Legislative action on police reform since 2020

George Floyd’s murder and the subsequent protests focused national attention on policing like never before. At first, the focus on police abuse and police reform was universal and bipartisan. Corporate giants, including Coca-Cola and Papa John’s pizza, endorsed the BLM movement in national ad campaigns (Ace Metrix 2020; Nguyen Reference Nguyen2020). In polls, respondents from both parties increased their support for BLM and decreased their favorability toward the police (Reny and Newman Reference Reny and Newman2021; Tesler Reference Tesler2020). And Republicans in Congress condemned Floyd’s murder and voiced support for at least some type of policing reform (Edmondson and Fandos Reference Edmondson and Fandos2020; Grynbaum, Karni, and Peters Reference Grynbaum, Karni and Peters2020; Snell Reference Snell2020).

Soon, however, public attitudes toward BLM and the police became polarized (Tesler Reference Tesler2020), and Republican elites shifted from embracing reform in the abstract to denouncing alleged calls to “defund the police,” which few if any Democratic officials actually supported (Dale Reference Dale2022). While both parties introduced comprehensive reform legislation in Congress, only the Democrats’ bill – the George Floyd Justice in Policing Act of 2021 – received a vote, with nearly all Democrats voting in favor and nearly all Republicans voting against. Among other provisions, the bill would have allowed civilians to sue police officers for wrongful deaths; increased federal oversight of local police; banned no-knock warrants, chokeholds, and the transfer of military equipment to local police; and required the use of bodyworn cameras (Lampe Reference Lampe2023).

Notably, while attitudes toward BLM and police funding polarized, many of the provisions listed above continued to receive majority support from supporters of both parties. For instance, among Republicans surveyed in the 2020 and 2022 Cooperative Election Studies (CES), 88% favored requiring officers to wear body cameras, about 2/3 supported a federal misconduct database, about 60% supported an end to qualified immunity (a legal doctrine shielding officers from civil suit), and just over half supported banning chokeholds; support for all of these proposals was even higher in the general public and higher still among Democrats. Yet congressional Republicans explained their opposition to the bill by falsely linking it to the unpopular idea of reducing police budgets. As House Republican leader Kevin McCarthy pronounced on the bill’s passing, “Democrats just doubled down as the party of Defunding the Police” (Fandos, Edmondson, Zraick Reference Fandos, Edmondson and Zraick2021).

While partisan polarization doomed national police reform, this does not necessarily mean policy toward police did not change after 2020. This is because law enforcement in the United States is largely out of the hands of the federal government, but is instead highly decentralized (Thomson-DeVeaux and Koerth Reference Thomson-DeVeaux and Koerth2020).Footnote ¹ As policing is not one of Congress’s enumerated powers, the 10th Amendment to the US Constitution gives states sovereignty over local law enforcement (Congressional Research Service 2023). While state legislatures write the state statutes that local police enforce, manage the state courts that prosecute defendants, and administer the state correctional facilities that house the overwhelming majority of incarcerated persons in the US, they have traditionally delegated their authority to local police departments, of which there are about 18,000 (Banks et al. Reference Banks, Hendrix, Hickman and Kyckelhahn2016). Zimring (Reference Zimring2020) characterizes state governments’ traditional role as theoretically “the most powerful level of government for criminal laws and criminal punishments,” but practically as “the least important level of government in policing…no level of government in the United States has less administrative or operational presence in policing than states” (see also Robinson Reference Robinson2020).

To assess policy change, then, it is necessary to ask whether states shifted from their traditionally laissez-faire approach after 2020. Data from several sources suggest that they did. According to the National Conference of State Legislatures (NCSL), which tracks police-related legislation in the states since 2020, states enacted over 1,000 such bills between 2020 and 2023 (National Conference of State Legislatures 2024). Another organization, the Wilson Center for Science and Justice at Duke Law, counts nearly 400 bills in the 2020–2021 period alone, up from about 100 in 2018–2019 (Garrett Reference Garrett2023). A third tally, by the Howard Center for Investigative Journalism at the University of Maryland and using the NCSL database, counts 300 reform bills between May 2020 and May 2022 (Monnay Reference Monnay2022).

