The Pandemic,Protests,and Perceptions of the Police: How COVID-19 and the Murder of George Floyd Affected Public Trust in the Chicago Police Department

Introduction

The COVID-19 pandemic placed tremendous challenges on all aspects of society and resulted in an unfathomable loss of life, an altering of economic trajectories, and populations of people still struggling to adjust to the continued presence of the virus in our society. When examining the agencies and institutions entrusted to respond to that public health emergency, much attention has been placed on police departments and their officers who were tasked with enforcing a voluminous and ever-changing set of regulations that were implemented to stem the spread of the virus (Shjarback & Magny, 2022; Van Dijk et al., 2022). Interactions with community members also changed as departments altered their operations and officers faced varying levels of citizen compliance with those regulations amidst ongoing fears of contracting and spreading the virus (McCarthy et al., 2021). Relations with the community were further altered in the summer of 2020, as shocking video of George Floyd's murder by a Minneapolis (Minnesota) Police Department officer catalyzed one of the largest movements against police brutality in the United States (Pryce & Gainey, 2021).

Amidst those challenges, questions exist about whether the pandemic, and the changes it ushered in for law enforcement agencies, coupled with antipolice protests in the summer of 2020, influenced public attitudes about the police. Furthermore, as the unrest receded and pandemic regulations were dropped entirely, questions remain about whether any changes in the trajectory of public attitudes influenced by those monumental events have dissipated or whether they endure today.

Research questions about public attitudes toward the police are always important. When the public expresses trust in police and believes in their legitimacy, they are more likely to follow the law and cooperate with authorities (Sunshine & Tyler, 2003). During crisis periods, such as the pandemic and widespread protests, views of the police and their impact on law-abiding behavior become critically important as public adherence to emergency measures and other enforcement functions have direct implications for the health and lives of both officers and community members (Perry et al., 2021).

This study thus seeks to understand the evolution of public attitudes toward the police in Chicago at multiple inflection points primarily centered on stages of the pandemic and the widespread protests over the murder of George Floyd. While the effects of the pandemic and George Floyd's murder reverberated throughout the entire nation, Chicago has recently been at the epicenter of controversies related to police violence, defunding, and reform (U.S. Department of Justice, 2017; Vanecko, 2020). As such, this study's sole focus on public attitudes in such a critical city provides a localized perspective that is missing in some of the nationwide (e.g., Vitro et al., 2022) or multi-jurisdictional (e.g., Pryce & Gainey, 2021) studies that have been undertaken. Using monthly district-level data from the Chicago Police Department's (CPD) community sentiment portal from 2017 to 2023, this study poses two research questions. First, did views of the police in Chicago change significantly during different stages of the pandemic and after the release of footage depicting George Floyd's murder? Second, were changes in views of the police conditioned by race and ethnicity? In addition to answering those questions, this study also uses a robust set of control variables to better understand factors that significantly affect public perceptions of the police.

This report begins with a literature review related to public views of the police and how those may have been altered by the realities of pandemic policing and massive protests against police brutality. It then describes the contributions of the study and the data and methods used to answer the research questions. Finally, it presents the findings and discusses their implications for research and practice.

Public Views of the Police

Earning and keeping a high level of trust with members of the public is a critically important goal for police in democratic societies (see Perry et al., 2021). It enables police departments and their officers to conduct a multitude of functions from order maintenance to crime control through being viewed as legitimate by members of the public. In the context of policing, legitimacy reflects “…the belief that legal authorities are entitled to be obeyed and that the individual ought to defer to their judgments” (Tyler & Huo, 2002, p. 14). When perceptions of police are positive and they are viewed as legitimate, the practical implications are also vast, as people are more willing to report crime, cooperate with the police, and serve as witnesses to criminal justice investigations or proceedings (Taylor et al., 2015, p. 234).

Conversely, when perceptions of police are negative and views of their legitimacy are lowered, crime and violence can increase as individuals feel increasingly alienated from law enforcement authorities and are more likely to respond to conflict through an escalation of interpersonal violence (Rosenfeld et al., 2021).

In assessing what drives or determines the level of legitimacy the public affords the police, research has neither indicated that perceptions are fully formed through the explicit legality or lawfulness of police actions (Meares et al., 2016) nor are they heavily influenced by a purely technical interpretation of how effective police are at responding to the crisis or achieving other numerically measurable outcomes like crime or arrest rates (McCarthy & Rosenbaum, 2015). Instead, perceptions of police are filtered through the way in which members of the public feel that they are treated by police officers and whether those interactions take place in a procedurally just manner (Meares et al., 2016). According to Wood et al. (2020, p. 9815), procedural justice is a model of policing that emphasizes listening to people, treating them with respect, making decisions equitably, and being trustworthy. The public's sensitivities, then, are not as much toward the question of legality as they are toward the perceived legality of the actions or how appropriately or fairly they felt the officer acted.

Views of the police are also likely to vary as a function of race and ethnicity. Research has found members of racial and ethnic minority groups, in particular those who are Black, view the police more negatively than those who are White (see Weitzer & Tuch, 2005). Kruis et al. (2023), for example, found Black survey respondents viewed police in their communities as significantly less procedurally just than White survey respondents, even after controlling for a variety of relevant variables. Views of Hispanic populations toward the police are less prominently covered in the literature, although studies have found their views to be less favorable than White respondents but more favorable than Black respondents (see Schuck & Rosenbaum, 2005). In line with the procedural justice model of policing, less favorable perceptions of police held by racial and ethnic minorities reflect differences in how they are treated by the police and how that filters through to their perceptions of legitimacy (Kruis et al., 2023).

