Abstract
Positive public perceptions are a critical pillar of the criminal justice system, but the literature addressing them often fails to offer clear advice regarding the important constructs or the relationships among them. The research reported here sought to take an important step toward this clarity by recruiting a national convenience sample to complete an online survey about the police in the respondent’s community, which included measures of the process-based model of legitimacy and the classic model of trust. Our results suggest that although both are predictive, the models can be integrated in a way that allows the strengths of each model to address the weaknesses of the other. We therefore present this model as a first step toward an Integrated Framework of Police Legitimacy that can meaningfully incorporate much of the existing scholarship and provide clearer guidance for those who seek to address these constructs in research and practice.
Whether by coming forward with information about a crime or by refusing to obey directives, the actions of the public have important roles to play in facilitating or impeding the criminal justice system. Experts widely recognize that the effectiveness and efficiency of these institutions rely heavily on their ability to encourage cooperative behaviors that range from compliance with mandatory orders to voluntary engagement in advancing the institution’s goals (Skogan & Frydl, 2004). In the policing context specifically, this growing emphasis represents a deliberate move away from an older focus on instrumental approaches in which the police worked to foster public belief in their ability to control crime, largely through a fear of punishment. Increasingly, however, experts recognize that police are reliant on a truly engaged public that not only passively but actively consents to its ability to hold and wield power (Ramsey & Robinson, 2015; Skogan et al., 1999; Tyler, 2009). Indeed, it has become almost painfully clear that the coercive and aggressive tactics that previously dominated policing in the United States actually hinder public cooperation and often foster resentment and counterproductive behavior (Meares & Neyroud, 2015; Schulhofer, Tyler, & Huq, 2011). The riots in Ferguson (Missouri) and Baltimore (Maryland) are extreme examples, but the strain from which they arose is not constrained to major incidents such as these. In fact, a recent Gallup Poll found that national levels of confidence in the police reached a 22-year low, with substantially lower levels in communities of color (Jones, 2015).
In an effort to address this, modern policing has generally adopted a focus on improving public perceptions of law enforcement. As Attorney General Eric Holder (2014) noted, “. . . good law enforcement requires forging bonds of trust between the police and the public” (see also Obama, 2014) and this perspective has had a growing influence on the practice of contemporary policing (Ramsey & Robinson, 2015). In fact, it could be argued that virtually all modern approaches to policing claim a core emphasis on enhancing, or at least addressing, perceptions of law enforcement (Walker & Katz, 2005), and most contemporary national efforts to address the pervasive strain between the public and the criminal justice system have followed suit (e.g., the President’s Task Force on 21st Century Policing, the National Initiative for Building Community Trust and Justice, and the Community Engagement in the State Courts Initiative).
This situation creates an important policy window (Solecki & Michaels, 1994) through which researchers can have greater impact on the practice of policing. However, to best take advantage of this opportunity, scholars need to ensure that their work is on the cutting edge of understanding both the nature of public perceptions of the police and the mechanisms by which those perceptions shape relevant public behavior. As would be hoped, research has worked to meet this challenge. A considerable body of scholarship has investigated the social dynamics of the public–police relationship and consistently suggests that perceptions of the police are, in fact, major determinants of public cooperation (e.g., Jackson, Bradford, Hough, & Murray, 2010; Paternoster, Brame, Bachman, & Sherman, 1997; Tyler & Jackson, 2014). Scholarly accounts of the mechanism of this effect often center on what can be referred to as the process-based model of policing (top panel of Figure 1). 1 This account posits that because individuals care deeply about how they are treated by authorities, they rely heavily on their perception of the fairness of this treatment in determining the appropriateness of the authority’s power and, therefore, the extent to which it should be obeyed (e.g., Tyler, 2006b; Tyler, Goff, & MacCoun, 2015).
Models of Public Perceptions and Cooperation
Importantly, trust is also thought by many to play a major role in the appropriate administration of justice. Trust is widely recognized as a critical part of the process-based model, but its specific nature and place in the model are widely disputed (e.g., Gau, 2011; Jackson & Gau, 2016; Pryce, Johnson, & Maguire, 2016; Tyler, 2004; Tyler & Jackson, 2013). For clarity in understanding this construct, we therefore look to a widely influential theory of trust from social psychology. This account hypothesizes that trust, as a willingness to accept vulnerability, is driven by perceptions of the target’s trustworthiness and the trustor’s propensity to trust (Mayer, Davis, & Schoorman, 1995; see the bottom panel of Figure 1). Although initially posited as a model of trust in individuals, 2 the considerable body of work investigating this classic model suggests that it may also explain trust in a variety of institutions (see Schoorman, Mayer, & Davis, 2007), including the police (Hamm et al., 2016), but it has been largely ignored by criminal justice scholars.
The current study advances this literature in two ways. First, we simultaneously test the classic model of trust alongside the well-established process-based model of policing. As a result, the current analyses permit us to clarify the extent to which public cooperation is better explained by the propositions of the process or classic trust models. Importantly, however, we believe the situation will be somewhat more complicated than this. As we discuss in greater detail below, there is substantial overlap between the two perspectives. Although process-based model scholarship has traditionally focused on the particularities of the policing context while the classic trust model has emphasized the nature and drivers of the constructs more generally, there remains important overlap in their conceptualizations and operationalizations. We, therefore, also advance the literature by presenting and testing a theoretical model that brings together both perspectives into a single account. This integrated model capitalizes on the strengths of each perspective, and our results are suggestive of a theoretically and empirically sound framework that provides a clearer and more complete understanding of the process by which public perceptions of the police impact cooperation. As a result, the current work takes an important step in clarifying the nature and relationships among these constructs and, therefore, toward improving and protecting law enforcement’s relationship with the public.
Process-Based Model of Policing
The core argument of the process-based model is that the public’s attitudinal and behavioral reactions to criminal justice institutions are driven primarily by evaluations of the fairness of interactions with them (Tyler, 2006b; Tyler et al., 2015). Rooted in classic procedural fairness theory (Lind & Tyler, 1988), the process-based model hypothesizes a weaker role for outcome-focused, instrumental concerns than for the evaluations of the processes that determine those outcomes (see Pryce et al., 2016). In the context of policing, this suggests that although the public certainly attends to the outcomes of interactions with law enforcement, people are often more influenced by how they feel they were treated during the interaction. These procedural fairness assessments (also referred to as procedural justice or process fairness) have been consistently shown to affect public behavior and perceptions (e.g., Beijersbergen, Dirkzwager, & Nieuwbeerta, 2016; Benesh & Howell, 2001; Colquitt, Conlin, Wesson, Porter, & Ng, 2001; Franke, Bierie, & MacKenzie, 2010; Murphy & Tyler, 2008; Rottman, 2007; Tyler, 2006b) and are often more predictive than instrumental concerns such as the possibility of punishment or perceived effectiveness (Sunshine & Tyler, 2003b).
