Policy Congruence and Policy Culture-making in American States

To understand the entrenchment of policy cultures in American state-level policymaking, one must consider three factors: policy congruence, diffusion, and feedback (feed-forward) effects. Social construction theorists argue policy designs—the decision to benefit or burden particular groups—are typically arrived at based on a consideration of two factors: (1) how much political power a group has, and (2) how positively or negatively a group is perceived to be by society (deLeon Reference deLeon2005; Ingram, Schneider, and DeLeon Reference Ingram, Schneider and DeLeon2007; Kreitzer and Smith Reference Kreitzer and Smith2018; Schneider and Ingram Reference Schneider and Ingram1993; Schneider and Ingram Reference Schneider and Ingram2005). Benefits or burdens are disbursed depending on where the target falls in one of four categories: groups are either advantaged (politically powerful and positively constructed), contenders (powerful and negatively constructed), dependents (weak and positively constructed), or deviants (weak and negatively constructed). Benefits are overprescribed to advantaged groups and underprescribed to “deviants”; benefits are seldom disbursed to “deviants,” and advantaged groups are rarely burdened (Ingram, Schneider, and DeLeon Reference Ingram, Schneider and DeLeon2007; Pierce et al. Reference Pierce, Siddiki, Jones, Schumacher, Pattison and Peterson2014; Schneider and Ingram Reference Schneider and Ingram1993).

Congruent policies reinforce dominant stereotypes that encompass target populations. Advantaged groups are perceived as more “deserving,” and any benefits received were “earned,” were the product of good behavior, or to justify their political influence. “Deviants,” on the other hand, deserve the burdens they receive because they are perceived as bad actors or norm violators. As a result, congruent policies routinely amass widespread support among the public and political elites (Johnson Reference Johnson2009; Schneider and Ingram Reference Schneider and Ingram1993; Simmons Reference Simmons2017). Group constructions become a part of the electoral calculus, as politicians are ever-aware of the power of the group as well as the reactions from voters looking to express their support for (or disapproval of) different policy decisions (Schneider and Ingram Reference Schneider and Ingram1993; Schneider and Sidney Reference Schneider and Sidney2009).

Once congruent policies are created, they routinely spread—or diffuse—beyond the borders of the original innovator. States often derive their own policy actions from past adoptions from other states. Early studies of policy diffusion suggested geographic similarities and contiguity fueled state-to-state policy spread (Walker Reference Walker1969). In the time since, scholars have identified multiple mechanisms of diffusion (i.e., Berry and Berry Reference Berry and Berry1999; Boehmke and Witmer Reference Boehmke and Witmer2004; Shipan and Volden Reference Shipan and Volden2008). States look to learn from past experiments with policy innovations and craft their own actions based on previous successes or failures (Boehmke and Witmer Reference Boehmke and Witmer2004; Butler et al. Reference Butler, Volden, Dynes and Shor2017). Others use heuristics to mimic the innovations of others based merely on partisan or ideological similarities that they may share (Shipan and Volden Reference Shipan and Volden2008). There are a few barriers to congruent policy diffusion. If states seek to learn from previous adoptions, congruent policies provide a range of information to justify emulation. Congruent policies provide political information for prospective states, as “they enjoy widespread public support, engender minimal counter-mobilization by the target population, and promise strong electoral returns for policymakers (Boushey Reference Boushey2016, 199).” If states base their emulation on shared values, congruent policies are easy to justify because the values that shape those decisions are aligned and reinforced by societal perceptions of deservingness.

The interactions between politics and policies entrench a culture of congruence. Scholars have long sought to settle the “chicken-or-egg” debate concerning the relationship between policies and politics. Some argue that politics dictate policies—that lawmakers work to appease voters and are responsive to the will of the public (Hacker and Pierson Reference Hacker and Pierson2014; Mayhew Reference Mayhew1974). Those who adopt a pluralist perspective would argue that congruent policies emerge because societal views and popular policy preferences dictate them. Policymakers are merely being responsive to their wishes. Others take a more policy-centric approach to explain how “policies create politics” (Campbell Reference Campbell2012; Hacker and Pierson Reference Hacker and Pierson2014; Mettler Reference Mettler2002; Schattschneider Reference Schattschneider1975; Schneider and Ingram Reference Schneider and Ingram2019). This perspective contends that politicians use their decisions to elicit specific responses from the mass public. Here, congruence becomes engrained in the political culture because policymakers intentionally design policies to invoke specific reactions from the public. Some even use deception to ensure a desired public output (Schneider and Ingram Reference Schneider and Ingram2019). Figure 1A reveals the dilemma for those seeking to enact noncongruent policy change. Policy cultures emerge from cycles of congruent policy innovations feeding the social and political dynamics that fuel the spread of policies across jurisdictions. The diffusion of policies then reinforces the social and political dynamics that motivate decision-makers to create new congruent policy innovations. This is an unrelenting cycle.

Figure 1. The entrenchment of congruent policy cultures

Motivating Noncongruent Policy Change: The Case of the State-Level Response to the #BlackLivesMatter Movement

Policy cultures are virtually, but not entirely, intractable. Occasionally, advantaged groups are burdened, and “deviants” are benefitted. However, noncongruent policies are rare, as they run counter to the logic of advantaged populations being rewarded and “deviant” groups being burdened. Further, policymakers are typically disincentivized from punishing advantaged groups because advantaged groups wield tremendous political power and resources to countermobilize politicians aiming to burden them (Schneider and Ingram Reference Schneider and Ingram1993). There are tremendous barriers to noncongruent policy diffusion. However, several studies shed light on how noncongruent policies may spread across jurisdictions. Boushey (Reference Boushey2016) finds that states were eleven percent more likely to enact noncongruent policies when their neighbors adopted similar policies. States are also likely to mimic those that are ideologically similar. Blanton and Jones (Reference Blanton and Jones2021) find similar effects in the diffusion of anti-trafficking laws. However, they also find that noncongruent policies diffuse much slower than congruent policies.