In Figure 1, I display the annual counts of policing-related bills introduced (left panel) and enacted (right panel) in state legislatures. These data come from two sources. First, the Wilson Center, which tallies bills between 2018 and 2022 (Garrett Reference Garrett2023). Second, the NCSL, which tallies bills from 2020 to the present. While I rely primarily on the NCSL’s bill data in my analysis below, I include the Wilson Center time series here as it includes counts prior to the unrest of 2020. Both series show a high degree of state legislative activity around policing after 2020, especially in 2020 and 2021, and the pre-2020 data from Wilson suggests the post-2020 focus on policing is abnormal relative to prior years. Further, the figures suggest that police reform remained a salient topic in state legislatures well beyond the 2020 protests.

Figure 1.

Number of policing-related bills introduced and enacted in state legislatures, 2018–2023.

The role of party control

Assessing police reform at the state level also helps us understand how the politics of policing in the US have changed since 2020. For one, we can look to the states to see the extent to which policing has become nationalized, or correlated with national party divisions (Hopkins Reference Hopkins2018). Such an examination is difficult using national-level data given congressional inaction on this issue. But given the rhetoric of national partisan elites in the media, as well as the polarization on the single reform bill voted on in the House, we would expect that states controlled by Democrats would be more likely to pass reforms, while states controlled by Republicans would be less likely. Such a pattern would be consistent with past work showing state party control is an important predictor of general state policy liberalism (Caughey, Warshaw, and Xu Reference Caughey, Xu and Warshaw2017), and that party control structures states’ policy reactions to other salient events like election denialism (James and Stewart Reference James and Stewart2022) and mass shootings (Luca, Malhotra, and Poliquin Reference Luca, Malhotra and Poliquin2020).

On the other hand, policing and police reform may not be as polarized as some other issues. Speaking of the local level, but with relevance to state politics, Anzia (Reference Anzia2021) theorizes partisan divides may be less important for policing due to police unions’ strong influence on the Democratic Party. Similarly, Eckhouse (Reference Eckhouse2019) points to disagreements between white and Black Democrats as a cause of party elites’ inaction on reform. Consistent with the view that party is less important for policing than it is for other issues, at least one analyst characterizes the wave of post-2020 state action as “bipartisan,” citing successful examples of reform from both red and blue states (Friedrich Reference Friedrich2022).

Does police reform follow public opinion?

Scholars are only recently turning to the study of public opinion toward the police (Boudreau, MacKenzie, and Simmons Reference Boudreau, MacKenzie and Simmons2022; Sances Reference Sances2024a), with even less work focused on the connection between police-related opinions and policing policy (though see Enns Reference Enns2014 for a partial exception). Again, the numerous reforms enacted at the state level allow us to learn about this connection much easier than would be the case at the national level.

Existing studies of state policy have found evidence that public opinion matters, though often the public’s role can be quite limited. These quantitative assessments usually focus on two metrics of representation: responsiveness, or whether state adoption of liberal policy is more likely in states where public opinion is more liberal; and congruence, or whether states tend to adopt the policies favored by the majority of their citizens. Both assessments are important, and one does not necessarily imply the other (Achen Reference Achen1978; Simonovits, Guess, and Nagler Reference Simonovits, Guess and Nagler2019).

Past work has also found far more evidence of responsiveness than congruence. States with more liberal publics have been found to adopt more liberal state policies in general, and on specific issues (Caughey and Warshaw Reference Caughey and Warshaw2018; Enns Reference Enns2014; Lax and Phillips Reference Lax and Phillips2012; Wright, Erikson, and McIver Reference Wright, Erikson and McIver1987). Responsiveness is found to be especially strong on issues that are salient (Lax and Phillips Reference Lax and Phillips2012), as Black Lives Matter and police reform certainly have been in recent years.

