Operationalizing a Conceptual Model of Colorism in Local Policing

Colorism is “the discrimination of individuals with darker skin” (Hunter, 2013; Russell, Wilson, & Hall, 2013). Burke and Embrich (2008) described colorism as a system of privilege and disadvantage that is based on “the lightness or darkness of one’s skin, with favoritism typically granted to those with lighter skin” (p. 17). It (colorism) is a nuanced form of racism that can be expressed from the conscious or subconscious psyche, and it is centered on individual decisions that are informed by perceptions of skin color. The phenomenon can occur even when the target’s race is unknown (Harris, 2008), and the discriminatory behavior can originate from the target’s assigned in-group or from an out-group (Herring, Keith, & Horton, 2004; Hochschild & Weaver, 2007) (Fig. 1).

Previous studies have found evidence of colorism in criminal sentencing outcomes (Eberhardt, Davies, Purdie-Vaughns, & Johnson, 2006; Viglione, Hannon & DeFina, 2011). For example, Viglione et al. reviewed 12,000 prison records of female inmates from 1995 to 2009 in the state of North Carolina. Each inmate’s record included a notation of their skin color category—“0” for non-light skin and “1” for light skin. The authors found that inmates coded as having light skin received significantly shorter sentences and had shorter periods of time served (Viglione et al., 2011). Where we lack analysis about this phenomenon is in local policing. We should expect to see similar evidence of colorism in local policing. However, most of our policing studies (Engel, Calnon, & Bernard, 2002; Epp et al., 2014; Leitzel, 2001) explore racial discrimination at the group level. Limited focus has been placed on the variability of discrimination within a target group, and what this level of specificity might tell us about local policing and how local policing behaviors might contribute to the perpetuation of bias throughout the criminal justice system.

Colorism offers an alternative explanation for some of the racial disparities that are perpetuated by public organizations, especially those that directly interact with the public on a routine basis. Therefore, I will use the concept of colorism as a vehicle for establishing some alternative approaches to understanding how racial discrimination might function during common interactions between public servants and citizens. The specific organizational context for this project is local policing in the USA. The previous research has provided cases of colorism in the more developed stages of the criminal justice process. However, I would suspect that the conditions of local policing—rapid decision making, face-to-face contact with the public and high stress—would further intensify the reliance on biases. If we can address such matters at the entry point of the criminal justice system, we may be able to reduce the magnitude of biased treatment throughout the system.

Prior to starting this work, I had a few informal conversations with police departments about the concept of colorism. Some departments expressed concern for managing the consequences of detected colorism. Their concerns are valid, and yet we do not have an appropriate response to such practical questions—if we detect it, then what? The second research question is an attempt to lay the groundwork for addressing this concern: RQ2—What might occur when managers apply the typical dilemmatic responses to address detected colorism? The computer simulations designed here are intended to simulate three common change strategies—incrementalism, transitional, and transformational—that an agency might assume when colorism has been detected in their organization. These strategies will be used to explore the conceptual model. Each change strategy will be described in detail in the Methods section of the article. The effort to answer this research question should yield some insights about the conceptual model and lay the groundwork for future hypothesizing.

While there is potential value in combining group and nuanced dynamics in our future studies and practices, we must first establish a few general understandings about how racial biases like colorism might function between members of the same organization and between the instigator and the target. To that end, a conceptual model was developed to help bridge disparate thought from multiple disciplines that have explored the concept of colorism. For the sake of brevity, an abbreviated explanation of the conceptual model is offered here. The conceptual model makes use of the available descriptions of colorism at the individual, interactive, and institutional level. The general premise of the model is that individual biases, like colorism, can influence a police officer’s decision-making model. By way of organizational socialization—framed here as a vehicle for interactive colorism—police officers are subject to mirror the biased decision-making models of their colleagues (Van Maanen, 1975). Overtime, these biased behaviors can become shared conventions (Douglas, 1986) with an end result of biased policing outcomes. A more detailed explanation of the rationale behind the conceptual model can be found in Appendix A. The conceptual model is also depicted in Fig. 2.