Aside from a few articles, the literature pays little attention to how noncongruent policies take shape (see Blanton and Jones Reference Blanton and Jones2021; Boushey Reference Boushey2016). Even fewer explore the potential for marginalized groups to initiate noncongruent change amidst a culture of congruence (i.e., Taylor Reference Taylor2016). Political decisions are often majoritarian, and those in the minority (numerical, racial, or otherwise) are at a distinct disadvantage when influencing policy decisions (Gilens Reference Gilens2005; Gilens and Page Reference Gilens and Page2014; Schattschneider Reference Schattschneider1975). I contend that for noncongruent policies to take hold, there must (1) be an intervention in the form of demands for noncongruent policies and (2) exist pathways to motivate the innovation and diffusion of noncongruent policy action. To further expound on this notion, I examine the adoption and diffusion of non-congruent police reforms in the aftermath of the #BlackLivesMatter movement.

#BlackLivesMatter’s Intervention: Demands for Noncongruent Police Reforms

Issues often exist in the environment but have yet to demand a public or government response. Governing bodies in Michigan knew for months of the lead-tainted water supply in Flint before videos of brown water surfaced on social media (Kennedy Reference Kennedy2016). Governments knew of a deadly virus spreading throughout China and Europe well before implementing protective measures in America (University of California—Davis Health 2022). Black and brown folk in America have long been subject to disproportionate uses of deadly force at the hands of police (DeGue, Fowler, and Calkins Reference DeGue, Fowler and Calkins2016). It is not enough that people are simply made aware of a problem to initiate policy action. An “alarmed discovery” must be accompanied by emphatic calls to action—those calls must be loud enough to garner the attention of both politicians and the public (Downs Reference Downs1972) (Fig. 1B). From the moment Darren Wilson shot and killed an unarmed Michael Brown in Ferguson, Missouri, a flow of viral videos was followed by immediate calls to “do something” about the extrajudicial killing of unarmed Black Americans (see Downs Reference Downs1972). The “something” in this instance was to enact policies that would challenge the cultures and practices that lead to police killings, change how police interact with Black and Brown communities, and provide accountability measures for those who engage in unjustified uses of deadly force.

The qualities of the demands matter. Robert Smith (Reference Smith1996) argues that the nature of the demands and the mode of delivery are directly linked to the government’s response to those demands. Demands can be systemic or non-systemic. Systemic demands—those that are “of the system”—are more likely to result in substantive policy responses. By “system,” I am referring to “the entire American complex of basic institutions, values, beliefs, etc.” that drive political, economic, and social governance (Ture and Hamilton Reference Ture and Hamilton1992, 41). Naturally, non-systemic demands (i.e., “abolish the police,” “overturn the election,” and “end capitalism”) pose a threat to any of those systems and are more likely to be neglected or repressed with force. Like demands, the methods used to forward demands can also be systemic or non-systemic. Non-systemic methods (violence, riots, etc.) are far more likely to be rejected than systemic methods (i.e., lobbying and voting) (Smith Reference Smith1996). Demands must be sustained long enough to ensure the problem does not fall out of public or elite attention. This may be especially true for noncongruent demands—decision-makers may be inclined to “wait them out” rather than change course.

These demands took aim at one of the most advantaged target groups there is—law enforcement. By most accounts, police are generally positively viewed by society and are, thus, more deserving of benefits (Kreitzer and Smith Reference Kreitzer and Smith2018; Schneider and Ingram Reference Schneider and Ingram1993). Police have also amassed an impressive amount of political power as punitive politics have become engrained over time. They are well-organized, well-supported by segments of the public, and have tremendous leverage over elements of both major political parties (Barkan and Cohn Reference Barkan and Cohn1998; Bies Reference Bies2017; Marks Reference Marks2007; Wilson and Buckler Reference Wilson and Buckler2010). Consequentially, criminal justice policies—like most policy designs—are also overwhelmingly congruent (Boushey Reference Boushey2016; Owens and Gunderson Reference Owens and Gunderson2022). Punitive criminal justice policies are typically reserved for “deviants,” and decisions to do so are supported by society (Chiricos, Welch, and Gertz Reference Chiricos, Welch and Gertz2004; Welch and Payne Reference Welch and Payne2010). In the rare instances that benefits are dispersed, they are generally reserved for advantaged groups. Federal, state, and local governments have invested billions in policing in recent decades (Naylor Reference Naylor2020). Corporations benefit from the proliferation of private, for-profit prisons and the prison manufacturing complex (Hallett Reference Hallett2006; Logan and Rausch Reference Logan and Rausch1985).

The contents of the demands also matter. At their core, demands can carry information that has the potential to motivate and shape policy decisions by reducing information costs associated with policy action (Baumgartner and Jones Reference Baumgartner and Jones2015; Workman, Jones, and Jochim Reference Workman, Jones and Jochim2009). Demands often communicate the nature of a problem, the root causes, the impacted populations, the scope of the problem, and suggested courses of action (Baumgartner and Jones Reference Baumgartner and Jones2015; Downs Reference Downs1972; Gillion Reference Gillion2013). Demands also relay information regarding the sense of urgency that should be devoted to solving the problem (Lesch and Millar Reference Lesch and Millar2021). Noncongruent demands also communicate policy-centered information about the problem in relation to the status-quo political culture. At a minimum, demands for noncongruent policy actions express that congruent policy designs cannot solve the problem. At the extreme, noncongruent demands communicate that congruent policy designs cause the problem.

Demands can also inform decision-makers that the political calculus of inaction in noncongruent areas has changed. Activating potentially supportive states could involve altering the political incentives and transaction costs that motivate most political decisions. Politicians are keenly aware of fluctuations in the electoral landscape (Gordon Reference Gordon2007; Mayhew Reference Mayhew1974). Groups can communicate that they are willing to organize and mobilize around the issue by applying strategic pressures on politicians and institutions (i.e., lobbying, voting, protesting, etc.) (Ainsworth Reference Ainsworth1997; Burstein and Linton Reference Burstein and Linton2002; Gause Reference Gause2022; Gillion Reference Gillion2013; Victor and Koger Reference Victor and Koger2016).