Numerous accounts of the components of procedural fairness exist in the literature (Bies, 2005; Greenberg & Lind, 2000; Leventhal, 1980; Tyler, 2000; see also Colquitt, 2001; Colquitt et al., 2001), and although consistency has been limited (Gau, 2011), measures typically revolve around a presence or lack of voice, respect, honesty, impartiality, accountability, and participation in public–police encounters. In an attempt to reduce these arguments to their essential differences, Tyler and Blader (2000, 2003; Blader & Tyler, 2009; see also Pryce et al., 2016) suggested that the components of procedural fairness could be categorized into two distinct but interrelated dimensions. The first dimension involves interpersonal issues and focuses on how legal authorities treat citizens during encounters. It, therefore, encompasses concerns like whether authorities are polite, dignified, honest, respectful, and responsive to citizens’ needs. Contrastingly, the second dimension focuses on how legal authorities make decisions and encompasses concerns about whether citizens are given voice, allowed to participate in the process, and whether authorities are neutral, impartial, transparent, and accountable for their actions
More recent work (e.g., Trinkner & Tyler, 2016; Tyler & Trinkner, in press) has highlighted a third dimension of procedural justice: the boundaries of authority. This work argues that people do not cede infinite power to authorities. Instead, they place limits on where and the extent to which an authority can exert its power. From this perspective, procedural justice encompasses issues of not only how police interact with citizens (i.e., treatment and decision-making concerns) but also when and where they interact with them and the types of power they wield (Trinkner, Jackson, & Tyler, 2016). When these limits are violated, individuals view the process as unfair and legal institutions as illegitimate, even when the interpersonal treatment and decision making were perceived more positively (Huq, Jackson, & Trinkner, 2016; Trinkner et al., 2016).
The process-based model goes on to suggest that these procedural fairness perceptions drive a belief that the authority itself is legitimate, that is, a sense that it should be in a position of power (Tyler, 2006b). Currently, there is considerable debate in the literature about how legitimacy should be conceptualized (see Bottoms & Tankebe, 2012; Johnson, Maguire, & Kuhns, 2014; Tankebe, 2013; Tyler & Jackson, 2013), but the analysis presented here is grounded in the work of Tyler (2006b) and Jackson, Bradford, Stanko, and Hohl (2013). Together, this scholarship emphasizes two central components of legitimacy judgments. The first is that legitimacy reflects a sense of shared values, or normative alignment, between the public and law enforcement (Jackson et al., 2013). From this perspective, legitimacy is an evaluation of interactions in which power holders make claims to the right to wield power over non–power holders (Beetham, 1991). Legitimacy is then earned when authorities behave in ways that are aligned with people’s beliefs, principles, and values regarding the way power should be administered in society. That is to say that it reflects a sense of aligned values between the public and law enforcement and subsequently promotes a normatively justified base of power. The second component of legitimacy reflects a felt obligation on the part of citizens to obey the authority (Tyler, 2006b). From this perspective, when people perceive an authority to be legitimate, they grant it the right to dictate appropriate behavior and to command others (Weber, 1968). As a result, they feel obligated to abide by the authority’s directives and to follow its rules.
It is important at this point to note that trust is considered by some to be another component of police legitimacy (e.g., Sunshine & Tyler, 2003b), but recent scholarship in the area points convincingly to this as being unjustified empirically (Gau, 2011, 2014; Reisig, Bratton, & Gertz, 2007; Tankebe, 2013). As a result, the current study operationalizes legitimacy using only the first two components. Nonetheless, given its centrality to the arguments of the current research, we will return to this issue.
Legitimacy serves as a basis of positive attitudinal and behavioral reactions, thereby promoting a healthy and mutually beneficial relationship between legal authorities and the public they serve (Jackson et al., 2013; Skogan & Frydl, 2004; Tyler, 2006a, 2006b; Tyler & Huo, 2002). When citizens view the police as legitimate, they are more likely to work with them (Bradford, 2012; De Cremer & Tyler, 2007; Tankebe, 2009; Taylor & Lawton, 2012; Tyler & Fagan, 2008) and comply with laws (Jackson et al., 2012; Tyler, 2006b), even when the police are not physically present (Sunshine & Tyler, 2003b). Indeed, the link between legitimacy and a wide variety of positive reactions to the police has been consistently supported by an impressive body of research conducted around the globe (e.g., Bradford, Huq, Jackson, & Roberts, 2014; Jackson, Asif, Bradford, & Zakar, 2014; Jackson et al., 2013; Jonathan-Zamir & Weisburd, 2013; Murphy, 2004; Tankebe, 2008, 2013; Tankebe, Reisig, & Wang, 2016; Tyler, 2006b; Van Craen & Skogan, 2015).
Classic Model of Trust
Trust has long been a construct of interest for scholars in a number of contexts, but arguably, the most developed of these literatures is the scholarship addressing organizational trust. Social psychologists with expertise in organizations—groups of individuals with a common goal—have focused heavily on the mechanisms by which these individuals can be encouraged to cooperate optimally and trust has been widely recognized as critical for this process. Although several somewhat distinct perspectives exist (see PytlikZillig & Kimbrough, 2016), the most widely replicated and certainly most accepted account of trust in this (McEvily & Tortoriello, 2011) and, increasingly, other contexts (Earle, 2010; Levi & Stoker, 2000) was seminally posited by Mayer and colleagues (1995). Their model argues that, when facing a situation in which there is some potential for broadly defined harm, trust is the psychological state within the trustor that predisposes them to act in ways that actually accept their vulnerability. This willingness to accept vulnerability is driven primarily by the trustor’s evaluation of the target’s worthiness of that trust, but the model also identifies a smaller role for dispositional propensities within the trustor. Although initially posited for the interpersonal level (individual to individual), this account of trust has been tested across levels in the organizational context (individual to organization and between organizations) and has been consistently supported (Schoorman et al., 2007; see also Colquitt, Scott, & LePine, 2007). In addition, scholarship from other contexts suggests that this operationalization is explanatory in a number of other relationships (e.g., Burns, Mearns, & McGeorge, 2006; Carter & Bélanger, 2005; McKnight & Chervany, 2005; Van Slyke, Bélanger, & Comunale, 2009), leading to an argument that it may be ideally applicable across situations (Hamm et al., 2016).
The central concept within this classic model of trust is vulnerability (Bigley & Pearce, 1998; Nienaber, Hofeditz, & Romeike, 2015; Pirson & Malhotra, 2011). Vulnerability is fundamental to all human interactions (Lind, 2001) and is widely recognized as a necessary condition of trust (PytlikZillig & Kimbrough, 2016; Rousseau, Sitkin, Burt, & Camerer, 1998). From the actions of parents regarding their children to those of a government regarding its constituency, the target of trust often has some level of agency or ability to make decisions that impact the actual or perceived probability that harm would come to the trustor. For example, parents can abandon their children and governments can disregard or act against the interests of their constituency (Frederiksen, 2012). In the organizational context specifically, the agency of the target creates a potential for harms like a supervisor inappropriately firing a hardworking employee or failing to recognize an individual who frequently compensates for lazy coworkers (Robinson, 1996).
Vulnerability refers to the potential for harms that run from concrete, individual injury to more nebulous psychological harms that can affect larger segments of society (Sarewitz, Piekle, & Keykhah, 2003; Schmidtlein, Deutsch, Piegorsch, & Cutter, 2008). Regarding the police, vulnerability most obviously includes the potential for justified personal harms ranging from getting a ticket or being arrested, to more serious and unjustifiable harms such as experiencing excessive violence, bias, or disrespect at the hands of the police. In addition, however, vulnerability in this context would go so far as to include more amorphous/abstract harms such as violations of beliefs regarding the appropriate role of the police in society. Consider, for example, an ethnic majority member who believes that the police consistently respond disparately to minorities. For this individual, the salient vulnerability is not likely to be that she would personally be mistreated by the police. Instead, the salient harm may be a belief that the police are violating notions about their appropriate role as protectors of a fair and just society. Vulnerability to the police is therefore relevant, not only for individuals—such that officers have the agency to justifiably or unjustifiably cause harm to specific members of the public—but also for society generally when the harm centers on more abstract but no less important concerns. When trust is lacking, these vulnerabilities can become insurmountable barriers to a positive relationship. When it is present, trust provides a mechanism by which these concerns can be overcome by creating a psychological state within the trustor that disposes him or her to accept the potential for harm. This is especially important for policing because the genuine need for considerable discretion makes it unlikely that the public will be able to significantly reduce its vulnerability. When the public is willing to accept that potential for harm, however, trust can still facilitate a positive relationship.
