash (Hutchings and Valentino, Reference Hutchings and Valentino2004; Tesler, Reference Tesler2016; Wasow, Reference Wasow2020). Assertions of racial struggle may be ignored or stimulate backlash by suggesting marginalized groups want to challenge the status quo (Wilkins and Kaiser, Reference Wilkins and Kaiser2014; Wilkins et al., Reference Wilkins, Hirsch, Kaiser and Inkles2017).
Importantly, media coverage of BLM-related events need not strictly offer positive portrayals (Jacobs, Reference Jacobs2000; Kilgo and Harlow, Reference Kilgo and Harlow2019; Wasow, Reference Wasow2020). News outlets, especially partisan-affiliated ones, likely offer divergent interpretations of salient events (Levendusky Reference Levendusky2013). Indeed, MSNBC and Fox News offered divergent interpretations of the 2014 protests in Ferguson, Missouri, following Michael Brown’s death. MSNBC highlighted discrimination and racism present in instances of police violence and Fox News denied their relevance (Engelhardt, Reference Engelhardt2021). We thus expect that the political left framed the summer 2020 events to sympathize with racial activists’ stances while the right emphasized backlash and social disorder. As the ‘All Lives Matter’ and ‘Back the Blue’ slogans suggest, a clear counterpoint to BLM activists’ interpretations of George Floyd’s murder emerged in political discourse and in the news media. This polarized information environment and its competing event depictions mean that the magnitude and durability of attitude change may vary across subgroups of Americans, urging a more nuanced and conditional theorizing of event-driven racial attitude change than currently suggested (e.g., Sawyer and Gampa, Reference Sawyer and Gampa2018).Footnote 1 This produces an Event-Driven Polarization Hypothesis: Floyd’s murder and ensuing protests triggered countervailing attitude change whereby individuals predisposed to be racially sympathetic will report more racially sympathetic views while those predisposed to be racially hostile will report more racially hostile views.
Importantly, George Floyd’s murder may have differentially altered the discourse environment across partisan elites and news outlets only briefly before other news stories took over. Events can ‘explode’ problems onto the agenda, with media organizations then looking for, and reporting on, related issues (Boydstun, Reference Boydstun2013). But the news cycle is fickle, and new events inevitably take center stage. To capture if attitude change occurs but fades alongside media attention, we also investigate whether the magnitude and duration of change differs for racial liberals versus racial conservatives. This correspondence between attitude change durability and media coverage amount motivates the Elastic Opinion Change Hypothesis: Attitude change reflects situational pressures that, once removed, reverts attitudes to their prior (pre-Floyd) levels (Cialdini et al., Reference Cialdini, Levy, Herman, Kozlowski and Petty1976; Thomas et al., Reference Thomas, Krosnick, Shook, Chiang, Krosnick, Chiang and Stark2016).
Finally, we articulate these hypotheses in contrast to the conventional wisdom about intergroup attitudes. Racial attitudes are said to form early in life. Infants as young as three months old categorize by race (Sangrigoli and de Schonen, Reference Sangrigoli and de Schonen2004) and family, peer, and school influences shape and reinforce these categories’ meaning into adulthood (Jennings and Niemi, Reference Jennings and Niemi1974; Sears, Reference Sears, Iyengar and McGuire1993; Hailey and Olson, Reference Hailey and Olson2013). Generational attitude transmission (Jennings and Niemi, Reference Jennings and Niemi1974; Acharya et al., Reference Acharya, Blackwell and Sen2018), bolstered by social and residential segregation and external information environments (Entman and Rojecki, Reference Entman and Rojecki2000), can create settings where group attitudes persist (Henry and Sears, Reference Henry and Sears2009). Early-learned, affectively charged group orientations are often difficult to change (Sears, Reference Sears, Iyengar and McGuire1993; Tesler, Reference Tesler2015; Archer and Kam, Reference Archer and Kam2020): even Barack Obama’s election and inauguration did not appear to affect explicit or implicit racial attitudes (Schmidt and Nosek, Reference Schmidt and Nosek2010). Per Tesler (Reference Tesler2015), when highly salient events occur, racial attitudes may be more likely to influence people’s responses to those events, rather than being altered by them. Even Reny and Newman’s (Reference Reny and Newman2021) study of opinion change around George Floyd fails to uncover changes in prejudice within their regression discontinuity analysis of Nationscape data. Thus, we juxtapose our proposed hypotheses against the following Status Quo Hypothesis: Despite widespread attention to Floyd’s murder and the ensuing protests, racial attitudes remain unchanged.