This section will be followed by the Methods section, which will describe how the conceptual model was operationalized into a simulation model. The Methods section will summarize the steps taken throughout the simulation. The subsequent section will include the results of the simulation/thought experiment along with some summary statistics, the major observations, and a few additional insights. This will be followed by the Discussion section where I will discuss the results, the significance of the research, the implications for research and practice, and the limitations of the study. The article will conclude with a summary of the research aims and contributions.

Concurrently, agent-based modeling (ABM) will be used to simulate the conceptual model (Iba, Matsuzawa, & Aoyama, 2004), challenge the tenets of colorism, and explore potential responses to detected colorism. ABM is a type of software programming that simulates decision-making models that are based on a set of simplified rules (Wilensky & Rand, 2015). These rules can be generated from rules of thumb, empirical evidence, descriptions of a concept, and/or a set of theories. With ABM, we can explore a collection of nonlinear behaviors (Bonabeau, 2002) and anticipated macro-behavior (Schelling, 2006). One way to imagine how ABMs operate is to think of a petri dish. Now, place all the theories that you think help to explain a very complex issue into the petri dish. Depending on the issue you have selected, your petri dish could contain hundreds of associated rules. What might emerge from your petri dish (ABM) are: (1) adaptive behaviors that you might not have noticed in a standard thought experiment; (2) anomalies—trends that run counter to the simulated theories; and/or (3) a collection of individual behaviors that form patterns that confirm/challenge an initial hunch (Wilensky & Rand, 2015). Just as we can grow a fungus in a petri dish, we can use ABM to further develop theories about emergent behavior.

ABM is appropriate for this project because it can be used to explore disparate knowledge within a safe environment. What can come of ABM thought experiments are plausible explanations for the proposed connections between various schools of thought (Helbing, 2012), which is the case for this project. From the literature, we have a grasp on how to describe individual and institutional colorism. However, the literature has yet to describe interactive colorism in a meaningful way. In addition, we have yet to explore the relationship between the three levels of colorism—individual, interactive, and institutional. We can use ABM to explore the proposed relationship between the three levels of colorism along with the other linkages that were established in the conceptual model. More importantly, ABM will allow us to conduct this exploration in an environment that will not put humans at risk.

There are a few predominate applications of ABM, which include models that focus on prediction, diffusion, and organizational design (Bonabeau, 2002). The model for this project applies ABM to an organizational context with a variant of diffusion behavior; the simulations will explore how might a collection of biased individual decisions, and the transfer of the bias among police officers, influence organizational outcomes. Concurrently, the thought experiment will simulate potential policy responses to detected colorism. In short, the ABM models will simulate a policy response to detected colorism, while colorism continues to spread between police officers. The decision-making models of police, both fair and biased, will be operationalized using rules intended to simulate individual, interactive, and institutional colorism. Before we shift to the next subsection, this last point is worth a second mention. The simulations are designed to challenge our descriptions of colorism, while exploring potential responses to detected colorism. These two things will occur at the same time.

This section provides a general overview of the simulated environment and how the thought experiment was operationalized. For each run of a simulation, citizens were assigned to a random space in the simulated environment. The only time citizens moved about is when they were arrested, when they were transferred to temporary detainment, or when they were transferred to permanent incarceration. Police officers randomly moved about the simulated space while seeking to make an arrest. Once all of the citizens were arrested, the simulation came to a halt. Greater detail will be provided in the following subsections.

The parameters of the model derive from “Appendix A.” For the sake of brevity, only the major parameters will be described in detail. Please see Table 1 to review a complete list of the model parameters and assumptions.

NetLogo 5.3.1—the selected ABM software program—was used to design the simulation model. The simulated environment was divided into three distinct zones. The first zone is designated as free society, which is representative of basic freedom. The second zone is the detainment area, which is an abstraction of all criminal justice functions except for incarceration (Minton & Zeng, 2016). The third zone is reserved for incarceration, which is representative of permanent imprisonment. Each zone contains several patches—segmented spaces (see Fig. 3). At all times, citizen-agents and police-agents will occupy a single patch within one of the three zones. If we liken a patch to the real world, it would be the space a person is occupying (e.g., the space a person is standing in) at any given moment. The rationale for setting the model up in this manner was to centralize the focus on how biased policing might impact individual freedoms. The three distinct zones were also used to operationalize colorism at the three levels. Individual colorism occurs as police officers engage the citizenry in the free society zone. Interactive colorism takes place by way of socialization (simulated learning) between police officers, and institutional colorism is the simulated outcome of the biased policing and the socialization of the bias.