Groups can also motivate noncongruent actions by communicating to policymakers that the social arrangements that motive designs have changed. Groups can relay that there is a lack of consensus around how groups are constructed (Kreitzer and Smith Reference Kreitzer and Smith2018). They can also communicate to strategic partners that groups have actually changed from positively to negatively constructed (or vice versa) or that groups have gained or lost power (deLeon Reference deLeon2005; Schneider and Ingram Reference Schneider and Ingram2005). Target populations are multidimensional, as are the constructions that define them (Oorschot Reference Oorschot2000). Demands may be able to communicate that group construction should be considered along a different axis to justify the desired noncongruent design.

BLM protests carried valuable information to motivate action beyond bringing attention to police killings and outlining plans to remedy them. The movement’s initial battle cry, “Hands Up, Don’t Shoot!”—communicated that it would rely on systemic methods—peaceful protests—to forward their demands (Smith Reference Smith1996). The movement’s demands also communicated that the problem, itself, was a function of the overprescription of benefits to police and that those benefits (i.e., qualified immunity, internal investigations, paid administrative leave) stood in the way of real accountability. Lawmakers could also derive information on the political ramifications of action (or inaction). Over time, protesters clearly showed that those bringing the demands enjoyed popular support beyond the impacted target population. The movement represented a growing, rapidly diversifying, well-organized segment of the public. They also displayed that the public was willing to mobilize at a moment’s notice using social media to organize and activate marches (Tillery Reference Tillery2020). Protests during election years applied electoral pressures on lawmakers faced with the potential for mobilization on the streets to transfer to mobilization at the ballot boxes (Gillion Reference Gillion2012; Gillion and Soule Reference Gillion and Soule2018). Protesters also regularly showed that they were willing to risk real bodily harm—or even death—to enact change (via a militarized police response to BLM or through exposure to a global pandemic).

The demands also relayed information concerning the social dynamics contributing to the problem. They first argued that police killings were directly related to the punitive practices engrained in the police culture. The disproportionate rate at which Black and brown people are intrinsically linked to the social arrangements that dictate criminal justice policies. They also challenged the construction of the police, arguing that those who are tasked with “protecting and serving” communities were engaging in unsavory behaviors and, if left unchecked, would continue to do so. This reconstruction justifies their stance that police were less deserving of the benefits that they routinely receive and more deserving of burdens that could reign in bad actors. Finally, the demands of BLM showed that punitive policing policies were not constrained to “deviants”; unarmed citizens were subject to the ultimate burden—death at the hands of the police.

A Network Approach to Examining Noncongruent Interventions

I rely on social network analysis (SNA) to identify and examine innovation and diffusion patterns before and during the BLM movement. The development of a network application is the latest evolution in the half-century since Walker (Reference Walker1969) first posited scholars could potentially use “innovativeness scores” to examine what drives a state’s propensity to experiment with novel policy solutions and influence future adoptions from their peers. However, as Berry (Reference Berry1994) points out, modeling a complex process such as policy diffusion presents a unique set of challenges. Namely, prior to the emergence of a network application, diffusion studies struggled to (1) capture the multidimensionality of diffusion processes, (3) capture the temporal and relational aspects of diffusion, and (3) speak broadly to the different influences on general diffusion patterns across a range of policy issues. These challenges persist over decades of scholarship that relied heavily on event history analysis (or survival models) to trace the spread of policies across jurisdictions (Berry Reference Berry1994; Berry and Berry Reference Berry and Berry1990; Boehmke Reference Boehmke2009).

The social network analysis approach that grounds this study is a relatively novel advancement that is better equipped to address each of those challenges. First, SNA has the capacity to examine the internal, external, and structural factors that influence the sharing of policies across states in ways that previous models proved insufficient (Desmarais, Harden, and Boehmke Reference Desmarais, Harden and Boehmke2015; Handcock et al. Reference Handcock, Hunter, Butts, Goodreau, Krivitsky and Morris2018; Hunter et al. Reference Hunter, Handcock, Butts, Goodreau and Morris2008). SNA is also designed to capture both the temporal aspects of policy diffusion as well as the relation dimensions. Studies that relied on Event History Analysis, for example, were effective at answering questions related to the sequence of adoptions—i.e., who are “leaders” or “laggards” in diffusion processes—or using dyadic variants to assess how commonalities between two jurisdictions influenced the likelihood that they share policies. However, those studies were less successful at determining pathways of actual influence. There was a low level of certainty surrounding the ability to comment on who is actually influencing who in diffusion processes. SNA, by nature, is the study of relationships, transmission, and connections between entities (Barabási Reference Barabási2016). The use of graph theory paints a fuller picture of which states are influencing the spread of policies, who is receptive to policies, as well as how the entangled web of policy diffusion is constructed. The ability to both confidently visualize the relationships between jurisdictions and model the propensity for connections to be made provides SNA a distinct advantage over other empirical strategies that lack such capabilities (Desmarais, Harden, and Boehmke Reference Desmarais, Harden and Boehmke2015).

Most importantly, a network approach provides a means to speak to the broad patterns in diffusion processes in ways that decades of studies struggled to. We have learned a great deal regarding the characteristics of innovative states, the mechanisms of policy diffusion, the roles that politics play in the sharing of policies across jurisdictions, and how these dynamics interact with one another (i.e., Berry Reference Berry1994; Berry and Berry Reference Berry and Berry1999; Boehmke and Witmer Reference Boehmke and Witmer2004; Boushey Reference Boushey2010; Reference Boushey2016; Butler et al. Reference Butler, Volden, Dynes and Shor2017; Mintrom Reference Mintrom1997; Shipan and Volden Reference Shipan and Volden2008; Walker Reference Walker1969). However, the bulk of what we now know is the product of case studies of single-issue diffusion processes. Because of this dominant practice, “almost no significant progress has been made towards answering Walker’s (Reference Walker1969) second question about general patterns of policy diffusion” (Desmarais, Harden, and Boehmke Reference Desmarais, Harden and Boehmke2015, 394). Network analysis, however, provides a means to uncover persistent pathways of policy diffusion across a virtually limitless number of policies. As a result, scholars have been given a tool designed to tackle inquiries into the broader nature of policy diffusion in ways unmatched by early studies of innovation and diffusion.