A unique feature of this article is our expansive analysis of a variety of racial attitude measures. Racial attitudes vary in their emphasis on cognition or affect (DeSante and Smith, Reference DeSante and Smith2020) and in whether they capture automatic associations or consciously considered thoughts (e.g., Kinder and Ryan, Reference Kinder and Ryan2017). Still further complicating things, self-reports include beliefs in biological or cultural differences, stereotypes, explanations for racial inequality, and even judgments about race’s social and political importance (Schuman et al, Reference Schuman, Steeh, Bobo and Krysan1997; Kinder, Reference Kinder, Huddy, Sears and Levy2013). Critically, while all represent racial attitudes, they emerge from different processes and have different social and political implications. Focusing on just one instantiation of racial attitudes could provide a tidy yet seriously incomplete picture of whether and how racial attitudes more generally shift following salient events.
By probing multiple attitude manifestations, we can engage in several important contrasts with theoretical, measurement, and policy implications. Importance judgments and favorability ratings are two quite different self-reported racial attitudes. While the former asks whether people see race as a problem or something worth public concern, the latter probes directly someone’s evaluation of a racialized group. An event could change one, both, or neither, and focusing on just one undercuts our ability to think holistically about patterns of change. Favorability, too, can be assessed indirectly, not just via self-reports, with this implicit-explicit distinction important for interrogating sources of public opinion and contributors to political decision-making (e.g., Pérez, Reference Pérez2016; Kinder and Ryan, Reference Kinder and Ryan2017). Accounting for this distinction likewise improves our ability to characterize, broadly, attitudinal responses.
Before testing these hypotheses, we investigate the partisan media environment surrounding George Floyd’s murder and the ensuing nationwide racial justice protests to shed light on the degree to which media interpretations of the event were more univalent or polarized. Afterwards, we evaluate the degree to which our hypotheses receive support based on the type of racial attitude considered, drawing on data from Gallup, then the Democracy Fund + UCLA, and finally, Project Implicit, as each offer different racial attitude operationalizations.
2. A death that shook the nation
George Floyd was murdered on May 25, 2020. That he died was not in question. But how to understand his death was very much in question. Ordinary Americans could receive information about Floyd’s murder and the ensuing protests via myriad sources including activist-directed campaigns, peer-to-peer social media, and corporate partisan news. But in an increasingly racially polarized information ecosystem, the interpretations could differ widely.
To trace the volume and nature of the discourse environment surrounding George Floyd’s death and suggest whether the Event-Driven Progress or Event-Driven Polarization hypotheses are more likely, we content-analyze mentions of Black Americans in MSNBC and Fox News. While these two television outlets constitute just two of many possible sources of information and interpretation, they are places where reactions to this event likely diverged by beliefs about race and racism (Engelhardt, Reference Engelhardt2021).Footnote 2 We do not assume that all or even the majority of Americans consume news from these sources; rather, we want to identify reactions among exemplars in a polarized information ecosystem (see also Manning and Masella, Reference Manning and Masella2018), with the assumption that they reflect frames also disseminated by other outlets and across interpersonal contacts (Levendusky Reference Levendusky2013). We thus offer a more expansive, and richer, characterization of the summer 2020 events and their connection to views of Black Americans than presently available (e.g., Griffin et al., Reference Griffin, Quasem, Sides and Tesler2021).
We collected all available transcripts for all shows on these channels from 2020 using the NexisUni database. We then created a corpus by taking the transcript for each show episode, segmenting by speaker and turn, and then dividing these into rough paragraphs within each speaker’s remarks to capture a speaker’s approximately complete thoughts (Gillion, Reference Gillion2016). The unit of analysis is each speaker-turn-paragraph.