Pandemic Policing, George Floyd, and the Altering of Police-Community Relations

Through this procedural justice lens, the current study argues that views of the police are likely to shift or evolve within the context of tremendous crises or other emergency situations, like the pandemic and the George Floyd video release and subsequent mass protests. For several reasons, events like those can alter the nature and extent to which the public interacts with and perceives the police. In addition, the effects of these events on perceptions of the police may further vary depending on the race and ethnicity of community members.

With regard to COVID-19, the early stage of that public health emergency was marked by police departments altering their operations and pulling back from many law enforcement activities (see Lum et al., 2020a; 2020b; Nix et al., 2021; Perry et al., 2021). Lum et al. (2020a; 2020b), for example, utilized two waves of surveys with over 1,000 agencies and found that a majority provided formal guidance to their officers to reduce traffic stops, pedestrian stops, and even arrests for minor offenses—all of which may have altered the police-community dynamic and influenced public perceptions.

Relatedly, the broader literature on public attitudes during crisis situations often shows that during the onset and early stages of an emergency, support for authorities is strong as community members unite against the threat. This phenomenon is often referred to as the “rally to the flag” effect (Mueller, 1970). The rally to the flag effect may be short-lived as the threat becomes normalized or individuals feel as if other core functions have been jeopardized or ignored (Parker, 1995). Applying that logic to the pandemic, interactions with police may have become more fraught as the public built up resentment towards continued masking regulations and the expanded role of law enforcement in matters that may have been perceived as being more appropriate for public health officials (Sandrin & Simpson, 2022). In fact, officers even experienced outright defiance from citizens due to pandemic-related stress and frustration, especially if it was perceived that officers were enforcing those regulations in procedurally unjust ways (McCarthy et al., 2021).

Views of pandemic-related regulations were also contingent on community context, with skepticism being higher in disadvantaged places where perceptions of police legitimacy were already low (Jones, 2020), like what characterizes many neighborhoods in Chicago (see Jain et al., 2022). In addition, research has indicated that the regulations themselves may have been enforced differently as a function of the racial and ethnic composition of certain places. Leal et al. (2023), for example, found that the enforcement of COVID-19 public health violations through citations was more likely to occur in zip codes in San Antonio, Texas, with above average percent of Black and Hispanic populations. Additionally, Kajeepeta et al. (2022, p. 62) found that New York City zip codes with higher percentages of Black residents had significantly higher criminal court summonses and public health and nuisance arrests than other zip codes, which indicated the racialized inequities in policing during that public health emergency.

Within this already altered police-community dynamic came another transformative event with the release of video footage on May 26, 2020, that showed George Floyd being killed by an officer working for the Minneapolis Police Department. As Pryce and Gainey (2021, p. 2) explained, the horrors depicted in that video served as a tipping point that led to widespread protests, calls to defund or abolish the police, and large segments of the population to galvanize against police brutality. Viewed through a procedural justice lens, the brutal treatment of George Floyd and the perceived illegality of actions taken by those police officers, formalized through later criminal convictions, are likely to change the ways in which the public views the police. The release of that video also had a direct impact on police operations, with Cassell (2020), for example, finding a significant reduction in proactive officer behaviors, which is another route through which that event could alter public perceptions of authorities.

Pandemic Policing, George Floyd, and Evolving Attitudes Toward the Police

The implications of the changes described above on public attitudes toward the police have been explored in a growing number of studies (e.g., Chamberlain, 2020; Chenane et al., 2023; Nix et al., 2021; Perry et al., 2021; Reny & Newman, 2021). These studies, conducted in various countries, have found that views of the police were favorable, at least early in the pandemic, which is consistent with the rally to the flag effect. Chamberlain (2020), for example, utilized survey data of United Kingdom residents collected during the early stages of the pandemic in 2020 and found most respondents felt that local police were doing a fair, good, or excellent job. Using online surveys in Serbia between March and April of 2020, Jankovic and Cvetkovic (2020) found that over 60% of the 325 respondents felt full confidence in the police and that they did their jobs well. The authors also found that this pandemic measure of confidence was higher than a comparable measure collected prior to the pandemic, providing support for the idea that perceptions of the police can improve during the initial stages of a crisis period. In New Zealand, Sibley et al. (2020) utilized propensity scoring to match a sample of over 1,000 people surveyed prior to the pandemic with a sample of over 1,000 people surveyed during the lockdown period. The analysis indicated that the lockdown sample reported a statistically significant higher level of trust in the police than the prepandemic sample, leading the authors to conclude that people are likely to place more trust in the police, and other authority figures, in the midst of a cohesive national response to a period of crisis, like what transpired during the early stages of COVID-19 (Sibley et al., 2020, p. 628).

There is also evidence that the initial rally to the flag effect waned as time elapsed and the pandemic entered different stages. Perry et al. (2021), for instance, used a panel design and compared public views across the first (i.e., April 2020) and third (i.e., December 2020) peaks of the pandemic in Israel. That study found that public attitudes towards the police in Israel significantly weakened and that survey respondents found the police to be less effective, fair, and trustworthy during the third peak than in the first peak. Additionally, respondents expressed less willingness to report pandemic-related violations to the police and less inclination to comply and cooperate with the police. Similarly, using three waves of data in Israel, Kimhi et al. (2021) found that trust in the police was lower among respondents in the latter two waves of the survey compared to the initial administration of the survey during May 2020.

In contrast to the initial, yet fleeting, rally to the flag effect seen in the aftermath of the pandemic, the George Floyd video release generated significant opposition to policing and support for substantive reform (Boudreau et al., 2022). In terms of studies that assessed perceptions of the police during that period, Reny and Newman (2021) used survey data collected on a weekly basis to demonstrate that unfavorable views of the police increased significantly and rapidly following the protests over the video release. However, the unfavorable views of White respondents shifted more quickly toward baseline levels, whereas the unfavorable views of Black, Latino, and Asian American respondents appeared more durable (Reny & Newman, 2021, p. 1502).