The following descriptions provide the rationale behind the initial parameters for five (5) simulated policy responses (PR) to detected colorism. The five simulated policy responses represent what a police department might do after they have detected colorism within their organization; the detection is assumed to have occurred prior to the start of each simulation. The do-nothing response (PR1) simulates absolute colorism, and the utopian state simulates absolute fair policing (PR5). The do-nothing response might occur when a police department detects colorism but takes no action to address the issue. The simulations for the do-nothing response started with a police-agent count of 0 police and 6 c-police, and this ratio was maintained for the entire simulation run. The utopian state is the ideal state of having no skin color bias within the police department; no action is required by the police department. The simulations for the utopian state started with a police-agent count of 6 police and 0 c-police, and this ratio was maintained for the entire simulation run.

The other three PRs explored three change strategies that might occur after the detection of colorism—passive incrementalism (PR2), counterbalancing (PR3), and aggressive dilution (PR4). Passive incrementalism is meant to simulate incremental/developmental change. In the change management literature, incremental change has been described as slow, strategic, or evolutionary (Dunphy & Stace, 1988). With passive incrementalism, an organization might seek new hires to offset skin color bias, but the effort might lack adequate capacity to counter adverse policing behaviors. The simulations for the passive incrementalism approach started with a police-agent count of 2 police and 4 c-police. Counterbalancing is meant to simulate transitional change. Transitional change has been described as planned rearrangement, or a middle-of-the-road change strategy (Ackerman, 1986). With transitional change, a manager might attempt to match the number of previously hired biased police officers with an equal number of new hires who do not exhibit signs of skin color bias. The simulations for the counterbalancing approach started with a police-agent count of 3 police and 3 c-police. Aggressive dilution is meant to simulate transformational change. Transformational change has been described as radical, revolutionary, or traumatic (Ackerman, 1986; Dunphy & Stace, 1988). A manager might decide to outweigh the effects of biased policing. This might be achieved via retraining and/or a hiring process that detects and filters out candidates who would adversely tip the scale (Kindler, 1979). The simulations for the aggressive dilution approach started with a police-agent count of 4 police and 2 c-police. In terms of countering detected colorism, it was assumed that the policy response of passive incrementalism would be the least effective strategy and that aggressive dilution would be the most effective strategy. To get a sense of what the simulation model looks like with these parameters and assumptions incorporated, see Fig. 4.

Six conditional statements were formulated to explore the general assumptions about how colorism might function in an organizational setting. Burke and Embrich’s (2008) definition of colorism is focused on the two most extreme skin color groups, those who are defined as light- and dark-skinned. Therefore, the experiments conducted here were primarily centered on the citizen-groups of darks and lights. The following conditional statements were designed to explore the assumptions about the darks and lights in the simulated environments for passive incrementalism, counterbalancing, and aggressive dilution.

We should expect an increase in colorism to have the following result: (1) The darks will experience the highest incarceration when compared to the mediums and the lights; (2) the mediums will experience higher incarceration than the lights but lower incarceration than the darks; and (3) the lights will experience the lowest incarceration when compared to the darks and the mediums. The strategies to counter detected colorism should yield and equalizing effect: (1) lower incarceration for the darks; (2) slightly lower incarceration for the mediums; and (3) higher incarceration for the lights.