This account of trust also draws a fundamental distinction between trust itself and the drivers of that trust. That is to say that within this model, trust refers specifically to the psychological state and not the attitudes or cognitions that form its basis. These bases are instead referred to as trustworthiness, a multidimensional construct that encompasses the characteristics of the target that encourage a belief that it can or should be trusted (Dietz, 2011; Dietz & Den Hartog, 2006; McEvily & Tortoriello, 2011; Pirson & Malhotra, 2011). Scholars have, therefore, argued that trustworthiness incorporates a wide variety of evaluations, including identification, reliability, transparency, and, in some models, even procedural fairness itself (for a review, see PytlikZillig et al., 2016). Within the classic model, however, trustworthiness is made up of evaluations of the target’s ability (the technical competence of the target to do what it is being trusted to do), benevolence (a belief that the target cares about the trustor), and integrity (a belief that the target adheres to an internal moral code that the trustor finds acceptable; Mayer et al., 1995). Although numerous other constructs have been proposed and tested, these three typically account for considerable independent variance in the trustor’s willingness to accept vulnerability to the agency of the target (Colquitt & Rodell, 2011; Colquitt et al., 2007).
The final major construct of the classic model of trust is the individual’s personality-trait-level propensity to trust “most people” (also called dispositional trust). This predisposition is important because it provides a baseline level of trust that is afforded a new target when specific trustworthiness information is not available (Hamm, PytlikZillig, Herian, Tomkins, et al., 2013). As a result, it is often most predictive when the trustor is especially lacking in knowledge and experience (sophistication) with the target (Hamm, PytlikZillig, Herian, Bornstein, et al., 2013; McKnight, Cummings, & Chervany, 1998).
Distinct Models?
The process and classic trust models paint slightly different pictures of the critical constructs for linking perceptions of the police to reactions to them. Fundamentally, the process-based model highlights procedural fairness as the primary driver of legitimacy such that when the police treat people fairly, the public will see them as valid authority figures. Contrastingly, the classic model highlights trustworthiness as the primary driver of trust such that when the police are viewed as trustworthy, the public will be willing to accept their vulnerability. These literatures, therefore, seem to suggest that these models are somewhat distinct. We argue that it is likely that there is considerable overlap, especially given that humans tend to be motivated to hold relatively unified evaluations of a single target (Quinn, Macrae, & Bodenhausen, 2003). Problematically, however, the state of the literature provides no clear guidance as to how the perspectives could most profitably be integrated. As a result, it remains as likely that one model essentially leads to the other as it is that the mechanisms from each more directly interact. For example, it could be that procedural fairness fosters legitimacy which creates a perception of trustworthiness that leads to trust. Alternatively, trustworthiness or trust itself could color the perception of fairness such that judgments of procedural fairness or even their subsequent effect on legitimacy are dependent on the degree to which people view the police as trustworthy or actually trust them.
We believe, however, that much of the difference between the two models is the result of the distinct scholarly perspectives that they come from, rather than fundamental differences in the processes themselves. Thus, the best way to integrate these two perspectives may be to combine the constructs that are theoretically similar, while maintaining the distinctiveness of those that are more unique. Of the four major constructs in these two models, procedural fairness and trust are likely most separable (at least as conceptualized here). As discussed above, procedural fairness refers to evaluations of previous or expectations of future interactions with the police. As a result, while some conceptualizations of procedural fairness push beyond the specific interaction (e.g., global procedural justice; Gau, 2014), these concerns typically focus on interactions and should therefore be distinct from the internal psychological state of trust. This may also be true for trustworthiness and legitimacy as neither of these constructs address the interactions themselves but are instead influenced by them, thereby suggesting that they must occur later in the process.
Where there is likely much more potential for overlap is between legitimacy and trustworthiness as, at their core, both constructs are evaluations of the target (PytlikZillig et al., 2016; Tankebe, 2013). This is especially noteworthy when contrasted with procedural fairness, a perception of the interactions (Tyler, 2006b), and trust, an internalized psychological state within the trustor that arises from these evaluations (Mayer et al., 1995). Thus, trustworthiness and legitimacy occur at roughly the same point of the process (after the evaluation of the interaction but before the internalization) and can be distinguished from procedural fairness and trust on at least that basis. Further evidence of the potential for overlap between legitimacy and trustworthiness arises from the fact that the two constructs often overlap in their conceptualizations. Although legitimacy (an evaluation of the target’s “right” to authority) and trustworthiness (an evaluation of whether that the target is worthy of trust) are conceptually distinct, closer inspection suggests that this may not always be the case, especially in the minds of the public. It could be argued that because government’s authority creates a power differential with the governed, it will always create vulnerability. Thus, acknowledging the authority of any agent of government may carry with it, at least a tacit preparedness to accept the potential for harm. From this perspective, the line between legitimacy, trustworthiness, and even trust itself becomes difficult. That is to say that perceiving an authority to be legitimate may infer that the individual has identified some level of trustworthiness because it is difficult or even impossible to acknowledge the authority of an institution that is perceived to be unworthy of trust.
An additional argument for the overlap between legitimacy and trustworthiness follows from discussions of the nature of trust itself within the process-based model scholarship. Despite its central place in most discussions of the process-based model (e.g., Tyler, 2001, 2004; Tyler & Huo, 2002), trust’s conceptualization, measurement, and even its role relative to procedural fairness and legitimacy remain unclear (Gau, 2011; Jackson & Gau, 2016; Reisig et al., 2007; Tyler, 2004; Tyler & Jackson, 2013). Reflecting a wider imprecision in the discussion of trust across criminal justice research generally (Cao, 2015), process-based model scholarship often conflates it with confidence, satisfaction, and legitimacy (Jackson & Gau, 2016) such that the constructs are often treated as interchangeable, even in relatively recent work (e.g., Tyler & Jackson, 2014). Of the process-based model scholarship that has been more precise, trust is most often discussed as a component of legitimacy such that an individual who perceives the police to be legitimate must feel some level of trust in them (Chrusciel, Wolfe, Hansen, Rojek, & Kaminski, 2015; Reisig & Bain, 2016; Tyler & Huo, 2002; Tyler & Jackson, 2014; Wolfe, Nix, Kaminski, & Rojek, 2015). Importantly, however, a growing body of empirical (e.g., Gau, 2011, 2014; Tankebe et al., 2016) and theoretical (e.g., Bottoms & Tankebe, 2012; Jackson & Gau, 2016; Tankebe, 2013) work challenges this, suggesting that at least obligation to obey is too distinct from trust to be a component of the same construct. Bottoms and Tankebe (2012) drew a distinction between trust and legitimacy such that while legitimacy is primarily concerned with the present appropriateness of the authority, trust is concerned with expectations of its future behavior. Jackson and Gau (2016) followed on this distinction, suggesting that trust is “the subjective judgment that a trustor makes about the likelihood of the trustee following through with an expected and valued action under conditions of uncertainty” (p. 53), but in so doing, they exposed a potentially more profitable distinction between the constructs than using the temporal focus alone. In citing organizational research that argues that trust requires a “leap of faith” (Möllering, 2001, p. 404; see also Bradford, Sargeant, Murphy, & Jackson, 2017), they suggest that trust is actually wrapped up in an evaluation of the “intentions and capabilities” of the police (Jackson & Gau, 2016, p. 53) and two other common conceptualizations of trust in the process-based model literature follow a similar line of reasoning. Specifically, Tyler (2005) referred to motive-based trust, which “involves inferences about the motives and intentions of the police” (p. 325), and institutional trust, which is a belief “about the degree to which the police are honest and care for the members of the communities they police” (p. 324; see also Pryce et al., 2016; Tyler & Blader, 2000; Tyler & Huo, 2002). Thus, although somewhat different, the various flavors of trust in the process-based model literature consistently refer to evaluations of the characteristics of the police, and this conceptualization lines up well with the notion of trustworthiness from the classic trust model. That is to say that both of these literatures highlight the importance of evaluations of the target that provide a basis for a positive relationship, and the most salient difference between them is that in process-based model scholarship the construct is referred to as trust while the work addressing the classic trust model calls it trustworthiness.