Other published work has focused on attitudes after the release of the video, without comparative data from a prevideo release period. Vitro et al. (2022) assessed changes in attitudes toward the police over five waves of survey data beginning in June 2020, which was the first full month after the George Floyd video was released. Surprisingly, the authors found small increases in propolice attitudes over time from June 2020 to October 2021 and evidence that the rise was more pronounced among Black survey respondents. Still, Black participants had lower average levels of propolice attitudes than White participants, and those differences were consistent across time despite the larger average level increase over time among Black participants relative to White participants (Vitro et al., 2022, p. 17). In addition, using data from over 1,000 online surveys completed by Virginia residents in the summer of 2020, after the George Floyd video was released, Pryce and Gainey (2021) found a significant effect of race on perceptions of trust, with Black respondents being more distrustful of police than other racial and ethnic groups.

Study Rationale and Contributions

While the studies cited above have quantified changing attitudes toward the police during the pandemic and periods of antipolice protests in the summer of 2020, the current study seeks to answer two research questions to fill gaps in existing knowledge. First, did views of the police in Chicago change significantly during different stages of the pandemic and after the release of footage depicting George Floyd's murder? Second, were changes in views of the police conditioned by race and ethnicity?

Answering those questions fills several important gaps in the existing literature. For one, building on academic work that has found markedly different perceptions of the police are held by different racial and ethnic groups (Weitzer & Tuch, 2005), this study separately examines views held by White, Black, and Hispanic respondents in the wake of the pandemic and George Floyd's murder. Since research has found that the enforcement of pandemic regulations has been conditioned upon the race of community members or the racial composition of spatial units (Kajeepeta et al., 2022; Leal et al., 2023), a procedural justice framework would suggest that the selective enforcement of pandemic regulations—or even the perception of disproportional enforcement—among different groups of people in different contexts may result in markedly different perceptions of police as the pandemic unfolded. A similar argument is true with regard to the effect of the George Floyd video release, as the literature summarized above (i.e., Pryce & Gainey, 2021; Reny & Newman, 2021; Vitro et al., 2022) has consistently found that views of the police in the aftermath of high-profile incidents of misconduct are racialized. As such, the ability of this study to assess measures of public perception of the police by race and ethnicity is one of its primary contributions to the literature.

Second, many of the studies cited above have used a limited number of waves of survey data to assess how these pivotal events influence the views of the police, or they only include data collected after the event without corresponding baseline data. In response, this study leverages over six years of public sentiment data in Chicago collected each month from November 2017 to November 2023 to assess how attitudes towards the police have changed in response to different stages of the pandemic and the release of the George Floyd video.

Third, since policing in the United States is often a localized phenomenon marked by the unique people in a given city and the history of the department therein, the study's focus on public attitudes towards the police in a single site, Chicago, is advantageous for research and practice. This point is especially true in Chicago, which has been under scrutiny for some time in the aftermath of the release of video footage in 2015 that showed the murder of Laquan McDonald by a Chicago police officer (U.S. Department of Justice, 2017). The release of the video resulted in a Department of Justice (DOJ) investigation and the department entering a consent decree that remains in effect today. As such, studying the evolution of public attitudes toward the police in Chicago during this critical period of time is an important contribution to the literature. Further, the focus on Chicago allows for the inclusion of a broader set of variables to better isolate significant changes in public sentiment. For example, control variables tabulated at the police district level of analysis include the number of officers assigned, stops conducted, and homicides and nonfatal shootings, in addition to Chicago-specific events like the implementation of the consent decree in 2019.

Finally, as a practical matter, while a municipal agency like Chicago is anomalous due to its size, smaller police agencies in the country may look to larger departments to understand potential issues that may later manifest in their agencies or even in terms of strategies for data collection, practices for engaging with their residents or even novel ways to measure and communicate about their successes. As such, this sort of context in Chicago not only speaks to issues of public trust in the police but also illustrates the salience of this specific city in answering these important research questions. The data and methods are described in more depth in the next section of this report.

Method

Measures

The data described above are used to operationalize the dependent, independent, and control variables outlined below.

Dependent Variables

We focus this analysis on three dependent variables: perceptions of trust by White respondents, perceptions of trust by Black respondents, and perceptions of trust by Hispanic respondents. These measures are each comprised of two survey items. First, respondents were asked how much they agreed with the statement: The police in my neighborhood treat local residents with respect. Second, respondents were asked how much they agreed with the statement: The police in my neighborhood listen to and take into account the concerns of local residents. The two items closely resemble elements of procedural justice identified by Tyler (2003). For both survey items, respondents were asked to reply by specifying a level of agreement from 0 to 10. Then, the two responses were averaged and multiplied by 10 so that they were reported on a 100-point scale from 0 to 100. These individual responses were spatially located in different geographic units (e.g., the police district) by Elucd (Zencity) using geotags or zip codes associated with the survey. Then, weighted average scores for White, Black, and Hispanic respondents, among other demographic characteristics, were compiled by Elucd (Zencity) and made available through the CPD's public sentiment dashboard. For the purposes of this article, we utilize monthly trust scores by White, Black, and Hispanic respondents, respectively, to the spatial unit of the police district from November 2017 to November 2023.

Independent Variables

We include two dichotomous independent variables assessing distinct stages of the pandemic in Chicago. First, the duration of the most stringent stay-at-home (SAH) order period, reflecting Illinois Governor J.B. Pritzker's statewide order that went into effect March 21, 2020 (https://www.illinois.gov/news/press-release.21288.html), and was lifted June 21, 2020 (https://www2.illinois.gov/Documents/ExecOrders/2020/ExecutiveOrder-2020-43.pdf), is captured through coding the months of March 2020 through June 2020 as one (1) to indicate this most restrictive period of the pandemic, while all other months are coded as zero (0). Second, the duration of the relaxed SAH order, in which a greater level of gathering and more freedom of movement existed, is captured through coding the months of July 2020 through March 2021 as one (1) to indicate the presence of the relaxed SAH order, and otherwise zero (0). Following March of 2021, Chicago high schools reopened, vaccines were widely available, and most of the pandemic restrictions were lifted (https://www.chicagotribune.com/2022/04/19/two-years-of-the-covid-pandemic-in-chicago-take-a-look-back/).