We first investigate coverage volume by considering trends in explicit mentions of Black Americans. We conducted a keyword search for terms including ‘Blacks,’ ‘Black people,’ and ‘African American,’ and counted how many texts included one or more of these terms, doing this separately for Fox and MSNBC. While clearly capturing explicit group mentions, the approach is limited by relying on texts with explicit references. It does not capture the breadth of conversation if other speakers, or parts of a speaker’s own remarks, exclude this reference. While restrictive, the data provide a glimpse of these sources’ narrative patterns.Footnote 3
We note first both Fox and MSNBC increased discussion of Black Americans immediately following Floyd’s murder. In the month preceding Floyd’s murder, Fox mentioned Black people on average 4 times per day and MSNBC 5 times, as defined by our speaker-turn-paragraph unit of analysis. But in the month after, we find these daily averages increase to 24 and 20, respectively. These are increases of 500% and 300%, respectively. We also find this coverage tracks accounts of media attention as explosive, but fleeting (Boydstun, Reference Boydstun2013). In September, average daily mentions decline to 11 on Fox and 8 on MSNBC, with still further declines through the rest of 2020. We refer interested readers to the appendix for more detail on these trends.
Two broad themes emerge in mentions of Black Americans after Floyd’s murder. One reflected racially sympathetic considerations surrounding activism, its virtues, and its motivations. The other reflected racially hostile considerations and backlash in response to this activism. These frames also appeared in coverage of past periods of police violence and urban protest including Watts (Jacobs, Reference Jacobs2000), Rodney King (Jacobs, Reference Jacobs2000; Lawrence, Reference Lawrence2000), and prior BLM protests (Kilgo and Harlow, Reference Kilgo and Harlow2019).
To capture each frame, we constructed counts within the set of mentions of Black Americans. We code activism frames as mentions including terms like: discrimin*, racism, and antiracism (Jacobs, Reference Jacobs2000; Lawrence, Reference Lawrence2000). We code backlash frames as mentions of Black Americans also including terms like: violenc*, crimin*, and riot (Jacobs, Reference Jacobs2000; Lawrence, Reference Lawrence2000). Whereas the activism frame emphasizes grievances over injustice and racial discrimination, the backlash frame features ‘delegitimizing’ language portraying self-serving and lawless actions following Floyd’s murder (Kilgo and Harlow, Reference Kilgo and Harlow2019).
We create daily counts of these frames’ prevalence, separately for MSNBC and Fox News. Figure 1 plots trends in coverage for each frame using smoothed trends. The lefthand panels report raw counts of the daily number of episodes where these frames appear and the righthand panels report the proportion of mentions of Black Americans where these frames appear. Before May, neither Fox News nor MSNBC mentioned either the activism or backlash frames to significant extent. But this changed in May, potentially linked to the May 5 release of video regarding the murder of Ahmaud Arbery, and persisted through the summer and fall following Floyd’s murder and the ensuing protests.
Racial activism and racial backlash frames in Fox and MSNBC, 2020.
On the activism frame, little differentiates Fox and MSNBC in the number of mentions or the share of all mentions of Black Americans incorporating this frame. Both outlets show a discontinuity around George Floyd’s murder, and as time progresses, each outlet’s tendency to use the frame decreases, and to similar degrees.
A critical difference across outlets appears for the backlash frame, in the bottom panels of Figure 1. Mentions of ‘violence’ and ‘mob’ sharply spike, more so on Fox than MSNBC, after Floyd’s murder. Further, the share of mentions of Black Americans including a negative descriptor is much higher on Fox throughout the post-May 25 period. As the bottom right panel shows, from June to October, nearly half of Fox News episodes that mentioned Black Americans included the backlash frame. The respective proportion of MSNBC episodes is half of that. Put together, our rough assessment indicates the political right emphasized backlash, while the political left gave a slight edge to activism.