Trust as discussed in the process-based model scholarship is therefore a distinct construct from that discussed in the classic trust model scholarship. As reviewed above, trust in the process-based model scholarship generally refers to an evaluation of the target while trust within the classic model scholarship refers directly to the psychological state that follows these evaluations. Indeed, the classic model notion of a willingness to accept vulnerability only shares noteworthy overlap with Tyler’s mention of empowering the police (Sunshine & Tyler, 2003a, 2003b; Tyler & Degoey, 1995; Tyler & Mitchell, 1994). The major difference between the constructs, however, is the point in the process in which they occur. Although both constructs are discussed as states within the individual, empowerment is typically measured with items that specifically address reactions to the police (e.g., “The police should have the right to stop and question people on the street”; Sunshine & Tyler, 2003a, p. 546). Thus, although the constructs overlap in principle, empowerment assumes the state within the individual while measuring the reaction. Trust, on the contrary, focuses specifically on the internalized state. The relationship between the constructs is, therefore, most accurately stated by saying that trust is a psychological state that is characterized by a willingness to accept vulnerability by, among other things, empowering the police.
The Current Research
As highlighted by our review above, the state of the literature is such that there remain important questions regarding the process by which perceptions of the police influence behavior. The current research sheds additional light on these questions by, for the first time, simultaneously testing two major accounts of this process using data from a national convenience sample. This simultaneous test allows this work to speak to the account that best addresses this process but has an additional benefit in its ability to shed light on their distinctiveness as there is good reason to believe that at least trustworthiness and legitimacy may be less distinct than the other constructs in the model. If supported, this may suggest that the most profitable integration of the two approaches may be to move legitimacy and trustworthiness to the same point in the process. Thus, procedural fairness would lead to legitimacy and trustworthiness which would, in turn, impact trust and then cooperation.
Method
Participants
To address the relative efficacy of the models and their potential integration in this context, a national convenience sample of participants (n = 610) was recruited using Mechanical Turk (MTurk) during the spring of 2015 to complete an online survey. MTurk is an online service provided by Amazon.com to connect survey takers (Workers) to survey administrators (Requesters) by posting tasks for Workers to complete and facilitating their payment. To reduce the likelihood of attention and motivation issues, participation was restricted to Master Workers (an MTurk designation for Workers who have received sufficiently positive ratings from previous Requesters). Because one common reason for rejecting MTurk survey data is evidence that the participant did not take the materials seriously (e.g., uniform responding, missed attention check questions), this rating likely helps in identifying participants who will provide quality data and is therefore typically associated with higher payments.
Although the use of MTurk as a data collection strategy is relatively new in social science, prior work has shown MTurk samples to actually be more representative of the general population than typical convenience samples (Buhrmester, Kwang, & Gosling, 2011). Despite this, it is important to note that Workers are not selected using the random or representative sampling techniques necessary for properly computing levels of measured variables that can be generalized to a specific population. Because our goal was to examine interrelationships among measured variables, however, the use of an online convenience sample like that provided by MTurk is not only sufficient but potentially optimal given the trade-off between cost and representativeness (see Berinsky, Huber, & Lenz, 2012). The survey consisted of an informed consent and measures of knowledge and experience with the police, the process-based and classic trust model constructs, cooperation, and demographics. The complete survey took approximately 20 min and participants were compensated US$2 for their time.
The majority of the sample self-reported as White (81%) with a household income less than US$60,000 per year (83%), either a partial (39%) or completed bachelor’s degree (34%), and had a mean age of 38 (SD = 10.8). A plurality identified as Democrat (44%; 17% Republican and 30% Independent) and the sample was somewhat more liberal than conservative on social (67% strongly liberal, liberal, or somewhat liberal) and economic issues (49% strongly liberal, liberal, or somewhat liberal), and in general (55% strongly liberal, liberal, or somewhat liberal). To account for the potential influence of the current public climate regarding the police, participants were also asked whether they had been involved in any of the antipolice sentiment on social media (e.g., by posting antipolice statuses or articles or by joining antipolice groups) or had been involved in an antipolice protest. Nearly all of our participants reported they had not (93% and 99%, respectively).
Measures
Construct measures for the process-based model (procedural fairness and legitimacy) and the classic trust model (trustworthiness, propensity to trust, and trust) were taken or amended from existing measures in the literature (see Table 1 for survey items). Procedural fairness was measured with nine items that asked participants to respond regarding whether police overstep their boundaries, their interpersonal treatment of the public, and decision making. In line with previous literature, legitimacy was measured using two distinct constructs. First, following traditional process-based model research (e.g., Tyler & Huo, 2002), participants’ obligation to obey was assessed with three items tapping into their internalized motivation to comply with the police. Second, following more recent work (e.g., Jackson et al., 2013; Tyler & Jackson, 2014), participants’ sense of normative alignment with the police was assessed with three items tapping the extent to which respondents believed the police shared their values.
Model Construct Measures
Note. All loadings were significant at p < .001. CFA = confirmatory factor analysis; var = variance; PUF = procedural unfairness; PF = procedural fairness; NA = normative alignment; TW = trustworthiness; Prop = Propensity to Trust; Oblig = Obligation to Obey.
Items scored on a 1 (never) to 5 (always) scale. bItems scored on a 1 (strongly disagree) to 7 (strongly agree) scale.
Regarding the classic trust model, trustworthiness was measured using a three-item scale that assessed the central components of ability, benevolence, and integrity (see Colquitt et al., 2007). Trust was measured using a three-item scale that probed the participants’ willingness to accept their vulnerability to police authority (derived from Mayer & Davis, 1999). Propensity to trust was measured using three items that assessed the participant’s beliefs about the motivations of “most people.” These items are commonly used in the General Social Survey and in other research on the legal system (e.g., Hamm, PytlikZillig, Herian, Bornstein, et al., 2013).
Regarding our primary dependent variable, cooperation, we used three slightly different measures (see Table 2). Our first, General Cooperation, asked participants how likely they would be to engage in a variety of cooperative activities that would advance the practice of policing (unpublished measure developed by the University of Nebraska–Lincoln Interdisciplinary Trust Team). The second measure, Specific Cooperation, focused on participants’ likelihood of assisting the police in their responsibility of controlling crime (Tyler & Jackson, 2014). We also utilized a third measure, Reaction to Crime, which asked participants to respond regarding how likely they would be to engage in one of three specific behaviors after witnessing a crime (Tyler & Fagan, 2008). The three measures were correlated, but analyses suggested they only shared approximately half of their variance (R2s < 56%) such that they measured somewhat different kinds of cooperation.