There are less direct tests of congruence, as such examinations require that policy and opinion are measured on a common scale. Those studies that avoid this issue tend to find congruence is much weaker than responsiveness. For instance, on a set of 39 policies, Lax and Phillips (Reference Lax and Phillips2012) found states were congruent with majority opinion just half the time, while Rogers (Reference Rogers2017) finds that individual state lawmakers are rarely punished electorally for their incongruence, a result he attributes to the lower salience of state lawmaking. A more recent example by Simonovits et al. (Reference Simonovits, Guess and Nagler2019) finds vast discrepancies between the majority-preferred minimum wage and the actual minimum wage across states, even though minimum wage policy is responsive. Again, however, there is evidence that congruence is more likely in the presence of greater salience (Lax and Phillips Reference Lax and Phillips2012; Rogers Reference Rogers2017).

Measuring police reform

While raw counts of police-related bills are interesting, they are noisy measures of police reform activity for a few reasons. First, a policing bill is not necessarily a policing reform bill, which likely is why the three organizations cited above report varying numbers of laws. Some policing bills may have nothing to do with reform, but might instead cover routine appropriations or technical regulations; other bills might be passed in opposition to reform, such as cracking down on police reform protests. Second, there is the issue of substantive magnitude. One state may pass a comprehensive reform package with 100 provisions, while another state may pass two separate laws containing one provision each. Thus, one measurement approach would be to read all 1,000 bills flagged by NCSL and other organizations, identify the universe of provisions passed, and then code the direction of each provision and which states enacted it.

As a more tractable alternative, I rely on summaries of state legislative activity over this period to both construct the universe of reforms and determine which states passed them. This includes the reports by NCSL and Wilson Center cited above, as well as several others. Table 1 lists the reforms included in my study, and Section 1 of the Appendix describes the sources used and gives more details on each individual reform.Footnote ²

Table 1.

Police reform policies included in the study

Note: Number of states enacting a given policy are in parentheses.

A related issue is just what types of policy changes should count as reform-related. Even if a policy change is highlighted in a report, it may not actually advance or detract from the goals of the reform movement. My working definition of police reform is laws regarding police accountability and police use of force, which I take to be the major concerns of the protests that emerged after George Floyd’s murder. This definition does capture almost all the policies mentioned by the policy reports discussed in the Appendix.Footnote ³

To generate a state-level measure of reform, I use two strategies. First, I simply add the number of reforms passed by each state, assigning anti-reform policies negative weight.Footnote ⁴ I call this resulting measure my Police Reform Index. Second, and in part because the additive measure relies on my own subjective coding of what is pro- and anti-reform, I follow recent work in state politics (see e.g. Caughey and Warshaw Reference Caughey and Warshaw2018) in using Item Response Theory to create a continuous reform scale. I refer to this measure as my Police Reform Scale. Footnote ⁵

In Figure 2, I plot the frequency distribution of the two policy measures. The Police Reform Index ranges from −2 to 12, with a mean of 3.9 and a standard deviation of 4. This variable is more easily interpretable, given a one-unit change represents a state passing, on net, one more pro-reform policy. However, in interpreting the results below, it is worth keeping in mind that this measure combines both pro- and anti-reform policies. For example, a state could score +2 as a result of passing four pro-reform policies and two anti-reform policies. Likewise, while the mode of the Index is zero, this is most commonly because a state passes an equivalent number of pro- and anti-reform policies.

Figure 2.

Frequency distribution of state-level police reform.

In comparison, the Police Reform Scale ranges from −1.7 to 1.8, with a mean of 0 and a standard deviation of 0.92. While less intuitive than the additive Index, this variable can be interpreted as our best guess of a state’s latent policy orientation toward police reform, with higher (lower) values indicating more (less) pro-reform policy.