The George Floyd video release and the subsequent protest period are measured through a number of variables. First, we include a dummy variable, taking the value of one (1) for the period from June 2020 to the end of the data, and zero (0) for the pre-Floyd period. This variable, labeled as “Floyd level” in the statistical models, presents the change in the level of trust scores occurring immediately after Floyd's death, compared with the pre-Floyd level. In addition, we include a linear trend (“Floyd linear trend”) to measure the effect of Floyd's death over time. Further, we include a quadratic (“Floyd quadratic T.”) trend to best fit the data that show a quadratic function of the post-Floyd time series for all races, given that the post-Floyd trend is not a straight line but a U-shaped curve, as seen in Figure 1.

OPEN IN VIEWER

Figure 1.

Public trust in the police in Chicago.

Control Variables

While views of police are driven by process-related concerns, they are also contingent on the context in which people live. To that point, various measures of the police districts of Chicago are included as control variables to account for what Lai and Zhao (2010) described as the neighborhood context model of attitudes toward the police. We include district-level monthly counts of, respectively, nonfatal shootings and homicides that attempt to capture the extent of particularly violent forms of crime that can affect how people view the police. The number of sworn officers assigned per district per month is also included to help capture police presence, which has been found to be related to perceptions of police legitimacy and overall attitudes toward the police (Bolaji & Metcalfe, 2024). In addition, we include the number of stops of community members per district per month to gauge the extent to which police are having contact with members of the public. The measures of officers assigned and stops also help to model changes in trust scores since the presence of officers and subsequent stops might positively or negatively influence public views of the police, depending on the nature of interactions between officers and residents.

We also control for the level of concentrated disadvantage in each district per year. We retrieved information from the ACS on the percent of residents that: live below the federal poverty line; are unemployed; are under age 18; and identify as Black. Then, an index of concentrated disadvantage was derived by creating a scale of those four sociodemographic measures and interpolated for each year assessed in the study period.

Furthermore, the analyses control for the ongoing presence of the consent decree, which was approved by a federal judge and implemented on March 1, 2019. The presence of the consent decree in Chicago stemmed from the video release of the murder of Laquan McDonald by a CPD officer and a subsequent Department of Justice (DOJ) investigation that highlighted a pattern of civil rights violations and systemic deficiencies in the department. The ongoing presence of the consent decree is measured through a dummy variable coded as (1) beginning with the month of March 2019 and extending through the end of the study period. All other months prior to March 2019 are coded as zero (0). Finally, a series of monthly dummy variables control for seasonal trends in the data. The model for each race includes a different reference category as the best contrast to other months: June for White respondents, April for Black respondents, and June for Hispanic respondents.

Statistical Analyses

There are three statistical procedures. First, descriptive statistics and visual illustrations of the time series at the city and district level offer preliminary evidence on changes in public views of the police coinciding with stages of the pandemic and the Floyd protests. Second, panel unit root tests analyze whether the dependent variables are nonstationary. Third, panel data analyses by race and ethnicity examine varying effects of the pandemic and Floyd protests on public views of the police, while a range of covariates are held constant. Given the nature of the data collected for each district over multiple time periods, the study conducts a series of pooled, random-effects, and fixed-effects regression models to decide which model is preferred. The study uses Driscoll and Kraay's (1998) method to obtain standard errors that are robust to heteroskedasticity, spatial dependence, and serial correlation.

The Lagrange multiplier (LM) test confirms that random-effects models are preferable to pooled OLS models: for White respondents, chisq. = 7,927.00, df = 1, p ≤ .01; for Black respondents, chisq. = 1,047.30, df = 1, p ≤ .01; and for Hispanic respondents, chisq. = 741.35, df = 1, p ≤ .01. In addition, the Hausman test reports that random-effects models are better than fixed-effects models: for White respondents, chisq. = 1.44, df = 21, p = 1; for Black respondents, chisq. = 13.07, df = 21, p = .91; and for Hispanic respondents, chisq.= 2.11, df = 21, p = 1. Although the Hausman test failed to reject the null hypothesis that the unique errors or district fixed effects are not correlated with the regressors, it is still worthwhile to conduct fixed-effects models as sensitivity analyses. In real terms, it is unlikely to assume that unobserved or unmeasured characteristics of each police district would not affect the regressors included in the models. By its design, the fixed-effects approach adjusts for all relevant covariates fixed over time within each district. It should be noted that concentrated disadvantage is excluded from the fixed-effects models due to its lack of within-group variation and a multicollinearity problem with fixed effects.

Results

Descriptive Statistics and Visual Illustrations

The goal of the study is to show how public views of the police changed in the wake of the pandemic and Floyd protests. As seen in Table 1, the data include 1,606 observations across 22 police districts over 73 months from November 2017 to November 2023. The CPD has 22 districts, although they are numbered between 1 and 25. Districts 13, 21, and 23 have been closed or subsumed under other districts and thus do not contain any data for the period analyzed here. The panel data are strongly balanced since each panel includes the same time points, which are monthly measured.

OPEN IN VIEWER

Table 1.

Descriptive Statistics.