Cooperation Construct Measures
Items scored on a 1 (strongly disagree) to 7 (strongly agree) scale. bItems scored on a 1 (very unlikely) to 7 (very likely) scale.
Analytic Strategy
Data were analyzed in latent variable analyses in Mplus (Version 6) using the maximum likelihood robust estimator, which includes a correction factor for slightly nonnormal data (in the presence of normal data—when the correction factor is equal to one—the results converge to those of maximum likelihood) (Kline, 2005). Latent models are an ideal approach to this work because they not only provide structural (simultaneous) tests of construct relationships but also provide p value tests of how well hypothesized item-level relationships account for the actual covariance in the data (i.e., global fit). Although the most common index of this global fit in early structural equation modeling (SEM) work was the chi-square test of exact fit (where nonsignificant values are indicative of good fit), these tests are often particularly susceptible to sample size and model complexity such that a nonsignificant value becomes less likely in larger and more complicated models regardless of the actual fit to the data (Kline, 2005). As a result, contemporary researchers are encouraged to evaluate alternative global fit indices like the comparative fit index (CFI) and Tucker–Lewis index (TLI; for which values greater than .95 are indicative of good fit), the root mean square error of approximation (RMSEA) or test of close fit (for which values less than .06 are indicative of good fit), and the standardized root mean square residual (SRMR; for which values less than .08 are indicative of good fit; Hu & Bentler, 1998). Upon achieving sufficient global fit, researchers are next encouraged to evaluate indices of local misfit (specific relationships between items or factors that were hypothesized by the model to be stronger or weaker than they actually are in the data). One of the primary indices of local misfit reported in Mplus is modification indices that suggest specific model adjustments that would result in a change in model fit greater than a specified value (available via the MODINDICES command). Notably, the literature does not provide specific guidance regarding when a recommended change is “big enough” to warrant evaluation. Instead, researchers are advised to balance the statistical and theoretical implications of the inclusion of any recommended modification (Kline, 2005).
When warranted, nested model comparisons were conducted using the difference in the null model’s log likelihood scaled to approximate a chi-square value (−2ΔLL) as a function of the change in degrees of freedom across models. It is important to note that all changes that involve adding model constraints mathematically must result in decreases in fit while changes that release model constraints must result in increases in fit. Using the scaled log likelihood difference test, however, allows researchers to determine whether the changes in global fit are statistically significant. Nonnested model comparisons were evaluated using the models’ Bayes Information Criterion (BIC; Schumacker & Lomax, 2010).
Results
Construct Analyses
Data from the six construct measures were initially subjected to confirmatory factor analysis (CFA) with each of the 24 items entered as indicators of their hypothesized latent factor. Factors were identified by setting the first item as a marker such that the latent factor took on the item’s mean and variance. All other loadings were estimated as were all correlations among the latent factors. The hypothesized model fit moderately to the data, χ2(237) = 810.62, p < .001; CFI = .94; TLI = .94; RMSEA = .06, p < .001; SRMR = .05, and suggested that all 24 items loaded significantly on their hypothesized factors. Evaluations of the loadings suggested that while most items loaded on their hypothesized factors greater than .7, the last Procedural Fairness item did not (λ [standardized loading] = .43, p < .001). Further evaluation of the modification indices revealed that a subset of Procedural Fairness items was, in fact, more related to each other than they were to the rest of the construct items, suggesting that the construct was importantly multidimensional. Evaluation of the items themselves highlighted methodological distinctions among them such that three items were all negative in valence. 3 Following on arguments from trust research (e.g., Bijlsma-Frankema, Sitkin, & Wiebel, 2015) and neuroimaging (Dimoka, 2010) regarding the importance of considering positive and negative trust constructs as distinct, a second model was estimated in which the original Procedural Fairness items were entered as indicators of two separate latent factors: Procedural Unfairness and Procedural Fairness. This final CFA model fit better than the initial model, −2ΔLL(6) = 230.87, p < .001, and fit well to the data overall, χ2(231) = 512.86, p < .001; CFI = .97; TLI = .97; RMSEA = .05, p = .95; SRMR = .05. As before, all 24 items loaded significantly on their factors and all but one (again, the last Procedural Fairness item) loaded greater than .7 (see Table 1). Evaluation of the modification indices revealed less local misfit than in the initial model but it is worthy of note that there were relatively large model improvements attached to adding cross-loadings for one Obligation to Obey item on Normative Alignment, Trustworthiness, and Trust. Because of the good overall fit and the importance of construct clarity for testing the two paths to cooperation distinctly, these cross-loadings were not added.
As reported in Table 3, the final CFA model revealed significant correlations among all seven latent constructs. Because of their ability to partial out error variance, latent variable correlations are often stronger than observed variable analyses and are often above .80 (e.g., Gau, 2011; Reisig et al., 2007; Tankebe, 2013), but several of the relationships identified here do bear further investigation. Specifically, the analyses revealed correlations of about .90 among the Normative Alignment, Trustworthiness, and Trust latent constructs. As recommended by trust researchers (Hamm & Hoffman, 2016), additional models were therefore estimated in which the highly correlated latent constructs were combined into fewer factors. Because these changes resulted in additional constraints, these models would mathematically fit worse than the final CFA model, but it is important to note that, as reported in Table 4, all of these decreases in fit were statistically significant. 4 The good overall fit of the model to the data coupled with this evidence of poorer fit after combining the factors suggested that, although the factors were highly correlated, the relationships within the data were statistically better represented by maintaining their independence. Stated differently, this suggests that, although there was strong correspondence among participants’ responses to them, participants did distinguish across the constructs. Given this statistical evidence and the importance of conceptual clarity in our initial test of the two accounts, we kept the constructs distinct. Nonetheless, the covariance among these constructs will be revisited below.
Final CFA Model Latent Construct Correlations
Note. All correlations were significant at p < .001. CFA = confirmatory factor analysis.
Alternative CFA Model Tests
Note. CFA = confirmatory factor analysis; CFI = comparative fit index; TLI = Tucker–Lewis index; RMSEA = root mean square error of approximation; SRMR = standardized root mean square residual; TW = trustworthiness; NA = normative alignment.
p < .01. ***p < .001.
Distinct Models Test
We next moved to simultaneously test the two pathways to cooperation (the process and classic trust accounts) in a single model. As in the CFA models, all 24 items were entered as indicators of their hypothesized factors. Cooperation was initially modeled as three separate latent constructs but resulted in considerable global and local misfit. Because the items ask about the likelihood of engaging in specific behaviors (some of which may well be oddly related), we added them to the model as three observed, item-total scales (see Table 2) such that higher values indicated that participants felt that they were more likely to engage in more of the activities (for Specific Cooperation and Reaction to Crime) or more agreement (for General Cooperation). Although the three observed cooperation variables were skewed (values between −1.32 and −0.68) and kurtotic (values between 0.54 and 1.90), research suggests that our use of the maximum likelihood robust estimator was appropriate (Hoogland, 1999). 5
For the models reported here, all three item-total cooperation variables were regressed on the two pathways distinctly (see Figure 2). The model fit well to the data, χ2(303) = 894.13, p < .001; CFI = .95; TLI = .94; RMSEA = .06, p = .006; SRMR = .05, and confirmed most of the relationships hypothesized by the models. Regarding the process-based model, Procedural Fairness predicted Normative Alignment (β = 1.23, p <.001) and Obligation (β = 0.91, p < .001). Procedural Unfairness did not significantly predict Normative Alignment or Obligation in the full model, but it is important to note that this may have been a result of collinearity in the predictors given the high correlation between Procedural Fairness and Procedural Unfairness (r = −.78). To address this, we reestimated the model with only Procedural Fairness and then only Procedural Unfairness. In both models, both Procedural Fairness scales were roughly equivalently predictive of Normative Alignment and Obligation to Obey.