Behavior space was used to produce the output for the verification of the simulation model, as well as the output for the analysis. Behavior space is a built-in analytical tool for NetLogo. Behavior space can run multiple simulations based on a set of predetermined parameters, and it can capture the current state of each agent at each time-tick for each run of a simulation. Before conducting the analysis, two verification tests were conducted to ensure the model was able to simulate the two forms of policing—fair and biased (Forrester & Senge, 1980; Iba et al., 2004). First, I ran ten simulations of each policy response—the do-nothing approach, passive incrementalism, counterbalancing, aggressive dilution, and the utopian state. The output from these simulations was used to compare the average time it took each citizen-group to reach complete incarceration. Complete incarceration for a citizen-group occurs when all the citizen-agents from the group (e.g., darks) have been incarcerated. Based on the rules of the simulation, the citizen-group with the darkest skin—the darks—should reach complete incarceration at a faster rate than the mediums and the lights. Based on the results of this test, it was confirmed that the model was able to simulate biased and fair policing.

For a moment, let us view all five of the policy responses along a continuum that leads to total fair policing (see Tables 6, 7 in Appendix B). Within that context, we can see that the plight of the darks improved with each increase in the intensity of the policy response; each time there was a lower number of c-police in the scenario, the incarceration outcomes for the darks improved. In the case of the mediums, there is minimal variation in the incarceration outcomes across all five of the policy responses. For the lights, there is no variation until aggressive dilution is the selected strategy. We will discuss these points in more detail in the following subsections.

For the first part of the phase analysis, selecting on two-thirds of incarcerated darks (260) provided the opportunity to observe the effects of interactive colorism prior to complete incarceration. The observations notated here are depicted in Fig. 5. When compared to passive incrementalism, the lights experienced higher incarceration with the do-nothing approach. However, as the intensity of the policy response to detected colorism increased, the lights did experience a higher rate of incarceration. Contrary to the assumptions of the model, the mediums experienced the highest rate of incarceration with aggressive dilution and the second highest rate of incarceration with counterbalancing. Across all three of the key policy responses (PR2-4), as the intensity of the policy response increased, so did the incarceration rate for the mediums. Let us turn our attention to the most intensive policy response, aggressive dilution (AD). For AD, the incarcerated 260 darks made up 66% of the total population for darks. Yet, at the same time-tick, only 45% (177) of the mediums and 24% (97) of the lights were incarcerated.

The second part of the phase analysis observed what occurred after each police-to c-police conversion for each of the key policy responses (PR2-4). The observations notated here are depicted in Fig. 6a–c and Table 8 in Appendix B. Phase I is based on the initial settings of the simulations, which is the state (c-police-to-police ratio) before the first phase of interactive colorism takes place. Prior to the first police-to-c-police conversion, we see a drastic difference in the prison outcomes. At the end of Phase I, the darks made up 61% of the prison population (pp) in the simulations for passive incrementalism (PI), 61% of population for counterbalancing (CB), and 52% of the population for aggressive dilution (AD). For Phase I, the mediums made up 27% of pp for PI, 27% for CB, and 30% for AD. The lights made up 12% of the pp for PI, 12% of the pp for CB, and 18% of the pp for AD. Across all three policy responses, and at different degrees, this notable bias is sustained throughout the simulations. More than half of the darks’ total population are incarcerated by the end of Phase II for PI (55%) and CB (54%). However, only 32% (PI) and 33% (CB) of the total population of mediums are incarcerated at the end of Phase II. Considering each phase and the overall outcomes, mediums and lights had the most favorable outcome with passive incrementalism, and the darks had the most favorable outcome with aggressive dilution.

The conditional statements that were designed for this inquiry were based on the rationale that any action taken by a police department to counter detected colorism would result in a more favorable outcome than doing nothing at all. Ideally, these favorable outcomes would resemble an equalizing effect—less incarceration for the darks and more incarceration for the lights. This rationale ties back to the definition of colorism authored by Burke and Embrich (2008), which describes a system of privileges and disadvantages afforded to individuals based on their skin color. Based on the results outlined in Table 3, I was able to simulate CS₁, CS₃, CS₅, and CS₆. However, I was not able to simulate CS₂—passive incrementalism will increase incarceration for the lights—and CS₄—counterbalancing will increase incarceration for the lights. In short, I was able to simulate a reduction in incarceration for the darks across the three key policy responses (passive incrementalism, counterbalancing, and aggressive dilution), but I was not able to conclusively simulate a reduction in privilege for the lights.³

There was the proposed notion that passive incrementalism (CS₁), counterbalancing (CS₃), and aggressive dilution (CS₅) would reduce the incarceration rate for the darks. Respectively, each of these policy responses significantly decreased the incarceration rate for the darks. The second proposed notion was that passive incrementalism (CS₂), counterbalancing (CS₄), and aggressive dilution (CS₆) would increase the incarceration rate for the lights. However, passive incrementalism and counterbalancing had no significant effect on the incarceration rate for the lights. Aggressive dilution was the only policy response that increased the incarceration rate (∆ + 2) for the lights.