In Figure 3, I map the distribution of the two measures, shading states according to quintiles. Focusing on the states appearing in the extremes in both maps, we see especially high levels of reform activity in Washington, Oregon, Colorado, Virginia, Massachusetts, and Connecticut; we see especially low levels of reform activity in Idaho, Montana, Wyoming, North Dakota, Kansas, Oklahoma, Arkansas, Michigan, Ohio, Mississippi, and Alabama.Footnote ⁶

Figure 3.

Spatial distribution of police reform.

Measuring party control and public opinion

My measure of party control comes from the NCSL, which assigns each state a value of party control equal to Republican if Republicans control both the legislature and the governor’s office, Democrat if Democrats control both, and divided if the legislature and the governor’s office are controlled by different parties. I combine annual values of these measures from 2020 to 2024.Footnote ⁷

I use two measures of state public opinion. First, I use Democrat Joe Biden’s share of the 2020 two-party presidential vote in each state (Wooley and Peters Reference Wooley and Peters2021). Second, I construct an index of pro-reform sentiment using seven police reform-related questions on the 2020 and 2022 Cooperative Election Surveys.Footnote ⁸ To construct this index, I first estimate the average state opinion on each question using multilevel regression and post-stratification (MRP) (Gelman and Hill Reference Gelman and Hill2007; Kastellec, Lax, and Phillips Reference Kastellec, Lax and Phillips2019); I then take the average of the seven estimates for each state, re-scaling the result to lie between zero and one. I give more details on the construction of this index in the Appendix (Section 7).Footnote ⁹

To measure congruence, I use specific policy questions from the CES, focusing on only the five questions that I can directly link to specific state policies. I estimate average support for each policy in each state using multilevel regression with post-stratification (MRP); more details on this procedure are in the Appendix (Section 7). While not exhaustive, these questions include some policies adopted by many states (neck restraints, misconduct data), some policies adopted by few states (officer immunity, body cameras, military equipment), and in general link to the specific provisions of the Democrats’ policing bill in Congress. That said, these questions are not a random sample of police reforms, and so in that sense are not representative. On the other hand, it is likely that these questions were placed on the CES due to the presumed salience of the policies, such that it should be more likely for me to find congruence.

Police reform, party control, and public opinion

In Figure 4, I display the average levels of police reform against state party control.Footnote ¹⁰ For both reform measures, there is a clear linear trend, with Republican-controlled states scoring lowest, divided state scoring somewhere in the middle, and Democratic-controlled states scoring the highest. For the Police Reform Index measure, Democratic-controlled states score about six points higher, or about 1.5 standard deviations; for the Police Reform Scale, the partisan difference is about one point, or a little more than a standard deviation.

Figure 4.

Police reform and state party control.

Thus while some state action on police reform may have occurred in both red and blue states, reforms in Republican-controlled states were exceptions to the general trend. State enactments on policing largely follow the same partisan divisions as have been observed in other state policy areas, as well as those evident in the national debate over reform.

In Figure 5, I plot my police reform measures against public opinion. The top panels in this figure use my Police Reform Index as the dependent variable, while the bottom panels use my Police Reform Scale measure. Meanwhile, the left panels in this figure use Democratic presidential vote as the independent variable, while the right panels use the MRP-based measure of reform opinion.

Figure 5.

Police reform and state public opinion: responsiveness.

Note: The top panel uses the police reform index as the dependent variable. The bottom panel uses the IRT-based measure. The left panels use democratic presidential vote as the independent variable. The right panel use pro-reform opinion estimated using MRP. Dashed lines represent moving averages estimated using linear regression.

Regardless of the measure, there is a strong positive relationship between reform policy and opinion that is suggestive of responsiveness. For example, in the top-left panel we see that states with the least Democratic publics score at about −1 on my reform index, while states with the most Democratic publics score about +8. Likewise, in the rop-right panel we see that states with the least pro-reform publics score about 0, while states with the most pro-reform publics score about +8. Results are substantively similar when using the IRT-based measure of policy, as shown in the bottom panels. These graphs provide preliminary evidence that states do respond to public opinion when passing police reform policies.