A two-step process serves as the empirical foundation for the remainder of this study, allowing me to maximize the potential of an SNA application to examine the diffusion of noncongruent police reforms in response to the BLM movement. First, I call on the NetworkInference package in R to construct three different diffusion networks (Desmarais, Harden, and Boehmke Reference Desmarais, Harden and Boehmke2015; Linder and Desmarais Reference Linder and Desmarais2017). The focal network is one that depicts the diffusion of reforms from August 2014 through December 2020. The remaining two networks that depict the broad spread of policies and the diffusion of criminal justice policies leading into the movement will serve as points of comparison. Once these networks are constructed, analyze the factors that shape the likelihood for states to share policies using the Exponential Random Graph Model (ERGM), a common modeling strategy in SNA applications (Cranmer and Desmarais Reference Cranmer and Desmarais2011; Handcock et al. Reference Handcock, Hunter, Butts, Goodreau, Krivitsky and Morris2018; Hunter et al. Reference Hunter, Handcock, Butts, Goodreau and Morris2008).

Step 1: Inferring Policy Diffusion Networks

The NetworkInference package in R (Linder and Desmarais Reference Linder and Desmarais2017) package converts data on policy innovations into event cascades and infers directed paths of influence between enacting states using the netInf function. Ties are established between States_ij dependent on (1) the number of times state i adopts before state j, (2) the length of time between i’s adoptions and j’s adoptions, and (3) the precision by which an adoption by i predicts an adoption by j. The probability of a tie between is discounted in the case that states infrequently adopt similar policies, states take longer to adopt similar policies, and when states frequently vacillate in who is innovating policies first. The NetworkInference package allows users to adjust the threshold by which ties are established based on the probability that influence exists.Footnote ¹ I test various thresholds for each network (p < 0.1, 0.05, 0.01, and 0.001).

Constructing BLM-Era Noncongruent Reform Networks

To assess reformative era policy diffusion, I examine the spread of police reform during the #BlackLivesMatter movement (2014–2020). Since the killing of Michael Brown in Ferguson, MO, on August 9. 2014, states have worked to enact various policies designed to increase transparency and accountability around unwarranted uses of deadly force and reimagining police interactions. National Conference of State Legislatures data points to forty (40) states enacting one or more noncongruent policies across three major categories and twelve subcategories. Panel A in Fig. 2 maps the total adoptions by state.

Figure 2. Post #BlackLivesMatter noncongruent reforms

Twenty-six states have enacted policies to either require police to wear body cameras (16 states), study the effectiveness of body cameras (10), or allow for the public release of body camera footage (10). Twenty-seven states enacted laws designed to reduce or punish unjustified uses of force by defining or redefining proper uses of force (12), mandating training for the use of force (10), creating or improving mandatory reporting processes (14), or making it easier to investigate and prosecute unlawful uses of force (14). Lastly, twenty-six states enacted various policies geared toward reimagining police contact in communities across the country. Seventeen states downgraded many victimless and nuisance crimes, allowing for citations to be issued instead of arresting suspects. Ten states created programs to divert first contact for cases involving mental crisis to specialists better suited to respond to those types of distress. Three states restructured police involvement in K-12 educational settings, and four states criminalized making racially motivated 911 calls. Ten states enacted no laws.

Previous studies suggest that the underlying attributes of policies can greatly influence the rate at which they diffuse (Makse and Volden Reference Makse and Volden2011; Rogers Reference Rogers2003). A policy diffuses more rapidly if it has a relative advantage over the status quo policy, if it is compatible with the existing values and needs of the eventual adopter, if there are opportunities to experiment with the policy on a limited basis, or if the effects of the previously adopted policy are highly visible. Complexity, on the other hand, tends to slow adoptions across states. One will likely note that there is a great deal of variation in the types of reforms that spread in the aftermath of the movement and that these differences may influence the rate at which individual policies spread. For example, states more readily adopted the use of body cameras—largely because of their high degrees of trialability, observability, and a perceived relative advantage over the previous policy stance. Compare the speed and breadth of officer-worn body cameras to the slower establishment of community policing programs—a complex policy with less observable benefits and a much higher cost of implementation and experimentation.

Policy attributes can also influence the mechanism that drives innovation and diffusion (Makse and Volden Reference Makse and Volden2011). Spatial and learning modes of diffusion are enhanced when policies are observable, have a high relative advantage, and are compatible and diminished for highly complex and trialable policies. This may explain the variation in the different mechanisms that drove post-BLM reforms. For example, body camera usage spread, in no small part, due to top-down pressures from the federal government, which offered monetary incentives for states and municipalities that complied (U.S. Department of Justice 2015). Policies designed to reduce police contact, on the other hand, were not a solution advocated on behalf of the federal government—states were left to their own devices to determine how to reform those practices.

Beyond these considerations, states may develop a natural tendency to resist diffusion based on the degree of congruence—or noncongruence, in this instance—that a policy embodies. Decision-makers are often hesitant to adopt policies that threaten advantaged groups or are perceived to bestow undue benefits to deviant populations (A. Schneider and Ingram Reference Schneider and Ingram1993). This is almost certainly to manifest in response to BLM’s demands for reforms. A number of states may also resist enacting policies that threaten the existence of systems and structures that have come to establish and preserve racial hierarchies in America (King and Smith Reference King and Smith2005; Smith Reference Smith1996; Ture and Hamilton Reference Ture and Hamilton1992). There are also varying degrees of public support behind BLM reforms that bump up against many of the ideological, partisan, and racial cleavages in America (Azevedo, Marques, and Micheli Reference Azevedo, Marques and Micheli2022; Bonilla and Tillery Reference Bonilla and Tillery2020; K. Drakulich et al. Reference Drakulich, Wozniak, Hagan and Johnson2020; K. Drakulich and Denver Reference Drakulich and Denver2022).

While I see a great deal of value in parsing out the nuanced differences in the ways that each individual reform diffused through states, that is not the goal of this particular analysis. In this article, in particular, I am most concerned with characterizing the persistent pathways of influence across the full range of policies. I am less interested in comparing the speed of innovativeness across policies and discerning the individual mechanisms at play in the diffusion of police reforms across states. In addition to the need to restrain the topical focus of analysis, I am both advantaged and limited by the empirical strategy deployed in this work. On the one hand, I can speak to the broad patterns of diffusion and how the spread of police reforms differs from previous eras of criminal justice and general policy diffusions. However, because of my reliance on the NetworkInference algorithm, I cannot speak as confidently to the more granular aspects that influence the diffusion of individual policies.