Distinct Paths Structural Equation Model
Regarding the classic model, Trustworthiness significantly predicted Trust (β = 0.91, p < .001) but Propensity to Trust did not. Most notably, the model suggested somewhat varied effects on cooperation such that while Trust (β = 0.37, p < .001) and Normative Alignment (β = 0.36, p < .001) significantly predicted General Cooperation, only Trust predicted Specific Cooperation (β = 0.35, p <.001) and only Normative Alignment predicted Reaction to Crime (β = 0.27, p < .001). Obligation was not predictive of any of the measures of cooperation. 6
Integrated Model Test
The distinct paths SEM model tested the utility of the process and classic trust accounts as distinct pathways to cooperation. As expected, these results suggested somewhat complex roles for the accounts but, as noted above, the modification indices suggested considerable overlap between Trustworthiness and Normative Alignment. 7 Because the statistical overlap identified in the final CFA model was theoretically consistent with the discussion of their conceptual overlap in the introduction, we next tested an integrated model (see Figure 3). The primary feature here is the combination of Normative Alignment and Trustworthiness into a single latent factor. As a result of this change, Procedural Fairness and Procedural Unfairness were now entered as predictors of the new Trustworthiness/Normative Alignment construct which subsequently predicted Trust. The three cooperation variables were then regressed solely on Trust. Although there has been some discussion in the field about the direct effects of procedural fairness on trust, it is important to remember that the conceptualization of trust in this previous work is distinct from that provided here such that in most process-based model scholarship, trust actually refers to what is here called trustworthiness. As a result, these effects were not hypothesized in the integrated model. In addition, because they were not predictive in the previous model nor in this one, Obligation to Obey and Propensity to Trust were removed from the model.
Integrated Model
The integrated model fit well to the data, χ2(182) = 431.92, p < .001; CFI = .97; TLI = .97; RMSEA = .05, p = .76; SRMR = .03, and supported all of the expected relationships (see Figure 3). Specifically, Procedural Fairness (β = 0.75, p < .001) and Procedural Unfairness (β = −0.16, p = .003) both predicted Trustworthiness/Normative Alignment which subsequently predicted Trust (β = 0.92, p < .001). Trust then significantly predicted all three cooperation measures but, as before, was most predictive of General Cooperation and least predictive of Reaction to Crime. As before, the model also provided little evidence of local misfit but this is of particular note because it suggests that direct effects that were not included in the model (Procedural Fairness on Trust or cooperation and Trustworthiness/Normative Alignment on cooperation) were not necessary for good fit. Indeed, adding these relationships revealed only negligible changes in the criterion variance accounted for by the regressions on Trustworthiness/Normative Alignment (R2 = .773, previous R2 = .772), Trust (R2 = .833, previous R2 = .840), General Cooperation (R2 = .529, previous R2 = .509), Specific Cooperation (R2 = .299, previous R2 = .288), and Reaction to Crime (R2 = .191, previous R2 = .171).
Alternative Models
To ensure that our final model best represented the relationships within the data, we next tested a series of alternative models. The high correlations among the factors and the cross-sectional nature of the data leave open the possibility that the relationships identified in this model were not falsifiable. Stated differently, the substantial covariance in the data may have meant that that no matter how the lines within the model were drawn, they would have been significant. It is important to note, however, that the relationships hypothesized in the model would still result in decreases in fit if they did not reflect the actual relationships in the data, even if they were significant. This means that even if there were several models within the data that could have yielded significant paths, only one model would fit the data best. We therefore tested a series of alternative theoretically and statistically driven models, including models in which relationships were directly reversed (such that, for example, Trustworthiness predicted Procedural (Un)Fairness which then predicted Trust and Cooperation). In addition, given the high correlation between Trust and both Normative Alignment and Trustworthiness, we also tested a model in which we combined all three constructs into a single latent factor. All of the models revealed similar relationships among constructs but, in every case, revealed poorer BIC values than the integrated model. Thus, we argue that the integrated model best captured the relationships in the data.
Discussion
The current research provides an initial test of the relative roles of the process-based model and the classic trust model. The final CFA model confirmed the hypothesized separability of the model constructs but was suggestive of noteworthy statistical overlap among Normative Alignment, Trustworthiness, and Trust. We therefore estimated several additional models in which the items were combined as indicators of fewer factors, but each new model configuration resulted in a statistically significant decrease in model fit. Thus, although the correlations among the constructs were strong enough to warrant evaluation, including the constructs in the model as distinct latent factors was not unjustified.
The subsequent distinct paths model results supported the general propositions of both models. First, consistent with an ever-increasing body of scholarship (see Tyler et al., 2015), process matters. To the extent that respondents believed the police interacted with the public in a fair and respectful manner, they were more likely to report higher normative alignment, obligation to obey, and cooperation. On the contrary, as indicated by a similar body of scholarship, this conceptualization of trust as a willingness to accept vulnerability also matters (see Schoorman et al., 2007). When respondents viewed the police as trustworthy, they also had more trust in law enforcement. Furthermore, as expected, this willingness was associated with increased cooperation. These results suggest that efforts to improve or protect the relationship of the police and the public that address either of these foci are likely to be effective. This is certainly good news for proponents of both models but it stops short of providing clear, actionable advice. Indeed, the best advice that these analyses can give to practitioners is that they must design their strategies around the kind of cooperation they want. If the kind of cooperation sought is general, both models appear to be predictive but increasing specific cooperation seems to implicate trust while reactions to crime implicate normative alignment. Interestingly, our measure of obligation to obey was never predictive but this is not inconsistent with previous scholarship (e.g., Gau, 2014; Tankebe, 2013; Tankebe et al., 2016).
The integrated model results sought to incorporate the basic arguments of the process and classic trust models. Following on the conceptual overlap discussed above and the statistical overlap identified in our analyses, we combined Trustworthiness and Normative Alignment into a single construct mediating the relationship between Procedural (Un)Fairness and Trust. These model adjustments were supported by the subsequent analyses, indicating that these constructs may be more similar than is often suggested in the literature. Considered together, the current research provides preliminary support for an account of legitimacy which suggests that evaluations of interactions with the police (here, procedural (un)fairness), drive evaluations of the institution and its agents (here, normative alignment and trustworthiness), that drive the internalization of positive psychological states within the individual (trust), and subsequently, facilitate positive reactions (cooperation).
Moving Toward an Integrated Framework of Police Legitimacy
Legitimacy has a long history in sociological and, subsequently, criminal justice scholarship. Arguably originating with the work of Weber (1968), modern notions of legitimacy have consistently been highlighted by scholars as a critical piece of the relationship between government and the governed (Beetham, 1991). Throughout this work, scholars widely agree that in the most mutually beneficial of these relationships, the authority of government is acknowledged and validated by the governed. They point to this as the essence of legitimacy but agreement in the construct largely ends there. Some conceptualizations of legitimacy are limited to the perception of the rightness of authority but most research actually measures proxies or components such as obligation to obey, normative alignment, trust, lawfulness, distributive and procedural fairness, and effectiveness (Gau, 2011, 2014; Reisig & Bain, 2016; Reisig et al., 2007; Tankebe, 2009; Tankebe et al., 2016; Tyler & Jackson, 2014), and others go so far as to argue that legitimacy is best understood as a dialogue (Bottoms & Tankebe, 2012).