Variable	Variation	Mean	S.D.	Min	Max	Obs.
Trust score—White	Overall	66.16	8.56	32.77	93.46	N = 1,606
	Between		5.44	58.06	79.75	n = 22
	Within		6.70	38.62	87.26	T = 73
Trust score—Black	Overall	55.14	9.55	25.05	86.73	N = 1,606
	Between		6.71	45.78	65.07	n = 22
	Within		6.95	33.75	82.33	T = 73
Trust score—Hispanic	Overall	61.50	8.67	35.18	88.19	N = 1,606
	Between		4.69	55.30	73.30	n = 22
	Within		7.36	35.92	85.87	T = 73
Police stops	Overall	374.54	251.86	45.00	1,637.00	N = 1,606
	Between		160.67	173.99	747.99	n = 22
	Within		196.91	−157.44	1,445.31	T = 73
Officers assigned	Overall	314.83	60.09	188.00	489.00	N = 1,606
	Between		51.01	228.49	401.85	n = 22
	Within		33.55	232.84	401.98	T = 73
Homicides	Overall	3.07	3.56	0.00	25.00	N = 1,606
	Between		2.31	0.22	8.93	n = 22
	Within		2.76	−5.86	23.32	T = 73
Nonfatal shootings	Overall	10.06	9.89	0.00	74.00	N = 1,606
	Between		7.56	0.90	30.82	n = 22
	Within		6.58	−13.76	53.79	T = 73
Conc. disadvantage	Overall	22.70	12.70	6.56	47.85	N = 1,606
	Between		12.95	7.00	43.16	n = 22
	Within		1.02	19.29	27.38	T = 73

In the table, each variable is divided into a between- and within-group variation. Overall, trust scores vary by race and ethnicity, with White respondents (66.16) reporting higher levels than Hispanic respondents (61.50) and Black respondents (55.14). Figure 1 visualizes a distribution of race-specific trust scores at the city level. In addition, trust scores for all race groups vary over time and across districts, as seen in Table 1 and Appendix A-C. There are greater between-group standard deviations in trust scores for Black respondents (S.D. = 6.71), followed by Whites (5.44) and Hispanics (4.69). Trust scores also differ within groups over time; the standard deviations are 7.36 for Hispanics, 6.95 for Black respondents, and 6.70 for Whites. For all races, overall, the within-group variance is higher than the between-group variance.

Unit Root Tests

The study checks nonstationarity in the dependent variables using the Levin-Lin-Chu, Im-Pesaran-Shin, and Hadri tests with individual intercepts and trend, as seen in Table 2. All unit root tests reject the null hypothesis of a unit root, indicating the absence of nonstationarity in all dependent variables. The data do not warrant differencing for stationarity and are thus analyzed in levels.

OPEN IN VIEWER

Table 2.

Unit Root Tests.

	Levin-Lin-Chu	Im-Pesaran-Shin	Hadri
Trust score—White	−11.11**	−10.66**	39.44**
Trust score—Black	−14.32**	−13.67**	32.23**
Trust score—Hispanic	−12.81**	−11.89**	38.54**

Note. Figures presented z or Wtbar statistics with individual intercepts and trend.

**Significant at α = .01.

Models for White Respondents

The results are reported in Table 3. ¹ The first panel of the table indicates the random-effects results of regression for White respondents. The SAH variable is positive and statistically significant, presenting a significant increase in trust scores for Whites by 2.49 (p ≤ .01) under the SAH order. While there is no significant effect of the Floyd level variable, the Floyd linear and quadratic trends are statistically significant in different directions. Since Floyd's murder, trust scores have gradually decreased by 0.72 (p ≤ .01) per month. However, there was a rebound in mid-2022, and trust scores slowly increased until the end of the data period in 2023 (b = 0.02, p ≤ .01). Among the covariates, police stops are statistically significant (b = −0.01, p ≤ .01). There was a 0.01 reduction in trust scores in response to a one-unit increase in police stops. In addition, the coefficient for the 2019 Consent Decree is 3.09 (p ≤ .01), implying an increase in trust scores of 3.09 over time, on average per district, following the court-mandated oversight in March 2019. Finally, the fixed-effects models produced the same results as the random-effects model in terms of direction, magnitude, and significance, as shown in the second panel of Table 3.

OPEN IN VIEWER

Table 3.

The Models for White Respondents.

	Random-effects			Fixed-effects
Variable	Estimate	Std. Error	Pr (> |t|)	Estimate	Std. Error	Pr (> |t|)
Intercept	69.54**	3.47	<0.01	-	-	-
SAH	2.49**	0.84	<0.01	2.44**	0.83	<0.01
Relaxed SAH	−0.81	1.20	0.50	−0.80	1.19	0.50
Floyd level	1.56	1.42	0.27	1.51	1.41	0.28
Floyd linear trend	−0.72**	0.15	<0.01	−0.71**	0.16	<0.01
Floyd quadratic T.	0.02**	<0.01	<0.01	0.02**	<0.01	<0.01
Police stops	−0.01**	<0.01	<0.01	−0.01**	<0.01	<0.01
Officers assigned	<0.01	0.01	0.81	<0.01	0.01	0.78
2019 Consent Decree	3.09**	1.22	0.01	3.21**	1.17	0.01
Homicides	−0.03	0.06	0.54	−0.03	0.06	0.53
Nonfatal shootings	0.03	0.03	0.34	0.03	0.03	0.29
Conc. disadvantage	−0.12	0.11	0.25	–	–	–
Adj. R-squared	0.14	–	–	0.13	–	–

Note. Eleven seasonal dummies, as opposed to June as the best contrast reference category, are not reported in the table for conciseness. Overall, trust scores are lower during the month of June, as opposed to other months.