In Figure 6 I describe congruence, plotting whether a state adopts the policy in question against state public opinion. In the subtitle to each plot, I report the share of states that adopt policies in line with the majority of their citizens. For instance, the top-right panel shows half the states are at “Yes” on the vertical axis, as half the states limited the use of neck restraints; all states are to the right of the zero line on the horizontal axis, as majorities in all states supported banning chokeholds. Thus congruence for this policy is just 50%, or what we would expect if policy was randomly generated without regard to public sentiment (Matsusaka Reference Matsusaka2010). Notably, the slope of policy to opinion, indicated by the solid line, is positive and consistent with responsiveness, demonstrating the importance of evaluating congruence as well.

Figure 6.

Police reform and public opinion on specific policies.

Congruence on body cameras and limiting officer immunity is even lower, about 12% and 14%, suggesting policy is systematically biased against what majorities want in these cases; while congruence is again roughly half (46%) on the issue of a national registry of officer misconduct. Congruence is unusually high, at 78%, for the issue of banning military equipment to local departments. However, this may be a coincidence of the fact that just five states adopted this policy, while majorities in only 13 states supported it; thus states did not need to exert much effort to please their publics in this case.

I use a linear regression to more precisely characterize these relationships and to quantify uncertainty. For predicting the Policy Index and Policy Scale variables, the basic regression specification I use is,

$$ {\mathrm{Policy}}_s=\alpha +{\beta}_1\hskip0.42em \mathrm{Split}\hskip0.42em {\mathrm{Control}}_s+{\beta}_2\;\mathrm{Democratic}\hskip0.42em {\mathrm{Control}}_s+{\beta}_3\;\mathrm{Public}\hskip0.42em {\mathrm{Opinion}}_s+{\varepsilon}_s $$

where $ {\mathrm{Policy}}_s $ represents the general policy adopted in state $ s $ , measured alternately using the additive Policy Index and the IRT-based Policy Scale; Republican Control is the reference category for party control; and Public Opinion is measured alternately using Democratic vote and MRP-based preferences. I use heteroskedasticity-robust standard error estimates in all specifications.

Column (1) of Table 2 quantifies the bivariate comparison of policy across party control shown previously. The constant of 1.39 represents the average reform index in unified Republican states; the coefficient of 3.19 on split control shows how much higher average reform is in states with divided control compared to the baseline of Republican control; and the coefficient of 5.74 on Democratic control shows how much higher the average reform index is in states with unified Democratic control compared to the Republican baseline. Both of these differences are significantly different from zero at at least the 0.05 level, with the Democratic-Republican difference more precisely estimated. As shown in the footer to the table, however, the roughly 2.5 point difference between Democratic and Split states is less precisely estimated, with a p-value of 0.07.

Table 2.

Regression of police reform index on party control and public opinion

+ p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001.

Note: Cell entries are coefficients from linear regressions. The reference category for party control is unified Republican control. Robust standard errors in parentheses.

Columns (2) and (3) quantify the bivariate relationships between policy and opinion, which I rescale such that zero represents the sample minimum and one the sample maximum. The least Democratic and least pro-reform states average − 1 and 0, respectively, on the reform index; the most Democratic and most pro-reform states have averages that are 9 or 8 points higher than these baselines, and both differences are very precisely estimated.

Column (4) shows results including both party control and Democratic presidential vote as predictors. Given the tendency of states with Democratic voters to elect Democratic politicians, the coefficients on both sets of predictors are attenuated – i.e., Democratic control falls from 5.7 to 3.5 and Democratic vote from 9.4 to 5.1 – though still significant at conventional levels. When adjusting for police-specific opinion in column (5), however, the coefficient on Democratic control is 4.6 and precisely estimated, while public opinion has a coefficient of 3.2 (compared to 8 in the bivariate regression) and a standard error of 2.2. Again, results using the IRT-base scale, presented in columns (6) through (10), tell a similar story.