The state of the literature on legitimacy therefore suggests a number of potential constructs and relationships and we find considerable merit to this argument. As Beetham (1991) argued, the “key to understanding the concept of legitimacy lies in the recognition that it is multi-dimensional in character” (p. 15). Indeed, evaluation of our work here suggests that legitimacy may be best understood as a framework or meta-construct in that it refers, not to a single idea but instead, to a collection of ideas that are theoretically linked in an account that, while subject to some variation, is widely consistent. It is for this reason that we argue for the development of an Integrated Framework of Police Legitimacy. We suggest that within this framework lie a variety of constructs which can be meaningfully grouped into at least four categories, namely, evaluations of the interaction, evaluations of the target, internalizations, and reactions (see Figure 4). Our results here lend credence to this general structure by showing that procedural fairness (an evaluation of the interaction) leads to trustworthiness and normative alignment (evaluations of the target) which provide a basis for trust (an internalization) which, in turn, facilitates cooperation (a reaction). To be sure, there are other constructs within these groupings that were not tested here and, in recognition of this, we include several example constructs in Figure 4. We suggest, however, that there is good reason to believe that the major constructs are included in our data and are therefore sufficient to support the implications noted below.
Integrated Framework of Police Legitimacy
Evaluations of the interaction refer most proximately to an individual’s evaluation of a specific interaction with a specific law enforcement officer but more distally refer to evaluations of the effects of law enforcement generally. As such, this group of constructs may include both more proximate (e.g., specific procedural justice; Gau, 2014) and more distal evaluations (e.g., the general effectiveness of the police; Tankebe et al., 2016). A major theme throughout the process-based model scholarship, however, is the centrality of procedural fairness which is supportive of its singular focus here. Indeed, work frequently suggests that procedural fairness outperforms instrumental concerns like distributive fairness or the risk of punishment in predicting reactions to the police (Tyler, 2006b) and, although recent work does suggest a relatively strong role for effectiveness (Tankebe et al., 2016), procedural fairness remains (slightly) more important in these models.
In addition, although it is discussed as roughly unidimensional here, procedural fairness is itself multifaceted. Our conceptual review of the construct revealed three substantive components (respecting boundaries, interpersonal treatment, and quality of decision making) and our statistical analyses were suggestive of two valence-related dimensions (fairness and unfairness). 8 In other work, procedural fairness has been further divided into evaluations of a specific previous interaction (e.g., “The officer treated me with respect”; Gau, 2014, p. 196), a generalized belief about the actions of most police (e.g., “The police make decisions based on the facts”; Reisig et al., 2007), or even expectations of future interactions (e.g., “‘How much (do) you agree or disagree [that]: They would treat you with respect if you had contact with them for any reason?” Bradford, 2011, p. 9). Our results here lend additional credence to the argument for the centrality of general procedural fairness as the starting point of police legitimacy as our measures accounted for almost 80% of the variance in the evaluations of the police. Thus, we stand with the majority of the work on policing to argue that while a number of evaluations of the interaction with the police will matter, procedural fairness (and especially general procedural fairness; see Gau, 2014) appears to be the key conceptual starting point.
Several evaluations of the police also exist in the literature, but a strong argument could be made that the current work subsumes the most important of these as well. Specifically, the process-based model scholarship focuses heavily on normative alignment and trust (here, trustworthiness) as the major evaluative constructs. Our work here lends credence to this argument but takes this a step farther by showing that, as in other contexts (PytlikZillig et al., 2016), these constructs are often strongly related. Across analyses, our models consistently revealed significant statistical overlap between the constructs which was substantiated by the conceptual overlap noted in the introduction. Thus, although discussed somewhat differently, the construct that process-based model scholars refer to as normative alignment is similar to what classic trust model scholars discuss as trustworthiness. It is important to note, however, that we do not argue that normative alignment and trustworthiness are exactly the same thing. At a conceptual level, normative alignment is a normative basis of power and authority that reflects understandings of both the proper position of the institution within society and their interrelationship with the tangible representations of that institution (Jackson et al., 2013). Trustworthiness, on the contrary, reflects an evaluation of the character of the institution, and therefore its representatives, by tapping into the motives and competencies that people believe it possesses (Mayer et al., 1995).
Despite this conceptual distinctiveness, however, it is also clear that the constructs are inextricably linked. Judgments about character will shape judgments about values alignment and vice versa and, as a result, the public may not actively distinguish between them. This postulation is supported by our results which were suggestive of a combined trustworthiness and normative alignment latent construct. In light of this, we suggest that distinguishing among evaluations of the police like normative alignment and trustworthiness may not be especially profitable. Again, this does not imply that individuals cannot tell the difference between them, but instead that people tend to hold relatively unified evaluations of criminal justice institutions. We suggest, therefore, that the utility of treating them as distinct in cross-sectional surveys of the public may be limited. As suggested by work on perceptions generally, humans are often strongly motivated to hold relatively unified evaluations of a single target (Quinn et al., 2003). Individuals will typically struggle, for example, in simultaneously believing the police to be high in trustworthiness and low in normative alignment and, as a result, research will consistently reveal high correlations among these and other evaluative constructs (e.g., benevolence, integrity, identification, reliability), likely to the point that measuring them separately may not yield any recognizable benefit. It is important to note, however, that when measurement is coupled with longitudinal or experimental data collection, attending to the various constructs distinctly may become much more important. Indeed, it is not hard to see how fostering normative alignment could be superior to a focus on, for example, reliability in building trust in the police. 9
The third construct grouping within the framework of legitimacy is made up of the resulting internalizations within the individual which, in the current study, included both trust and an obligation to obey. Drawing from the classic trust model (Mayer et al., 1995), we join with the increasing body of scholars who define trust as a psychological state within the individual that is characterized by a willingness to accept vulnerability to the agency of the target. As suggested by our review, policing scholarship has consistently struggled in disentangling trust from other related constructs (Jackson & Gau, 2016; Tyler & Jackson, 2013). By rooting our conceptualization of trust in the classic model, however, we provide a basis for understanding the nature of trust that is conceptually and operationally distinct from other related constructs (e.g., normative alignment, trustworthiness, procedural fairness). Rather than being part of the individuals’ evaluations, trust here is a result of it.
It is important to note that this place of trust is not unlike many conceptual discussions of obligation to obey. Although it is typically not modeled as such, obligation to obey is consistently discussed as an internalization such that an individual’s belief that an institution is trustworthy or normatively aligned with the public creates a state within them that is characterized by an obligation to obey the directives of the institution (Tyler, 2006b). Thus, the conceptualization of trust presented here differs importantly in the nature of the specific psychological state such that, for obligation, the state is characterized by a motivation to obey and, for trust, by a willingness to accept vulnerability. Our results here lend support to this distinction by revealing relatively little statistical overlap between the constructs and suggest that while trust was consistently predictive of cooperation, obligation to obey never was. As a result, we suggest that although obligation to obey may be an important internalization in some contexts (e.g., compliance with the law or officer directives), a willingness to accept vulnerability is likely to have a robust effect.