**Significant at α = .01.

Models for Black Respondents

In Table 4, ² the random-effects model shows that a significant decrease in trust scores occurred during the relaxation of social distancing (b = −1.95, p ≤ .05). Contrary to expectations, there is a significant and immediate increase in trust scores in the post-Floyd period compared to the pre-Floyd period, as indicated by the Floyd level variable (b = 6.56, p ≤ .01). Just like the models for White respondents, both Floyd linear and quadratic trend coefficients, b = −0.62 and 0.01, respectively, are statistically significant at the 0.01 level. The post-Floyd trend was initially downward but later started to increase in mid-2022. Furthermore, trust scores for Black respondents are positively associated with the 2019 Consent Decree (b = 2.37, p ≤ .01) and negatively associated with homicides (b = −0.09, p≤  .05), nonfatal shootings (b = −0.09, p ≤ .05), and concentrated disadvantage (b = −0.35, p ≤ .01). Finally, there is a similarity in the coefficients of most variables between the random- and fixed-effects models. Note that the relaxed SAH variable, in addition to the measures of homicides and nonfatal shootings, became nonsignificant in the fixed-effects model.

OPEN IN VIEWER

Table 4.

The Models for Black Respondents.

	Random-effects			Fixed-effects
Variable	Estimate	Std. Error	Pr (> |t|)	Estimate	Std. Error	Pr (> |t|)
Intercept	62.09**	3.74	<0.01	–	–	–
SAH	0.91	1.60	0.57	0.80	1.61	0.62
Relaxed SAH	−1.95**	0.97	0.05	−1.84	1.02	0.07
Floyd level	6.56**	2.12	<0.01	6.42**	2.12	<0.01
Floyd linear trend	−0.62**	0.17	<0.01	−0.58**	0.19	<0.01
Floyd quadratic T.	0.01**	<0.01	<0.01	0.01**	<0.01	<0.01
Police stops	<0.01	<0.01	0.08	<0.01	<0.01	0.07
Officers assigned	<0.01	0.01	1.00	<0.01	0.02	0.68
2019 Consent Decree	2.37**	0.82	<0.01	2.60**	0.85	<0.01
Homicides	−0.09**	0.04	0.04	−0.08	0.05	0.10
Nonfatal shootings	−0.09**	0.04	0.03	−0.07	0.05	0.10
Conc. disadvantage	−0.35**	0.08	<0.01	–	–	–
Adj. R-squared	0.12	–	–	0.07	–	–

Note. Eleven seasonal dummies, as opposed to April as the best contrast reference category, are not reported in the table for conciseness. Overall, trust scores are lower during the month of April, as opposed to other months.

*Significant at α = .05.

**Significant at α = .01.

Models for Hispanic Respondents

In Table 5, ³ the random-effects model reports a significant increase in trust scores in the SAH order period (b = 3.24, p ≤ .05), while no significant change occurred during the relaxation of social distancing. All Floyd-related variables are statistically significant at the 0.01 level. Contrary to our expectations and like the model results for Black respondents, there is a significant and immediate increase in trust scores in the post-Floyd period compared to the pre-Floyd period, as indicated by the Floyd level variable (b = 9.72, p ≤ .01). There are also significant month-to-month linear (b = −1.07, p ≤ .01) and quadratic (b = 0.02, p ≤ .01) trends following the Floyd murder and protests. Specifically, the relationship between trust scores and time was negative, but as time passed, it became positive, which indicates an upward trajectory in trust scores that visually appears to occur in 2022. Other statistically significant covariates include police stops (b ≤ 0.01, p ≤ .05), the 2019 Consent Decree (b = 2.05, p ≤ .01), homicide (b = −0.14, p ≤ .01), and concentrated disadvantage (b = −0.25, p ≤ .01). Finally, the fixed-effects model produced the same results as the random-effects model. The implications of these findings for research and practice are outlined below.

OPEN IN VIEWER

Table 5.

The Models for Hispanic Respondents.

	Random-effects			Fixed-effects
Variable	Estimate	Std. Error	Pr (> |t|)	Estimate	Std. Error	Pr (> |t|)
Intercept	64.91**	2.92	<0.01	–	–	–
SAH	3.24**	1.45	0.03	3.13**	1.44	0.03
Relaxed SAH	−1.50	1.11	0.18	−1.42	1.09	0.19
Floyd level	9.72**	1.66	<0.01	9.62**	1.65	<0.01
Floyd linear trend	−1.07**	0.17	<0.01	−1.04**	0.18	<0.01
Floyd quadratic T.	0.02**	<0.01	<0.01	0.02**	<0.01	<0.01
Police stops	<0.01**	<0.01	0.03	<0.01**	<0.01	0.02
Officers assigned	0.01	0.01	0.51	0.01	0.01	0.31
2019 Consent Decree	2.05**	0.54	<0.01	2.20**	0.55	<0.01
Homicides	−0.14**	0.05	0.01	−0.14**	0.06	0.01
Nonfatal shootings	<0.01	0.03	1.00	0.02	0.04	0.69
Conc. disadvantage	−0.25**	0.05	<0.01	–	–	–
Adj. R-squared	0.21	–	–	0.18	–	–

Note. Eleven seasonal dummies, as opposed to June as the best contrast reference category, are not reported in the table for conciseness. Overall, trust scores are lower during the month of June, as opposed to other months.

*Significant at α = .05.

**Significant at α = .01.

Discussion

Although years have elapsed since the emergency stages of the pandemic and periods of intense protest over police brutality in the summer of 2020, researchers are still trying to understand the implications of those events on a variety of outcome measures. This study has added to that growing literature through an examination of over six years of data on public sentiment toward the police in Chicago, IL. Specifically, framed through a procedural justice perspective, this analysis used survey data on trust in the police, collected by Elucd (Zencity) and made available through the CPD's sentiment dashboard, at the level of the police district to see how that metric changed during different stages of the pandemic and after the release of video footage depicting George Floyd's murder in 2020. In addition, the study assessed how those changing views of trust in the police were conditioned upon race and ethnicity, through using disaggregated perceptions of trust by White, Black, and Hispanic survey respondents.