Next, I examine responsiveness and congruence on five specific policies, earlier shown graphically in Figure 6. To do so, I first re-shape the data so that each observation is a state-policy. Next, I run two sets of regressions for two dependent variables. First, I use a binary indicator for whether the state adopts the policy as the dependent variable. Second, I use a binary indicator for whether state policy is congruent with majority opinion in that state. I regress each dependent variable on public opinion, party control, and policy fixed effects, as well as interactions between opinion, control, and policy fixed effects. Formally, the specification is:

$$ {\displaystyle \begin{array}{c}{\mathrm{Policy}}_{st}=\alpha +{\beta}_1\;\mathrm{Public}\hskip0.42em {\mathrm{Opinion}}_{s1}+\sum \limits_{k=2}^5\;{\beta}_k\mathrm{Public}\hskip0.42em {\mathrm{Opinion}}_{sk}\ast {1}_{\left\{t=k\right\}}\\ {}\hskip0.82em + {\mu}_1\;\mathrm{Split}\hskip0.42em {\mathrm{Control}}_s+\sum \limits_{k=2}^5\;{\mu}_k\;\mathrm{Split}\hskip0.42em {\mathrm{Control}}_s\ast {1}_{\left\{t=k\right\}}\\ {}\hskip7em + {\delta}_1\;\mathrm{Dem}\hskip0.42em {\mathrm{Control}}_s+\sum \limits_{k=2}^5{\delta}_k\mathrm{Dem}\hskip0.42em {\mathrm{Control}}_s\ast {1}_{\left\{t=k\right\}}+{\pi}_t+{\varepsilon}_{st}\end{array}} $$

where $ s $ indexes states and $ t $ policies; $ {1}_{\left\{.\right\}} $ is the indicator function; and $ {\pi}_t $ are policy fixed effects. I use robust standard error estimates, allowing for clustering within states. The opinion measure is on a 0 to 1 scale. Following Lax and Phillips (Reference Lax and Phillips2012), for the regression using congruence as the outcome, I measure public opinion using the size of the majority (i.e., the absolute value of the difference between state opinion and 50%).

I present my results in Table 3, starting with policy adoption and public opinion in column (1). These results mirror those seen in Figure 6 previously, with the coefficient of −17.84 on opinion reflecting the negative (but insignificant) relationship between opinion on body cameras and the adoption of body cameras, and the interactions reflecting the positive opinion-policy relationships for the other reforms, with slopes ranging from 1.6 (chokeholds) to 3.6 (immunity).Footnote ¹¹ Substantively, these estimates imply that a one percentage point increase in public support increases the probability of adoption by as much as 3.6 percentage points, or a high level of responsiveness; statistical tests of the relevant sums of coefficients show these slopes are only significant for immunity (p < 0.10) and military equipment (p < 0.05).

Table 3.

Regression of individual reforms and congruence on party control and public opinion

+ p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001.

Note: Cell entries are coefficients from linear regressions. The unit of observation is a policy-state. The reference category for policy is Body Cameras. The reference category for party control is unified Republican control. All specifications include policy fixed effects. Robust standard errors, clustered by state, in parentheses.

Column (2) shows that party control is also highly predictive of the adoption of specific policies. States with Democratic control are 33 percentage points (standard error of 13 points) more likely than states with Republican control to adopt body cameras, with partisan differences on other policies never falling below 0.26 (immunity), and these differences are significant at at least the 0.10 level with the exception of chokeholds. When adjusting for both opinion and party control in column (3), public opinion is either negative or close to zero and insignificant, while Democratic control is substantively large, but significant only for cameras and military equipment.

The models in columns (4) to (6) repeat the analysis, now using congruence as the dependent variable and size of majority as the opinion measure. With the exception of military equipment policy in the bivariate regression, where the size of the majority actually makes congruence less likely, public opinion is never significantly related to congruence in any specifications. For party control, results are more mixed. In column (5), without adjusting for opinion, states with Democratic control are more likely than states with Republican control to be congruent on cameras and immunity, but are less likely to be congruent on military equipment. When adjusting for opinion in column (6), the relationship for immunity is no longer significant.