To that end, Fig. 3 presents a directed network that maps the persistent pathways of diffusion from August 2014 through December 2020. The post-BLM reform network is established using a threshold (p < 0.1).Footnote ² The NetworkInference algorithm identified connections between 36 of 50 states—ten of the isolated states enacted zero reforms in the six-year period, and the remaining isolates only adopted a single policy. The most influential states in the diffusion of BLM reforms, according to the network, were California, Utah, Colorado, Maryland, and Texas. This network becomes the first point of analysis to be compared with previous policy adoptions from a more congruent era.

Figure 3. Inferred post-BLM diffusion network—Panel A represents the entire network, including isolate states. Panel B visualizes the connected component in the network. States are sized according to their outbound influence on other states

Constructing Pre-BLM Reform Networks

This case study is grounded in a comparative analysis between pre- and post-BLM policing policies using data from the State Policy Innovation and Diffusion Database (Boehmke et al. Reference Boehmke, Brockway, Desmarais, Harden, LaCombe, Linder and Wallach2020). This dataset catalogs the spread of 728 policies across a myriad of distinct substantive areas dating back to the 17th Century, making it the most comprehensive data collection on policy innovation available to date. The database allows researchers to test broad hypotheses on the mechanisms of diffusion or hone in on their focus by allowing them to subset the data into time splices or narrowing their concentration on specific topical areas. I will take on elements of both strategies by examining diffusion trends across the entire collection of substantive areas that spread between 1994 and 2014, as well as those in criminal justice reform during the same time.

This period is significant for two reasons. First, 1994 is the symbolic start to one of the more reformative periods of criminal justice reform. The infamous Clinton-era crime bill, rising crime, and civil unrest ushered in a period of punitive criminal justice policies. Second, the end of that period, 2014, symbolizes the start of a new era of criminal justice reform. The nation entered the first wave of the #BlackLivesMatter movement following the killing of Michael Brown in Ferguson, MO, on August 9, 2014. Studying this twenty-year span of diffusion cycles serves a practical function by providing a historical baseline to compare policies enacted due to the BLM movement.

Figure 4 presents the two resulting directed networks constructed using the NetworkInference process. The criminal justice diffusion network selected was at the traditional p < 0.05 threshold, meaning there is a 95% confidence that a pathway exists when a tie appears between States_ij. When testing at higher thresholds for this network, states began to drop out of the network (i.e. West Virginia at the p < 0.01 threshold). For the broader policy diffusion network, I selected the graph that was subject to the stricter p < 0.001 threshold (Fig. 4).Footnote ³

Figure 4. Inferred diffusion networks for broad and criminal justice policies from 1994 through 2014. States are sized according to their outbound influence on other states

Step 2: Modeling Policy Diffusion using the ERGM

I draw on the ERGM to examine the characteristics of the diffusion networks. The ERGM is frequently used to examine networks at nodal, dyadic, and structural levels (Butts Reference Butts2008; Handcock et al. Reference Handcock, Hunter, Butts, Goodreau and Morris2008; Reference Handcock, Hunter, Butts, Goodreau, Krivitsky and Morris2018). The foundation of the ERGM model is a multi-level logistic regression that evaluates the likelihood that a tie exists between States_ij. The ERGM takes this analysis one step further by allowing researchers the ability to identify endogenous structural characteristics that may exist in the broader network. Significance is determined if a characteristic exists in the observed graph to a degree that would not be uncovered at random.

I call on three standard functions within the ERGM to explore policy innovativeness and emulation. The first is the likelihood that a state adopts a position of influence within the network. In this case, influence represents a state that is the source or sender of a policy to another state. This is accomplished using the nodeocov function. I draw a straight line from the likelihood of a state to influence others to their broader propensity for innovativeness and to be senders, first-movers, or “leaders” in diffusion cycles. Second, I include a measure designed to capture the likelihood of a state being influenced by another state’s adoption. The nodeicov function is designed to identify exogenous trends that contribute to in-bound (receiver) edge formation. In terms of a diffusion network, this may signal that a state has taken on a position as a “laggard” (Desmarais, Harden, and Boehmke Reference Desmarais, Harden and Boehmke2015).

The third measure is homophily—the propensity for nodes in a network to form ties with nodes that are similar to them. Shared characteristics often lead to collaboration, communication, and shared political behaviors. Shared political beliefs, partisan affiliation, and gender and/or racial identity often increase the likelihood that political actors co-sponsor legislation, share information, and vote in a similar fashion (Bratton and Rouse Reference Bratton and Rouse2011; J. E. Campbell Reference Campbell1982; Craig et al. Reference Craig, Cranmer, Desmarais, Clark and Moscardelli2015; Fowler Reference Fowler2006b; Ringe, Victor, and Gross Reference Ringe, Victor and Gross2013; Victor and Ringe Reference Victor and Ringe2009; Victor and Koger Reference Victor and Koger2016). Relevant to this study, there is a long line of research that consistently finds state similarities—partisan, ideological, cultural, etc.—influence paths of influence, shared adoption, and even who states choose to learn from (i.e., Berry and Berry Reference Berry and Berry1990; Boehmke Reference Boehmke2009; Boehmke and Witmer Reference Boehmke and Witmer2004; Butler et al. Reference Butler, Volden, Dynes and Shor2017; Desmarais, Harden, and Boehmke Reference Desmarais, Harden and Boehmke2015; Shipan and Volden Reference Shipan and Volden2008). I use the absdiff (absolute difference) function for continuous variables and the nodematch function for categorical variables.

Key Variables

The variables of interest are three measures of state racial dynamics. The racial threat hypotheses expect innovation and diffusion to be dependent on the racial composition in a state. Therefore, I include a state’s average White population from the 1990 through the 2010 census. I also include a measure of the decline of the White population in a state from the same period to accommodate the racial diversification hypotheses. To assess the influence of Black representation on diffusion processes, I include the 2015 record of the percentage of Black lawmakers in state lawmaking chambers from the National Conference of State Legislatures survey.