In sum: While both Fox and MSNBC mentioned Black Americans more after George Floyd’s murder, these mentions diverged in kind. These outlets began with the same indisputable video evidence but subsequently engaged in different conversations, with the sequence of activism and backlash we find present in other policy areas (Manning and Masella, Reference Manning and Masella2018). These differences in coverage and content in these high-profile partisan outlets likely influence other outlets (Levendusky Reference Levendusky2013; Dreier and Martin, Reference Dreier and Martin2010), seeding a polarized information ecosystem. Our content analysis thus leads us to suspect that the Event-Driven Progress Hypothesis may be overly optimistic, and the Event-Driven Polarization Hypothesis may be more plausible. The racially sympathetic activist frame and the racially hostile backlash frame likely compete to change opinion. Further, the attention surge, followed by steady decline, suggests the plausibility of the Elastic Opinion Change Hypothesis, as attitudes may revert to pre-Floyd levels as media attention shifts elsewhere.
3. Race as the most important problem facing the country
Given this polarized information environment, how did Americans respond? The Agenda-Setting Hypothesis suggests increased coverage of news surrounding Black Americans will increase Americans’ likelihood of seeing racism or race relations as an important problem facing the country. To test this hypothesis, we analyze Gallup’s longstanding ‘Most Important Problem’ series. Generally on a monthly basis, Gallup respondents can provide up to three mentions of what they think are the most important problems facing the country. Gallup then codes these open-ends into one of a few-dozen topics. We focus on ‘racism and race relations.’ Increasing mentions after Floyd’s murder would suggest an agenda-setting effect.
Here and throughout, we test for racial attitude change through a variant of an interrupted time series (ITS) design. We compare attitudes before May 25 to attitudes measured after and examine trends. Observed discontinuities around Floyd’s murder would be consistent with event-driven attitude change.Footnote 4 Further, we estimate a series of models where we test for differences in the pre- and post- periods, across subgroups, to discern for whom change occurs and how long it lasts.
We accessed data from Gallup’s Key Indicators and Demographics series from January 2017-May 2021. Given the data collection schedule, we use monthly time indicators. The monthly sample sizes range from 961 to 4,563 with a median of 1,035. Preceding Floyd’s death, in early to mid-May of 2020, respondents focused on the COVID-19 pandemic (38.2%), governmental/political affairs (following the impeachment proceedings earlier that season) (19.5%), and the economy (6.9%). Only 3.6% of respondents mentioned race relations. Gallup’s next survey began June 1 and ran most of the month. While the COVID-19 pandemic remained the most important problem, only 18.2% of respondents mentioned it. Racism and race relations came next, mentioned by 16.5% of respondents. Governmental/political affairs came in third (15.5%). The percentage of respondents mentioning racism and race relations as the most important problem facing the country thus increased by more than 400% after Floyd’s murder.Footnote 5
Figure 2 visually depicts this temporal discontinuity. Each dot represents the proportion of respondents mentioning racism or race relations as the most important problem within each month, and each line represents a Lowess smoother, with a discontinuity set between May and June 2020. The first panel represents the data for the full sample. Statistical modelsFootnote 6 turn up a statistically significant discontinuity between the pre- and post-period (p < 0.001). Within the post-period, we observe a sharp rise then ensuing dip in mentions, consistent with our media coverage trends. However, the focus on racism and race relations still remains above its pre-period levels, an effect inconsistent with the Elastic Opinion Change Hypothesis (which predicts that attitudes will revert to pre-event levels).
Racism or race relations as the most important problem.
How much do racial predispositions condition these changes? To test the Event-Driven Polarization Hypothesis we use racial background and, among White respondents, political party as albeit imperfect proxies for racial predispositions.Footnote 7 Figure 2’s remaining panels display trends for these subgroups. All see sharp discontinuities following Floyd’s murder, but how much racism and race relations remain a high priority varies, with Black respondents and White Democrats demonstrating the most sustained, heightened priority. These patterns reflect, though are not completely explained by, shifting media attention.
Tracking temporal change following Floyd’s murder using monthly indicators relative to the period before, we find a significant 10-percentage point increase in June 2020 among White respondents in their view that racism/race relations are the most important problems facing the country. Black respondents demonstrate an even larger increase of 21-percentage points (significantly larger than that of Whites: Wald-test p < 0.05). Among White respondents, mentions increase by 14-perentage points among Democrats compared with a 6-percentage point rise among Republicans, a significant difference by party (Wald-test p < 0.01). Finally, we find that nearly every post-Floyd month shows a statistically significant and sustained increase in mentions of racism and race relations among Whites. Among Blacks, our tests are underpowered yet suggest greater attention persisting even through spring 2021. Among Whites, Democrats and Republicans show some indications of sustained prioritization of racism and race relations, although this is more pronounced among White Democrats.