Panel analysis indicates significant changes in public perceptions of trust were ushered in by the catastrophic events of 2020 and that those changes were conditioned by race. When examining perceptions of trust during the SAH order, we find empirical support for a rally to the flag effect, but only among White and Hispanic respondents. Specifically, perceptions of trust for both groups of survey respondents increased significantly in that 4-month period from March to June 2020, a result that is consistent with the rationale that people place more trust in authority figures, including the police, as part of a cohesive response to an immediate crisis (Mueller, 1970). This result also coincides with empirical studies of confidence in the police (Jankovic & Cvetkovic, 2020) and trust in the police (Sibley et al., 2020) that have found significant increases in those measures during the initial lockdown period of COVID-19, compared to baseline measures collected prior to the pandemic. Counter to existing research (e.g., Kimhi et al., 2021; Perry et al., 2021), however, we do not detect a statistically significant reduction in perceptions of trust during the relaxed SAH period, specifically among White and Hispanic respondents.

It is notable that the presence of a significant rally to the flag effect did not manifest among Black survey respondents. Instead, the random-effects model indicates that perceptions of trust among Black residents decreased significantly during the relaxed SAH order, a period that encompassed the months of July 2020 through March 2021. This significant reduction in perceptions of trust expressed by Black residents during the relaxed SAH period may reflect some of the tension embedded in the enforcement of pandemic-related regulations that were so contentious in this stage of the pandemic that occurred after the initial lockdown period but before the widespread availability of vaccines and a greater freedom of movement that started in April of 2021 (see Kajeepeta et al., 2022; Leal et al., 2023). While the study conducted here is not able to definitively link disproportionate enforcement of pandemic-related regulations in minority communities to diminished perceptions of public trust, it is one avenue through which the procedural justice framework would support such a statistical relationship.

As far as the implications of the George Floyd video release on public perceptions of trust, we see several important findings worthy of discussion. First, it is certainly surprising and counterintuitive to find a statistically significant and positive effect of the “Floyd level” variable for Black and Hispanic respondents. Interpreting this result is complex. Taken at face value, it may indicate that some aspect of the CPD's communication strategy or tactical response to the video release was received positively by these groups of survey respondents, although that conclusion strains credulity given the scathing report that was released by the inspector general for the city of Chicago (Ferguson & Witzburg, 2021). There is also a precedent in other published work. Vitro et al. (2022), for example, did find a rise in propolice attitudes among Black Americans in the summer and fall of 2020, although they lacked a comparative data point prior to that period. Those authors suggested that the numerical rise in their measure of propolice attitudes may be indicative of a “floor effect” more so than a meaningful improvement in perceptions of the police (p. 18).

Here, it is possible that antipolice sentiments might not yet have been reflected in the June 2020 survey, especially if the survey was administered to most respondents early in the month. To that point, the unexpected rise in trust scores detected here might still be a remnant of the increasing trends in the data ushered in by the 2019 consent decree. Overall, we see it as likely that this counterintuitive result is due to some artifact of the survey methodology or data collection itself, either in the timing of survey responses or in the nature of the sample that may have responded (or did not respond) during that particularly contentious period. As we discuss in more depth below, this specific result, which runs completely at odds with expectations, underscores the importance of greater collaboration with the private entities collecting these data, in addition to more transparency and availability surrounding these measures and underlying data so that researchers can better discern reasons for anomalous findings.

Other results related to the George Floyd video release are more consistent with a procedural justice framework. The analyses find significant month-to-month drops in trust scores for all races following June 2020, as reflected in the “Floyd linear trend” variables. Then, a significant upward reversal in each panel of data, as reflected in the “Floyd quadratic trend,” occurred in 2022 and trust scores continued upward at a significant level with minor fluctuations for White, Black, and Hispanic respondents. The downward linear trends in each data series from the video release through the end of 2021, which reflect significant declines in feelings of trust towards the police among Black, Hispanic, and White respondents, are consistent with Reny and Newman's (2021) work, although their study measured unfavorable views of the police, a slightly different concept. Unlike Reny and Newman's (2021) results; however, we do not see as quick of a rebound in the positive perceptions of White respondents towards the police. In fact, the results here suggest a great deal of consistency in the month-to-month patterns of trust scores among these three groups of respondents in that the measure started to rebound towards baseline levels at approximately the same time in 2022. In fact, by the end of the data series in 2023, average trust levels for each group were at, or above, historical averages, suggesting that the implications of the pandemic and the George Floyd video release have receded in the public's consciousness, at least according to these data.

Other results are worth discussing. For one, the enactment of the consent decree in the CPD in March of 2019 had a positive and statistically significant effect on perceptions of trust for each group of survey respondents. As we broaden the discussion to ways in which police departments can fortify perceptions of trust and procedural justice among the people they serve, especially those that have been historically marginalized, the positive effect of external monitoring and accountability on public trust detected in these data from Chicago is evidence that they can serve as an inflection point towards better police-community relations (see Greenberger, 2016). While municipal police departments have historically resisted those efforts, the study findings here demonstrate that methods of external accountability can significantly improve perceptions of trust among a broad set of people. Whether that mechanism is simply through the perception of accountability or the myriad ways the consent decree has filtered through to police-community relations is uncertain at this point, but future research should be cognizant of assessing the impact of such initiatives as police departments grapple with ways to improve their legitimacy in the eyes of the populations that they serve.

Additional model results related to the control variables are of note. Trust scores for Black and Hispanic respondents, as opposed to Whites, are significantly related to changes in crime and community covariates (Lai & Zhao, 2010). As expected, trust scores for racial and ethnic minorities tend to decrease over time across districts when homicides, nonfatal shootings, and/or concentrated disadvantage increase. There is also a notable disparity between White and Hispanic respondents in terms of the effect of police stops on trust scores. When police stops increase, trust scores for Whites significantly decrease, but those for Hispanic respondents significantly increase. However, the effects are marginal. Still, this finding begs additional research into understanding how different variables measuring aspects of police activity at spatial units of analysis relate to measures of police perceptions by distinct groups of people.