The first interpretation of these observations is related to how we think about the relationship between discrimination and privilege. To address this point, Burke and Embrich’s (2008) reference to disadvantages is likened to bias or discrimination. Given that all three of the policy responses—passive incrementalism, counterbalancing, and aggressive dilution—significantly reduced the influence on skin color bias for the darks and only one policy response significantly reduced signs of privilege (increased incarceration for the lights), we should question whether the two terms of privilege and discrimination should be used to define colorism. Privilege was programmed into the simulation model with the assumption that privilege might function as favoritism that can carry a less punitive interaction than universal fair policing. The observations imply that privilege is not bound to discrimination, and how we respond to both issues may require separate strategies. If this would hold true in the real world, coupling these two terms in our theory and practice decisions could prove to be problematic.

The words privilege and discrimination are such complex terms that including them into one description of a complex phenomenon (e.g., colorism) may misinform how we interpret the associated theories. The decoupling of these terms—privilege and discrimination—might help to streamline how we define colorism. This type of careful thought could help to reduce the chances of unintended collateral damage in studies that focus on bias at the individual and interactive level. The previous studies have taken similar care when selecting their research aims. For example, Blair, Judd, and Chapleau (2004) examined the relationship between Afrocentric facial features and criminal sentencing. The authors tied physical features to discrimination, not privilege. Within this frame, it can be assumed that privilege is functioning in the same environment. However, we cannot assume that the antecedents and the consequences associated with discrimination and privilege warrant the same treatment. Therefore, the same care that we apply to field experiments should be applied to the definitions we use to describe the phenomenon of interest.

A deeper look at the results from the phase analysis (see Table 8 in Appendix B) might help us to understand why a response to discrimination may not equate to a concurrent response to privilege. In Table 8, Phase II of the aggressive dilution model shows a 25%-point increase in incarceration for the darks and a 13%-point increase for the lights. The outcome here should have resembled more of an equalized result, less discrimination, and less privilege with the two outcomes collectively depicting equitable policing. This anomaly can best be addressed by asking questions that challenge the logic that undergirds the description of colorism. It is important to ask what we mean by privilege when we position the term as an alternative outcome to or as a by-product of a discriminatory system. The results have helped to point out some key flaws in how we use terms to describe complex behavior. The take-away here is that colorism is best described as scaled discrimination based on a skin color spectrum; colorism is not an “either or” phenomenon. In other words, justice for the target group does not equate to a reduction in privilege for outgroups.

The second interpretation is related to the first, but it is more grounded in practice than theory. Let us turn our attention back to Table 3. In the three PR comparisons between passive incrementalism, counterbalancing, and aggressive dilution, we notice that aggressive dilution significantly outperforms the other two policy responses. The results suggest that our efforts to counter contagion behaviors, in an organizational setting, will require an aggressive response. The simulated responses to detected colorism were up against the continual spread of skin color bias, so the more passive approach had minimal impact. While the observations for counterbalancing and passive incrementalism were favorable, these strategies would take longer to achieve the intended outcome and afford more space for police officers to learn adverse behavior (Van Maanen, 1975).