While Floyd’s murder and the ensuing protests may have breached the attention of most Americans, its influence was stronger and lasted longer for those predisposed to hold racially liberal views. These results support the Agenda-Setting Hypothesis and the asymmetries in degree of change partially support the Event-Driven Polarization Hypothesis as racial progressives move away from racial conservatives. Moreover, our analyses suggest that in the post-Floyd era, the priority placed on racism and race relations extended over time, evidence inconsistent with the Elastic Opinion Change Hypothesis.
Summer 2020 appears rather unique in how much Americans acknowledged that racism and race relations constituted a problem that needed to be addressed. Gallup’s own analysis revealed that the only period covered by its time series that reflected more engagement with racism and race relations was the 1960s Civil Rights Era (Newport, Reference Newport2020). Our analyses suggest that the summer 2020 activities and their associated media attention may have generated a new equilibrium in heightening Americans’ assessment of racism and race relations as an important issue.
4. What kind of problem? sympathy and backlash
We next assess how Americans feel about race relations by investigating group favorability. Here, we take advantage of the Democracy Fund + UCLA Nationscape survey,Footnote 8 an enormous dataset with weekly internet surveys of U.S. residents collected through the Lucid market research platform. We analyze a difference score reflecting favorability towards Whites relative to Blacks, ranging from -1 (completely favor Blacks over Whites) to +1 (completely favor Whites over Blacks), with 0 indicating no difference.Footnote 9
Figure 3 reports average relative favorability ratings by week for 2020 with separate Lowess averaged trends preceding and following Floyd’s murder (indicated by the vertical line at week 22). We again find racial attitudes clearly shifted after Floyd’s murder. On average, respondents felt more positively towards Black people relative to White people after Floyd’s murder compared to before (p < 0.01). Most weeks in the post-period indicate significantly lower levels of favorability towards Whites relative to Blacks throughout 2020, suggesting durability inconsistent with the Elastic Opinion Change Hypothesis.Footnote 10 We note this result diverges from Reny and Newman’s (Reference Reny and Newman2021) conclusion of no meaningful post-Floyd racial attitude change apart from discrimination perceptions, despite using the same data. Our diverging conclusion may come from our choice to use an interrupted time series design (see Footnote 4) and our use of the full time-span of 2020, compared to their use of a regression discontinuity in time approach that uses data only through the beginning of September. The designs make different pre-post comparisons, with ours motivated by an interest in both the potential for change and its durability across a longer period of time.
Average favorability towards Whites vs. Blacks.
We also find that responses vary by racial predispositions proxied by racial background and by political party among Whites. While both White and Black Americans hold relatively more favorable views of Blacks relative to Whites in the post- compared to pre-period, this effect is five times larger for Black Americans compared to White Americans (Wald test p < 0.001).
Disaggregating White respondents by party reveals countervailing patterns, consistent with the Event-Driven Polarization Hypothesis. White Republicans on average express significantly (p < 0.001) greater favorability towards Whites relative to Blacks in the post-period, whereas White Democrats move significantly in the opposite direction (significant differences across groups, Wald test p < 0.001). These shifts also appear relatively stable for the rest of 2020.
Generally, the results indicate that George Floyd’s murder and its concomitant protests not only heightened Americans’ acknowledgment of the importance of racism and race relations as a problem, but also further polarized intergroup attitudes (cf. Reny and Newman, Reference Reny and Newman2021). These patterns appear for additional survey items in Nationscape including: beliefs about levels of discrimination towards Blacks relative to Whites and racial resentment.Footnote 11 In perceived levels of discrimination faced by Blacks relative to Whites, we see change among all groups in the immediate aftermath, but long-lasting change only among White Democrats. For racial resentment, a sharp discontinuity appears in the full sample, but this pattern quickly dissipates among White Republicans while it durably remains among White Democrats. Blacks also appear to report lower levels of racial resentment with reversion by year’s end. These additional self-reported racial attitude measures also support the Event-Driven Polarization Hypothesis, primarily attributable to change among White Democrats.