In addition to the variables designed to account for state racial dynamics, I incorporate several measures that capture characteristics most often associated with innovation and diffusion. I account for the average statehouse ideology across chambers using the estimates from Shor and McCarty (Reference Shor and McCarty2011). Conservative states are those that fall on the higher end of their scale. I include the 2015 Squire (Reference Squire2007) measure to gauge the impact of legislative professionalism on the adoption and spread of policies, where state legislatures with higher measures are presumed to have more state resources, spend longer lengths of time in session, and have more staff at their disposal. Well-resourced legislatures often translate these features into a more innovative policy posture and regularly serve as a model to their peer states (Jansa, Hansen, and Gray Reference Jansa, Hansen and Gray2019; Squire Reference Squire2007).

Boushey (Reference Boushey2016) finds evidence that electoral competitiveness may contribute to noncongruent policy innovation. Therefore, I include Ranney’s most recent folded measure of electoral competitiveness (2010). The Ranney measure ranges from .5 (not competitive) to .99 (most competitive). To account for partisan control, I create a variable that captures the dynamic realignment that took place during this period. I include two variables for gubernatorial and legislative partisan control using a categorical scale—(non-partisan; solid Democrat, trending Democrat, trending Republican, solid Republican, competitive). I account for the potential that state contiguity impacts sending and sharing policies by including a matrix that indicates if states share borders.

There are subtle changes to the post-BLM models. The measures of racial dynamics essentially remain the same, except that I account for only the 2010 census measure in determining the racial composition of states and the change of racial composition from 2000 to 2010 in measuring racial diversification. There are, however, two new variables added to each model. Both are measures of a state’s influence in the historical broad and criminal justice diffusion processes. This is measured using each state’s out-degree centrality—or the number of outbound paths of influence on other states. I argue that (1) a state that has been historically influential in diffusion processes will be similarly influential in modern ones, and (2) past diffusion “leaders” will be more likely to influence past “laggards.” Given these assumptions, I expect to find a significant, positive relationship between the source measures and homophily measures.

In the pre-BLM ERGM models, I include three measures to capture the structural nature of the broader diffusion network. The first two account for the geometrically weighted in-degree and out-degree distribution. I also measure the network’s transitive nature by incorporating the geometrically weighted edgewise shared partners (GWESP). GWESP is a common measure of transitive clustering in single-mode networks. I also account for structural features of the network—using out- and in-star measures. These are typically used to account for network propensities to cluster in sparser networks. This also serves to improve model fit. Table 1 presents a table of the descriptive statistics of the key individual-level variables used in this study.

Table 1. Exponential random graph model results

***p < 0.001; **p < 0.01; *p < 0.05.

Findings

Table 1 presents the results of the Exponential Random Graph Models. Models 1 and 2 are evaluations of the pre-BLM broad and criminal justice diffusion networks, respectively. Model 3 is the examination of characteristics that shape post-BLM police reform. Each of the models is the result of 50,000 Markov Chain Monte-Carlo simulations following a burn-in of 1,000 permutations. The models are subject to standard goodness-of-fit tests, and each model fits reasonably well with each of the corresponding networks.

To better understand the movement’s role in shaping noncongruent policy, I begin by placing historical criminal justice diffusion within the context of broader diffusion to highlight its similarities and uniqueness. In many ways, criminal justice diffusion mimicked many of the patterns found in broader diffusion processes. The first pattern identified, which is both consistent between models and consistent with supporting literature, is that states will mimic those that are ideologically similar. Ideological homophily is a strong predictor if states share policies in both the broad diffusion network (coef. = −0.443; p < 0.01) as well as in the pre-BLM criminal justice network (coef. = −0.523; p < 0.01).

One dynamic that is often understudied is the influence of racial diversification on diffusion studies. While a small number of studies incorporate the racial composition of states influences the spread of policies, few (if any) have considered how changing racial dynamics in a state shapes innovation and emulation. In the broad network (coef.  = −0.061, p < 0.01), states experiencing the most precipitous declines in White population share have about a nineteen-percent probability of influencing another state, whereas states where the White population remained relatively steady had almost no chance of shaping other state’s policy adoptions. Like the broad network, states that experienced the greatest decline in White population share were most influential in the spread of criminal justice policies (coef.  = −0.082; p < 0.001). This equates to just under a three-percent decline in establishing an outbound tie to another state.

The broad and criminal justice networks also appear to share several structural characteristics. The positive, significant findings for the GW Out-degree measures suggest that there are relatively few states that are influencing diffusion. The negative, significant coefficients associated with the GW In-degree measure, however, suggest that influence is distributed broadly across both networks. Finally, the positive GWESP coefficients indicate that both diffusion patterns display transitive properties.

From Congruent Policies to Noncongruent Police Reforms

I proposed three potential pathways to enact noncongruent policies in response to demands from the #BlackLivesMatter movement. Model 3 presents results from the ERGM modeling of the diffusion network of those reforms. These findings can then be compared to the punitive-era criminal justice reform network to identify differences any differences that may exist.

Pathway 1: Reorienting Political Dynamics that Contribute to Innovation and Diffusion

The first proposed pathway suggests that the demands may be enough to operate on the political and social dynamics and motivate noncongruent innovations and diffusion. I find support for the proposition that the political dynamics that shape criminal justice policies shifted dramatically in the wake of the BLM movement. As previously discussed, criminal justice policy innovation during the pre-BLM era was dominated by conservative states (Fig. 5A). The projection suggests that this relationship may be quadratic, with some “liberal” states innovating punitive policies during this time (i.e., California and the “three strikes” laws). Alternatively, this may capture the rare instances where noncongruent policies cut through the dominant punitive, congruent culture (Boushey Reference Boushey2016).

Figure 5. Impact of statehouse ideology on pre- and post-BLM diffusion patterns

During the BLM era, noncongruent policies emerged largely—but not exclusively—from liberal states (coef.  = −1.294 p < 0.05). This supports assumptions derived from Boushey (Reference Boushey2016) and Blanton and Jones (Reference Blanton and Jones2021) that posit state liberalism as a source of noncongruent innovations. There is over a nine percent chance that the most liberal states influence any other state to adopt one of the twelve policies identified in this study. That probability decreases to zero for the most conservative states. This finding comes despite findings that Texas and Utah were among the most influential states (see Fig. 3). On average, the general tendency was for conservative states to lag behind their liberal counterparts. I also find evidence that political ideology also influenced how policies spread from state to state. The negative, significant coefficient associated with the nodeicov function of ideology suggests that liberal states were more receptive to the spread of noncongruent reforms (coef.  = −1.412; p < 0.05). Some may interpret these findings—taken together—that the diffusion of police reforms was siloed in only liberal states. However, this does not appear to be the case, given the nonsignificant findings associated with the ideological homophily measure.