How about the elephant with the medium skin color? The predominant definitions of colorism create a vague understanding of how to describe those who fall in the middle of the skin color spectrum. Based on the definitions used for this project, we can assume that the mediums of the world would benefit from fair policing but also be subject to skin color discrimination. The previous point about combining convoluted terms may have contributed to the lack of focus on the middle of the skin color spectrum. The simulation model was designed to stagger discrimination so that the darks experienced the highest degree of discrimination, the mediums the second highest degree of discrimination, and the lights the least degree of discrimination. The results do reflect this staggered application of discrimination. However, it was also expected that the mediums would experience a decrease in incarceration, especially with aggressive dilution. Counter to the expectations of the exercise, aggressive dilution had no significant influence on the incarceration rate for the mediums. This outcome runs counter to how individual colorism was operationalized in the model, and it stresses the need for further development of our definitions of colorism. Most descriptions of colorism have focused on the two extremes of the skin color spectrum, the darkest and the lightest of types. Collectively, scholars have made the argument for more inclusiveness, yet our descriptions of colorism have neglected an entire segment of people—those in the middle of the skin color spectrum. Moving forward, I think there should be a conscious effort to address the two points made here about privilege and those who fall in the middle of the skin color spectrum. To that end, I offer a modified conceptual model of colorism which includes a reference to the mediums of the world, and it circumvents the distorted term of privilege. See Fig. 7.

This effort contributes to our knowledge in two ways. First, it contributes to the conceptual development of colorism. The article provides the reader with an overview of the conceptual dimensions of colorism, and the underlying theories are unpacked and applied to several rigorous thought experiments. We discovered opportunities to revisit the predominant definitions of colorism and offered a more manageable conception of the phenomenon. In addition, two action items were offered that should inform future research—decouple privilege and discrimination in our descriptions of colorism and construct definitions that are more inclusive.

In addition, the interactive colorism analysis demonstrated how biased individual decisions could metastasize within an organization. The previous research has provided evidence of individual and institutional colorism, but little is known about how we get from individual colorism to institutional colorism. With the help of agent-based modeling, and the contributions of several disciplines, this article offers a roadmap that fills in the conceptual gap between individual and institutional colorism. Wilder’s discreet mention of interactional colorism has not been explained in a way that can be measured. Each of the trials performed in this thought experiment will equip scholars with a starting point for hypothesizing the interactive aspects of colorism. While there is room for improvement, the conceptual model designed here can serve as a starting point for testing the proposed explanations for individual, interactive, and institutional colorism.

Keep in mind that the results of this study are based on simulated data. Any claims made here cannot be used to credit colorism for any portion of the known or the unknown cases of racial discrimination in the criminal justice system. The only argument that can be made with surety is that the inclusion of colorism in our dialogues about race has the potential to account for what is not being counted. A thorough understanding of colorism can only complement our efforts to explain, document, and address known racial disparities in the criminal justice system.

There are several opportunities to expand the model and our treatment of the concept of colorism. For example, the description of organizational socialization within a policing context might be difficult to apply to other types of organizations. The simulated unidirectional learning that was embedded in the function of interactive colorism might be unique to police departments. If we consider organizations that may not have such stringent standards of socialization, we would need to alter how we describe the function of interactive colorism. Future studies should consider the effects of a bidirectional flow of influence on decision-making models. Bidirectional learning could be modeled by coding the police-agents to learn from each other versus the top-down approach assumed here. In the bidirectional model, a biased police officer would be able to adopt a fair police officer’s decision-making model and vice versa. In fact, a model that compares both unidirectional and bidirectional interactive colorism would be a valuable contribution.

This thought experiment explored a limited number of theories and responses to colorism. Scholars should consider other frames, such as the effects of the return to free society, the outcomes of fair policing but biased return to free society, or any combination of the available policing and justice theories. The policy responses explored in this version of the model are the first steps toward understanding how colorism might function at the individual, interactive, and institutional levels within the context of street-level bureaucracy. Scholars should also try to transport the framework to other organizational settings that require direct interaction with the public (e.g., the provision of social services). In addition, the previous studies have focused on the extreme ends of the skin color spectrum. Studies that attempt to explore what happens to those in the middle of the skin color spectrum would add a key component to our discussion.

It would be premature to imply that this work warrants an immediate policy change. However, there are a few potential broader actions that could originate from this exploratory work. With some customization, the conceptual framework can be applied to any public agency that has direct interaction with the citizenry. Agencies that are limited by the current frames we use to discuss racial disparities could use this work to initiate nuanced dialogue within their organizations. In addition, the work could help to inform the design of mechanisms intended to detect colorism within organizations. Such mechanisms could help managers with decisions related to hiring, retraining, and retention.