5. In the mind? implicit associations about Blacks and Whites
While we have found that more Americans reported that race was an important problem, and many shifted their relative favorability of Blacks viz Whites, these measures capture individuals’ consciously-articulated survey self-reports: explicit judgments about national problems and group favorability. To the extent that White Democrats demonstrate greater explicit self-reported favorability towards Black Americans, they might be simply ‘talking the talk’ or engaging in ‘virtue signaling’: mimicking dominant elite discourse rather than expressing genuinely held beliefs (cf. Engelhardt, Reference Engelhardt2023). Analogously, White Republicans may also be repeating the reactionary party line even if underlying beliefs have changed. Whereas these self-reports capture self-aware, deliberately articulated judgments, implicit measures capture effortless, automatic associations held in long-term memory. Such associations may reflect genuinely held beliefs or cultural knowledge absorbed from the information environment (Arkes and Tetlock, Reference Arkes and Tetlock2004; Pérez, Reference Pérez2016). Importantly, explicit and implicit racial attitudes are not always highly correlated nor influenced by the same factors; whether reported explicit bias corresponds with indirectly captured implicit bias can vary across targets of bias, individuals measured, measurement contexts, and situational and individual motivations (Gawronski and Bodenhausen, Reference Gawronski and Bodenhausen2006; Rydell and McConnell, Reference Rydell and McConnell2006).
The divergent portrayals of Black Americans in partisan media we observed run against what existing research would anticipate. Sawyer and Gampa (Reference Sawyer and Gampa2018) argue that attention to BLM repeatedly pairs Blacks with positive concepts like courage and agency, suggesting broadly pro-Black movement in implicit associations consistent with the Event-Driven Progress Hypothesis. Yet other work (Schmidt and Nosek, Reference Schmidt and Nosek2010) has found little to no change in implicit associations due to salient events, our Status-Quo Hypothesis. Yet according to our content analysis, if like-minded partisan media changes people’s implicit associations, then we may see Event-Driven Polarization instead: the left’s relatively more positive rhetoric may create more positive implicit attitudes on the left while the right’s relatively more negative rhetoric may produce more negative implicit attitudes on the right.
To test if Floyd’s murder shifted implicit associations, we use data on the Black-White faces IAT from Project Implicit, a well-known online platform housed at Harvard that collects data to facilitate research on societal bias.Footnote 12 In the IAT that we analyze, participants are exposed to a series of Black and White faces and positive and negative words, presented in random order and one at a time, and they are asked to sort these stimuli using designated keystrokes, as quickly as possible. Across sets of trials, the pairing of words and faces varies. In one trial, Black faces and negative words are sorted together using one keystroke (i.e., pressing a key like ‘f’) and White faces and positive words are sorted together using a different keystroke (i.e., pressing a key like ‘j’). In a subsequent trial, Black faces and positive words are sorted together using one keystroke and White faces and negative words are sorted together using a different keystroke. If participants are faster in the first trial (i.e., Black faces paired with negative words and White faces paired with positive words) compared with the second trial (Black faces paired with positive words and White faces paired with negative words), this suggests that the implicit associations for the first trial are more cognitively accessible, and thus more closely linked in memory, than the implicit associations for the second trial.