Boushey (Reference Boushey2016) found that electoral competitiveness contributed to the diffusion of noncongruent criminal justice reforms. The negative coefficient associated with the nodeocov variant of that measure supports this notion—suggesting that electorally stable states may have been more inclined to lead the way in post-BLM reforms (coef.  = −5.567; p < 0.05). The positive coefficient with the nodeicov measure of competitiveness suggests that competitive states may be more likely to be influenced by other state’s innovations. However, the measure fails to reach traditional measures of statistical significance. Beyond ideological and electoral considerations, I do not find support that other political dynamics influenced the speed or spread of police reforms. Shared partisanship in the statehouse or governor’s mansion had little substantive impact on the likelihood that states enacted similar policies. Legislative professionalism also had no significant impact on the diffusion of noncongruent reforms.

Pathway 2: Black Lawmakers as Policy Entrepreneurs

There may need to be strategic actors with access to the decision-making process who must serve as entrepreneurs to initiate policy change. I focus my attention on Black state lawmakers, who I propose play central—yet different—roles in the diffusion of criminal justice policies across eras. First, I draw on historical and empirical accounts that contest Black lawmakers played a critical role in the spread of criminal justice policies during the punitive era. Model 2 supports the expectations that statehouses with a larger presence of Black lawmakers were more receptive to punitive-era innovations (coef. = 0.059; p < 0.05). At the federal level, Black lawmakers were front and center in advocating for the punitive policies that followed the 1994 Crime Bill (Peay and Rackey Reference Peay and Rackey2022; Young Reference Young2016). These findings may provide support for similar dynamics at the state level. There is an eight-percent difference in the likelihood that a state was influenced by an outside adoption between the statehouses where Black Americans are most represented and those where they are least represented.

Model 3, however, provides significant evidence that the role of Black lawmakers may have shifted from providing key support to adopting policies from other states to occupying entrepreneurial roles needed to spark innovation. States where Black lawmakers occupy more seats were more likely to be first-movers in enacting noncongruent police reforms (coef. = 0.122; p < 0.05). This relationship, too, may be quadratic, as states where African Americans make up ten percent of their legislators are twice as likely to lead the way in criminal justice policy diffusion than the most and the least diverse legislatures. This is likely a relic of the fact that many of the most diverse statehouses are in the American South—a region that is notorious for conservative reign and racially regressive policies. Aggressive policing, harsh sentencing, mass incarceration, and other policies have long been used as tools of oppression against minority communities, and these conservative states may serve as a “red wall” that no critical mass of Black representation can break through.

An additional finding supports this notion that Black lawmakers served as entrepreneurs in statehouses. Mintrom (Reference Mintrom1997) argues that entrepreneurs are able to reach across state lines and develop relationships used to fuel policy diffusion. Model 3 finding states that enjoy similar levels of Black representation are more likely to share policies (coef.  = −0.103; p < 0.05). This small, yet significant, finding suggests that there may be somewhat of a critical mass in Black representation that needs to be met in order for similar policies to spread from state to state. This finding would be consistent with prior research from Mansbridge (Reference Mansbridge1999), Tate (Reference Tate2014), and countless others that have examined the role and utility of Black lawmakers in political institutions (i.e., Gamble Reference Gamble2007; Reference Gamble2011a; Reference Gamble2011b; Minta Reference Minta2011; Minta and Sinclair-Chapman Reference Minta and Sinclair-Chapman2013; Peay and Leasure Reference Peay and Leasure2023; Swain Reference Swain1993).

Pathway Three: Divergent Diffusion Trends

The third line of inquiry suggests that there would be a distinct difference between historical criminal justice diffusions and the diffusion of noncongruent policies in the wake of the #BlackLivesMatter movement. In a previous section, I explored the similarities and differences between criminal justice diffusion patterns and broad diffusion patterns. While some of the mechanisms of diffusion were similar (i.e., ideological similarity), there are sharp differences in which states occupy central roles in the innovation of policies. Pathway 3 proposes that noncongruent diffusion will be more likely to resemble broad diffusion patterns than the diffusion of criminal justice policies of years past. Model 3 finds significant support for this notion (see also Fig. 6).

Figure 6. Influence of past diffusion processes on post-BLM reforms

First, I find that BLM did not have to seek out new innovators. They simply activated states that were already innovative outside of criminal justice diffusion networks. The broad network suggests that the most influential states in broad diffusion patterns from 1994 to 2014 are California, Texas, Virginia, Washington, Arizona, Florida, Illinois, and Utah. This is consistent with overlapping results from Desmarais et al. (Reference Desmarais, Harden and Boehmke2015) who examined pathways of influence from 1960 through 2009. Many of these states were among the top fifteen leaders identified in their study. The six most central states in the police reform diffusion process are California, Utah, Texas, Colorado, Maryland, and Connecticut. The directed ERGM model also provides evidence that leaders in past diffusion cycles are more likely to remain leaders in the diffusion of noncongruent police reform (coef. = 0.157; p < 0.05). This equates to a fifteen-percent increase in the probability of past leaders influencing another state in the police reform diffusion network.

It also appears that innovative states in broad diffusion processes were more open to taking cues on police reform from states than the laggard states in that same network (coef. = 0.193; p < 0.01). However, this does not mean that the diffusion of noncongruent police reforms was solely between innovative states. In fact, Models in Table 3 suggest past diffusions influenced both the speed and spread of police reform adoptions during the BLM movement. This would be supported, first, by a significant heterogeneous relationship between past “leaders” and “laggards”—meaning leaders would be more likely to influence policy adoptions of laggards than they are of other leaders or vice versa. Leader states were over ten times more likely to form ties with the most laggard states than they are with other leaders (coef. = 0.103; p < 0.05).