In terms of the practical implications, we broadly see this study and its results as important to departments that are constantly expected to respond to a variety of emergency and crisis situations (Bonkiewicz & Ruback, 2012). During those periods when the public is fearful, resources are spread thin, and emergency rules and protocols are in place, views of the public, especially related to procedural justice and legitimacy, are critical. This study indicates, however, that the anticipated rally to the flag effect is not universal and did not significantly manifest with Black respondents during the initial period of the pandemic, suggesting a heightened need for police agencies to engage and communicate with historically marginalized populations during emergency and crisis periods.

In addition, we see this study and its results as an opportunity to encourage departments to continue to implement measurement systems to capture public sentiment surrounding procedural justice. Historically, police departments have measured their success in terms of law enforcement-driven metrics, like crime rates, arrest rates, and other measures of police output (McCarthy & Rosenbaum, 2015). Public sentiment surveys, like the one conducted by Elucd (Zencity) in Chicago, are a mechanism through which departments can increase accountability for the use of procedural justice. As McCarthy and Rosenbaum (2015) argued, there is a long history of police departments using data to hold command staff accountable for effectively responding to crime in their districts. A similar directive, which they labeled as a shift from Compstat to Respectstat, could also manifest in police departments whereby commanders could focus on improving accountability related to procedural justice using metrics of this nature.

In this same vein, a suggestion to improve these methods of collecting public sentiment data would center on the need to collect data from individuals who had recent contacts with the police. The data collected by Elucd (Zencity) and utilized in this study are general survey data that capture views from individuals, the vast majority of who have not had a recent experience with the Chicago Police Department. To bolster the quality of these data, and get a better sense of procedural justice concerns, we advocate for the use of public satisfaction tools that focus on individuals with a documented recent contact with the police. As one example, the police-community interaction survey (PCIS) has been fielded by researchers in a number of agencies and has been positioned as a mechanism to measure police performance in new ways (Rosenbaum et al., 2016). That survey involves a sample of respondents who reported a crime incident, had a traffic accident, or were stopped for a traffic violation. By focusing on those individuals who have had a recent contact with the police, surveys of this nature can be better attuned to elements of procedural justice as perceived by those who recently had contact with a police officer, thereby making them more practical for departments as they monitor prosocial behavior by their personnel.

We also encourage police departments, including the CPD, who collect these data to be more transparent about the ways in which they use these measures of sentiment to inform decisions about operations and initiatives. Considering this study's findings, especially that perceptions of trust are chronically low among people of color in police districts with elevated levels of disadvantage, it would seem as if initiatives focused on those areas would be especially beneficial. Data of this nature can then be used to try and assess if those initiatives delivered on their stated goals of bolstering trust in those communities. As an illustration of this, a recent study attempted to assess the effects of the Chicago Neighborhood Policing Initiative utilizing these public sentiment data (Center for Neighborhood Engaged Research & Science, 2023). While that study did not find an impact of that initiative on improved measures of public sentiment, it provides a model for practical applications of these data in the pursuit of improved relations between the police and the public.

It is important to acknowledge the limitations of this study. For one, we do not have access to the raw survey data, as it is collected and distributed by Elucd (Zencity). As such, the possible underlying data issues related to the exact timing of survey responses, the nature of the sample, its generalizability to the demographics of the police districts, and the exact weighting specifications utilized to produce these scores are not available for further analysis. This issue reflects the increased complexity of public-private partnerships between police departments and private companies (see White, 2014) and underscores our suggestions above about finding mutually agreeable ways in which more transparency can exist around the data while still protecting Elucd's (Zencity) rights as a private company. Second, the data are limited in that the smallest unit of aggregation to which they are tabulated, the police district, is still a large spatial unit in Chicago with populations of people that average 125,000. As a result, there is still likely greater heterogeneity that exists in public levels of trust at smaller units of analysis, which this study and its primary data source cannot detect.

With these limitations in mind, it is our hope that future studies continue to assess levels and predictors of public sentiment towards the police, either through addressing the limitations described above or by pursuing a variety of work that we attempt to identify here. As Reid et al. (2024) found, political anxieties related to government corruption, the state of democracy, the direction of the country, and a contentious 2020 presidential election in the United States all contributed to diminishing public trust in the government. In turn, it is not certain how views of the local police, as measured here in Chicago during the same tumultuous period, are related to, influenced by, or filtered through broader perceptions of trust in government. Future research should thus be purposeful in how it measures and delineates trust in the government, which passes law, from trust in the authorities who enforce the law (i.e., the police). Longitudinal studies that track those separate measures, and how they relate to one another, would also be advantageous in the effort to better reconcile trust in the government with trust in the police. In addition, an expanded focus on the views of Hispanic populations toward the police is warranted, as these residents are less prominently covered in the literature, other than a small number of studies (e.g., Lai & Zhao, 2010; Schuck & Rosenbaum, 2005) that have found their views to be less favorable than White respondents but more favorable than Black respondents—a pattern of results similar to what was discovered in this study. Relatedly, research that assesses the views of Asian populations, in light of the increased discrimination and criminal victimization that they endured during the pandemic (e.g., Han et al., 2023), would be a welcome addition to the existing literature. Further, studies that attempt to examine these measures of public sentiment as independent, as opposed to dependent, variables would be beneficial. As one example, a procedural justice framework would hold that higher levels of public trust should relate to a greater level of engagement with the police, operationalized through calls for service. In the quest for improved perceptions of procedural justice and a heightened belief in police legitimacy, in emergency situations and otherwise, researchers should continue to focus on measuring ways that people are treated by the police, and departments should prioritize ways to design operations, policies, and actionable measures around those central goals.

Supplemental Material