Participants receive a D-score capturing the difference in how quickly and correctly they complete these sorting tasks. Higher D-scores indicate that respondents completed the Black-Bad/White-Good pairings faster than the Black-Good/White-Bad pairings. Higher D-scores indicate stronger negative associations and weaker positive associations with Black faces along with stronger positive and weaker negative associations with White faces.Footnote 13
Because participants opt in to complete a study on Project Implicit, these data are not nationally representative. We address this sample self-selection three ways. First, we compare weekly trends in 2020 to weekly trends (at the same respective time of the year) in 2019. This accounts for potential participation seasonality (i.e., participation may increase during the school year) and gives us 2019 as a baseline for potential attitude change. Second, we follow other opt-in sample analyses by creating and applying weights for sample consistency (Gelman et al., Reference Gelman, Goel, Rivers and Rothschild2016; Caughey et al., Reference Caughey, Berinsky, Chatfield, Hartman, Schickler and Sekhon2020).Footnote 14 Third, we regress individuals’ IAT D-scores on the suite of demographic used to generate the survey weights along with week-of-completion indicators (and their interaction with year). Doing so allows us to take an additional step to address potential self-selection differences in our opt-in sample by controlling for the variables used to construct the survey weights (Berinsky, Reference Berinsky2009).Footnote 15
Figure 4 presents predicted IAT D-scores for a typical respondent for weeks with available data, by year.Footnote 16 We summarize trends with year-specific Lowess smoothed averages that separately reflect pre-May 25 trends and post-May 25 trends. We exclude completes on May 25, 2019, and May 25, 2020, to provide clearer weekly pre-post comparisons. The top left panel shows that implicit attitudes changed in 2020, after Floyd’s murder. While the solid dark line summarizing the 2020 estimates indicates an in-motion trend in decreasing anti-Black bias before Floyd’s murder, it also points to a change in predicted D-scores appearing a week later, patterns consistent with news coverage trends we found before and after Floyd’s murder. Comparing the week starting June 1 to the week starting May 18, we find an average change in D-scores consistent with previous work on general movement effects (Cohen’s d = 0.07), with the difference-in-difference between 2020 and 2019 larger still (d = 0.14).Footnote 17 While attitudes partially revert toward pre-Floyd levels, consistent with shifting media attention, implicit anti-Black bias remains on average reduced. This is inconsistent with the Elastic Opinion Change Hypothesis.Footnote 18
Predicted weekly average IAT D-scores.
The implicit associations indicate attitude changes occurred differently across subgroups. Considering first racial background, we see a sharp discontinuity in implicit attitudes around Floyd’s murder for White respondents: bias in the second half of 2020 remains lower than in the first half of 2020 (p < 0.001). In contrast, no clear discontinuity emerges among Black respondents.Footnote 19
Project Implicit only records ideological self-identification, so we consider heterogeneity among White liberals, neutrals, and conservatives rather than partisans (Levendusky, Reference Levendusky2009). Anti-Black bias declines among White respondents, but the magnitude of the shifts varies across ideological identification. Anti-Black bias appears to decline more for White conservatives and White ideological-neutrals than for White liberals, and these decreases generally persist throughout the year. Results in the appendix indicate that average pre-post changes in D-scores are somewhat larger for White ideological neutrals and conservatives (d = 0.097) than liberals (d = 0.077), a significant difference (p < 0.03).
These results support the Event-Driven Progress Hypothesis, for White respondents across the ideological spectrum. These results are consistent with Sawyer and Gampa’s (Reference Sawyer and Gampa2018) analyses of the earlier rise of BLM, and they may suggest that some positive media portrayals of Black agency may have penetrated the perceptual screen of even White conservatives and ideological-neutrals. Among Black respondents, the data are more consistent with the Status-Quo Hypothesis and may suggest that implicit associations are already ‘baked-in’ for a given in-group. Further, while we observe durable change through 2020, the rupture immediately after Floyd’s murder and slow decline toward, but not meeting, pre-Floyd levels reinforces the preceding sections’ insights: attitudes can change quickly as media coverage increases but begin to revert as attention declines. However, full reversion does not occur in our study period; our results suggest a new equilibrium in Americans’ implicit racial attitudes.
6. Conclusions
Conventional wisdom suggests racial attitudes are difficult to change. Our results consistently suggest otherwise: high-profile events can alter racial attitudes, but the direction, degree, and durability of change in our study period varies across attitude type and subgroup. Content analysis indicated increased media salience of race-related issues post-Floyd and diverging interpretations populating the information environment. We next showed that Americans increasingly said racism and race relations were problems facing the country, with larger and longer-lasting changes among those predisposed to hold racially progressive attitudes: Black Americans and White Democrats. This pattern aligned with durable changes in explicit attitudes, with Blacks and White Democrats exhibiting more favorable views of Blacks vis Whites and with White Republicans expressing more hostility towards Blacks vis Whites. Finally, we document a modest decrease in implicit anti-Black bias across the sample and for White Americans regardless of ideology. Even if survey-based responding is performative, with partisans ‘talking the talk,’ a progressive shift in implicit associations did actually occur across Whites—even among White conservatives. Table 1 summarizes our results.