Discussion: Toward a Theory of NonCongruent Policy Diffusion

Much of the early criticism of the BLM movement centers around a perceived lack of tangible policy change (i.e., Szetela Reference Szetela2020). Few ground their critique of the movement on a foundation that acknowledges (1) how deeply entrenched punitiveness and congruence are in the culture of criminal justice policymaking and (2) the seemingly insurmountable challenges that exist for those seeking noncongruent policy change. This work finds the BLM movement captured the attention of the American public, media, and governing elite and brought demands of noncongruent change to their doorsteps. Those demands reshaped the discourse around the American policing system, challenged the social and political conditions that contribute to a culture of punitiveness, and activated strategic agents within institutions to spark policy change that diverges from the dominant policy culture.

I offer three important caveats concerning the limits to efforts to initiate noncongruent change. First, this proposed path is a perfect storm of events. So much of the process is dependent on each component being in place and effective. Removing any piece of the puzzle drastically decreases the likelihood of wholesale noncongruent action. Without demands for noncongruent action, there is very little that could motivate decision-makers to deviate from the norm of congruent policymaking (Fig. 7B). Those demands must be heard—the process fails should powerful groups effectively silence, neglect, or repress them with force. Next, the demands likely need to operate on both political and social dynamics. Reorienting social constructions may be meaningless if there is no political incentive to change course. Likewise, politicians are unlikely to give in to political incentives if the demands run counter to the values that they hold dear. The demands, and the effects that those demands have on political and social arrangements, may not be enough to spark innovation. They may need strategically placed actors to convert policy ideas to government action. Even then, single innovations are not enough to effectively challenge the dominant policy cultures across the states. Policies need to diffuse after being innovated.

Figure 7. The multiple pathways to noncongruent policy innovation and diffusion after the #BlackLivesMatter Movement: Panel A displays the expected process during typical congruent innovation and diffusion. Panel B outlines proposed pathways to noncongruent action

Because of the uniqueness of the BLM movement, it would be prudent for future studies to explore these dynamics across a broader range of noncongruent policies and, likely, more importantly, across a range of movements. The BLM movement represented the most significant challenges to what may be the most deeply entrenched congruent policy culture imaginable. It is likely that there may be some policies and movements that do not need this “perfect storm” to enact substantive noncongruent change. Future studies should examine the conditionality of each mechanism of change. It is also important to delve into questions of how the nature of demands (i.e., level of violence, systemic versus non-systemic, level of inclusivity, longevity, etc.) influence the conditionality of each mechanism.

The second caveat is that the effects of noncongruent policy innovations and diffusions are likely confined to the policy problem in question. One noncongruent policy and diffusion process does not necessarily dismantle the entire culture of congruent policymaking. Demands are problem-specific, and even if demands are received and acted upon, the processes are unlikely to bleed into other areas. Political elites are working constantly to confine conflict into small, manageable arenas (Baumgartner and Jones Reference Baumgartner and Jones1991; Cobb and Ross Reference Cobb and Ross1997; Rochefort and Cobb Reference Rochefort and Cobb1994; Schattschneider Reference Schattschneider1975; Smith Reference Smith1996). Issue-specific demands are unlikely to alter the political calculus and social norms across all policy domains. Congruence will likely remain the norm, even within policy domains.

Third, noncongruent policies are unlikely to permanently reconstruct the culture of congruence, particularly in the American policy culture where congruence is so deeply embedded. BLM in no way fundamentally changed policing policy culture from punitive to reformative. Noncongruent change is an anomaly. It is the occasional one-off that runs alongside the dominant culture of congruent policymaking. It is a brief deviation from the norm, not an abolishment of the norm. Policy congruence is systemic—it is not only reinforced by dominant societal values and norms but also drives action in institutions tasked with preserving those values and norms (Ture and Hamilton Reference Ture and Hamilton1992). The powerful are unlikely to entertain demands that could threaten the entire system of governance and the arrangements that legitimize the power they hold.

There is certainly resistance to noncongruent police reforms. The data suggest that much of this resistance can be captured in the hesitance on the part of conservative states to both initiate reforms and be influenced by diffusing policies. This could be semblance of a broader trend of conservatives’ reluctance to depart from congruent policies, as congruence may be more aligned with conservative values. Alternatively, the hesitancy to adopt noncongruent police reforms may be a sign that conservative states may be more committed to preserving the racialized, punitive culture that informs criminal justice practices than they are to remedying extrajudicial police killing of Black and brown citizens. Future studies should seek to parse out the difference between a state’s adherence congruence versus its investment into the political and social arrangements that dictate the dominant policy culture.

Readers will likely note that the state-level reforms that resulted from the BLM movement were moderate, at best—the burdens placed on police are mild. More punitive demands (i.e., “defund the police”) were nonstarters, as politicians on both sides of the aisle balked at implementing reforms that would burden police forces any further. In fact, some politicians actively campaigned against further burdening police and offered, instead, to funnel more money into police departments—including Democratic President Biden (Blake Reference Blake2022; Nickeas Reference Nickeas2022). This speaks directly to the boundaries of noncongruence—there are limits that politicians are willing to burden advantaged groups and benefit deviants. Future studies should test for any variance in innovation and diffusion caused by the degree to which advantaged groups would be burdened (or deviants benefitted) by noncongruent policies.

Noncongruent policies can also result in backlash—not only from the electorate or from the countermobilization of advantaged groups but from policymakers themselves. States may work to re-entrench congruent policies in response to the small deviation into noncongruent policymaking. This was certainly the case with the BLM movement. Mass arrests and incarceration were the general response to largely peaceful, constitutional protests. Several states enacted “blue lives matter” hate crime laws that, ironically enough, created protections for police as a “targeted minority group” being threatened by Black citizens (Mason Reference Mason2022). Others implemented policies designed to surveil and criminalize BLM protesters, enhance charges to felonies, and even allow citizens to physically harm protesters (Quinton Reference Quinton2021). A number of states actually passed laws protecting motorists who run over protesters with their vehicles (Andone Reference Andone2017). Scholars interested in policy cultures should investigate the re-entrenchment of congruent policies in response to noncongruent policy ventures.