One important implication of this argument regarding the nature of trust within the framework of legitimacy is the central role it suggests for vulnerability. Vulnerability is an essential part of every human interaction (e.g., Lind, 2001) and it is especially salient in the criminal justice context. Law enforcement in particular is given a tremendous amount of power by society to regulate citizens (Walker, 1993, 2006). It can—appropriately or inappropriately—circumscribe behavior, suspend rights via arrest, extract property through fines and asset seizure, and use deadly force to resolve conflicts. While these direct harms are important, the potential for harm from the police is much broader and encompasses more amorphous and intangible harms as well. As representatives of the law, the police play a central role in organizing society (Sklansky, 2005; Tapp & Levine, 1974; Tyler & Trinkner, in press) and, as a result, the potential for harm to the public also resides in violations of what the police should be. Even when tangible harm is unlikely, as in the case of the majority group member who is concerned about disparate treatment of minorities, there remains a potential for harm that is certainly more abstract, but which may be no less salient. As suggested in the introduction, the necessary and considerable discretion of the police makes eliminating these vulnerabilities unlikely. Thus, our fourth category of constructs, reactions to the police, relies heavily on this willingness to accept vulnerability.
Reactions to the police that have been discussed in the literature include cooperation (as addressed here) as well as constructs like empowerment and compliance. The connection of vulnerability to cooperation and empowerment is probably most obvious as in both cases, the individual is voluntarily advancing the mission of the police by actively working to assist the police (cooperation) or consciously deciding to remove, or at least not add, barriers to their effectiveness (empowerment). Thus, an individual who cooperates with or empowers the police must, at some level, be willing to accept the potential for harm to themselves. The connection between compliance and vulnerability may seem more tenuous but stands to reason when considered in light of the importance of relational concerns in driving these reactions. Research consistently suggests that building strong relationships with the public is a major factor in driving compliance (Tapp & Levine, 1974; Tyler, 1997) and, without some level of acceptance of vulnerability, facilitating the give and take required for developing a strong positive relationship will be, at least, more difficult.
Challenges within the Framework
Although we believe that our data provide provocative evidence regarding the proposed framework of legitimacy, there are at least two challenges that are ripe for future research, namely, (a) the potential bidirectionality of the relationship between evaluations of the interaction and evaluations of the target and (b) the difficulty of statistically distinguishing between evaluations of the target and resultant internalizations. Regarding the first challenge, process-based model scholarship generally argues that procedural fairness drives legitimacy but there also exists an important minority argument regarding the potential for a reverse effect such that the level of procedural justice individuals believe the police provide may be a function their perceived legitimacy (Bottoms & Tankebe, 2012; see also Mondak, 1993). Although we stand with the majority view in specifying our model, we recognize that both processes are likely to occur. Conceptually, the root of perceptions of the police must start with some level of information from direct or vicarious interactions with the police. Thus, the initial process likely operates as modeled here such that one moves through Figure 4 from left to right. This, however, does not preclude the possibility of feedback loops within the framework. For example, it is likely that in some situations, the more salient relationship between evaluations of the police and evaluations of interactions with the police actually runs backward. This may be especially true when individuals either have or feel that they have insufficient information upon which to base their assessment of how the police treat “most people,” but the question of whether and when this relationship flips is an empirical one. Our results suggest that the effect is defensible as presented and, given the decrease in fit in the alternative model that flipped the direction, even provide some evidence that it is the superior explanation. Nonetheless, it is important to note that our analyses do not preclude the possibility of a reverse effect. As a result, we encourage future researchers to shed additional light on whether, and the conditions under which, the direction of the effect reliably changes.
The second challenge is equally ripe for investigation. The conceptual distinction between trust and trustworthiness lies at the core of the classic model of trust and, as a result, has been a particularly influential idea in the field of trust research for quite some time. This distinction marked the end of the first era of trust research in which trust was thought of as a characteristic of the trustor, such that some people were simply more trusting than others (e.g., Rotter, 1967). Distinguishing trust from trustworthiness suggested that there may, in fact, be things that a trust target can do that increase trust independent of the personality of the trustor and it is upon this assumption that the literatures addressing process-based police legitimacy and the classic model of trust rest. Problematically, however, it appears somewhat difficult to reliably separate these constructs empirically (see PytlikZillig et al., 2016). This situation mirrors that of distinguishing constructs within the evaluations of the institution that was discussed above but happily appears somewhat easier for participants. In our research, the model misfit attributable to combining trust with either trustworthiness or normative alignment was at least twice that of combining trustworthiness and normative alignment (see Table 4). This is suggestive of the relatively greater statistical distinctiveness of trust, but it certainly remains something less than would be ideal. Future research may be able to address this by applying suggestions regarding measurement artifacts like common-method variance (e.g., Podsakoff, MacKenzie, & Podsakoff, 2012), but a more profitable line of inquiry would be an investigation into the conditions under which these constructs become more separable. It may be, for example, that the distinctiveness is greater for individuals with a higher need for cognition, more relevant experience, or when measured more precisely. The state of the literature makes it clear that evaluations are not the same thing as internalizations but, given the difficulties in identifying these differences statistically, it may yet be a (conceptual) distinction without a (practical) difference.
Limitations and Conclusion
Recently, there have been calls for new approaches to policing that identify trust and legitimacy as essential pillars of effective law enforcement (Ramsey & Robinson, 2015; Schulhofer et al., 2011). Unlike traditional approaches that emphasize instrumental concerns, this approach recognizes that relational considerations are major predictors of following the law (Tyler, 2006b). In short, people obey the law when they view law enforcement as legitimate. Thus, it is clear why fostering positive perceptions is essential for an effective criminal justice system: It ultimately fosters internal motivations that facilitate upholding the social contracts that permit society (Sunshine & Tyler, 2003a). In response, agencies and policy makers increasingly focus on strategies and approaches that build these perceptions. The current work speaks directly to these efforts by taking an important step in clarifying and disentangling the roles of some of the plethora of related constructs in this literature. Specifically, we evaluated models that distinguish public evaluations, internalizations, and reactions in the policing context, and tested the relationships among them.
Despite this contribution, however, our work is not without limitations. Most notably, our use of MTurk, albeit appropriate in light of research addressing these samples, carries with it two potentially important concerns, namely, threats to generalizability and the potential for uniform responding. Regarding generalizability, it is, as yet, unclear who precisely MTurk samples represent. As discussed above, research does generally speak positively to the ability of these samples to approximate the United States generally (Berinsky et al., 2012; Buhrmester et al., 2011), but these samples are necessarily limited to individuals who not only have access to but, likely, are comfortable with computers, thereby potentially over-sampling more affluent and educated individuals. Contrastingly, though, some anecdotal, typically journalistic evidence suggests that MTurk workers may overrepresent individuals for whom regular employment is difficult, thereby overweighting individuals of lower socioeconomic status and education.
Regarding uniform responding, online samples do increase the risk that participants would be distracted or unmotivated during the completion of their task and, as a result, neglect the conceptual distinctions upon which articles like this are premised. We believe that we have taken sufficient protective measures to ensure that this concern does not undermine our conclusions through the use of Master Workers (who can lose their status and therefore the additional money it affords by submitting poor data) and through our rigorous testing of alternative models throughout our analyses which demonstrate that our theoretically driven models do, in fact, best approximate the data (a finding which is theoretically impossible in random data and unlikely in data that are artificially correlated across constructs). Thus, although appropriate safeguards were in place and similar data often reveal relationships as strong as ours (e.g., Gau, 2011; Reisig et al., 2007; Tankebe, 2013), it is nonetheless possible that our heightened correlations are the result of a higher degree of uniform responding than would be optimal. As a result, we suggest that, as with any single study, our results not be taken as the last word on the relationships among these constructs. Instead, we simply offer an initial step toward a framework of legitimacy with implications for both theory and practice but, as ever, future research should take up these questions with other samples and methodologies.