Summary of results
By considering multiple racial attitude measures, we provide the most comprehensive depiction to our knowledge of such shifts existing data allows, an important result given the diverging patterns we observe. These racial attitudes are not mere substitutes in political and social life (i.e., Kinder and Ryan, Reference Kinder and Ryan2017); one racial attitude measure should not be privileged over another.
So, did a racial reckoning occur? If by ‘racial reckoning’ we mean consistent progressive change in racial attitudes, then Floyd’s murder was indeed a reckoning for some Americans, particularly those predisposed towards racially liberal views. White Democrats manifested a long-lasting progressive change on all three outcomes. We find similar results for Black Americans, although their implicit views did not change. But among White Republicans we find countervailing and less durable change. They acknowledged racism and race relations as a more important problem, but expressed more anti-Black bias on self-reports but less anti-Black bias on implicit associations. Thus, while some durable change likely occurred among Americans, we note that what we consciously articulate does not always fit with what we implicitly associate.
This juxtaposition in results between explicit and implicit attitudes could come from myriad sources. First, recall our interest in implicit attitudes came in part from their potential connection to different motivational processes than explicit attitudes. Implicit attitudes appear to change from associative processes like the repeated pairing of objects and attributes while explicit attitudes relate more to conscious processing goals (Rydell and McConnell, Reference Rydell and McConnell2006). While our theorizing for any hypothesis did not require individuals to consciously accept the news coverage frames offered, the accumulated evidence suggests this may matter for direction of change. Our content analysis revealed divergent frames which, in light of standard models of opinion formation (e.g., Zaller, Reference Zaller1992), require acceptance for attitudes to change in the corresponding direction. Relatedly, Ryan and Krupnikov (Reference Ryan and Krupnikov2021) suggest mere exposure could change implicit attitudes but not explicit attitudes. The observed implicit attitude trends may then relate to positive phrasing of ‘Black Lives Matter,’ images of peaceful protestors at the events’ outset, or other such indirect cues mattered even as other information may have contradicted these visuals, pushing explicit attitudes in another direction. It is also possible the progressive change we see among Project Implicit conservatives is a function of the types of conservatives completing these studies. While we cannot adjudicate between these possibilities, the pattern of results reminds us of the importance of identifying the specific racial attitude in question given that related, but different, attitudes can respond to the same stimulus in different ways.
While our results suggest new racial attitude equilibria following George Floyd’s murder, we note also that even fleeting shifts can be consequential. Changes in political priorities and shifts towards more progressive racial attitudes can provide reformers with initial support, and citizens may demand that reluctant officials change policy. These attitudinal shifts could help explain why some White parents responded to the summer 2020 protests by changing their approach to teaching their children about race (Anoll et al., Reference Anoll, Engelhardt and Israel-Trummel2025). But just as the attitude changes we uncovered reveal both progress and backlash, the policy space also revealed progress and backlash. Policy reforms have been neither broad-based nor persistent (Jefferson and Ray, Reference Jefferson and Ray2022). The incremental shifts we observed in summer 2020 remind us that institutional structures and social relations are sticky and hard to change; their persistence can mean, as attention fades and the urgency of social movements loses steam, that many Americans retain negative views of Black Americans or endorse facially neutral or colorblind practices as an indirect means to retain advantage (Bell, Reference Bell1993; Feagin, Reference Feagin, Krysan and Lewis2004; Bonilla-Silva, Reference Bonilla-Silva2014). In the absence of significant structural shifts, the inch of racial progress that we have documented here may just as easily inch back.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/psrm.2025.10036. To obtain replication material for this article, https://doi.org/10.7910/DVN/11KPQY.
Funding declaration
No funding was received to support this research.
Competing interests declaration
The authors declare no competing interests.
Data availability
Most data and all code necessary to replicate the analyses will be hosted on the journal’s Dataverse page. Due to licensing restrictions we cannot share the data used for the Gallup analyses. We will post relevant code, but original data would have to be acquired from Gallup.
Open access
