Why Officer Attitudes Matter

Our work builds on a substantial body of research that examines the causes and effects of public servants’ attitudes. Work in public administration focuses on two categories of attitudes that correlate with the quality of services bureaucrats provide: bureaucrats’ assessment of clients and their adherence to bureaucratic mission (Keiser Reference Keiser2010; May and Winter Reference May and Winter2009; Maynard-Moody and Musheno Reference Maynard-Moody and Musheno2003). In political science, the most common approach is to treat these attitudes as mechanisms and proxy for their existence using data on bureaucrats’ demographic characteristics, though some notable exceptions measure bureaucrats’ attitudes directly (Kuipers Reference Kuipers2023; Sabatier, Loomis, and McCarthy Reference Sabatier, Loomis and McCarthy1995; Toral Reference Toral2024). Applied to the study of policing, research tends to focus on three aspects of client assessment that we label (1) sharing citizen concerns, (2) trust in citizens, and (3) empathy for citizens’ concerns. When it comes to adherence to bureaucratic mission, policing scholars are especially attuned to officers’ views on (4) organizational accountability for misconduct and (5) the seriousness of misconduct, especially corruption. Variation in these attitudes is important because patrol officers have substantial discretion over when to enforce laws and what sanctions to apply, similar to the discretion afforded other street-level bureaucrats (Lipsky Reference Lipsky1980). In particular, officers’ attitudes correspond to policing practices such as the quality of communication, the amount of effort devoted to investigating citizen tips, the prioritization of some citizens over others, and the propensity to harass or harm citizens.

Understanding and sharing citizen concerns is often considered the first step toward officers being able to do their job effectively. When officers are in touch with the issues that are important to citizens, they can better allocate their time to activities that make a substantial dent in improving order and earning citizens’ trust (Skogan and Hartnett Reference Skogan and Hartnett1997). Moreover, a better understanding of citizens’ concerns is considered to be the key mechanism that leads to improvements in other officer attitudes (Paoline, Myers, and Worden Reference Paoline, Myers and Worden2000). When officers better understand citizens’ perspective on public safety issues, they are less likely to resort to harmful stereotypes of civilians as being unpredictable and suspicious.

Next, research devotes substantial attention to officers’ trust in citizens, a concept that is particularly prevalent in work on “police culture.” We conceptualize trust as a range of officer beliefs that constitute an “us versus them” mentality, including perceptions of citizen intentions and the threat posed by citizens (Sparrow, Moore, and Kennedy Reference Sparrow, Moore and Kennedy1990). For example, Paoline (Reference Paoline2003, 203) defines a key aspect of police culture as the “prescription to be suspicious and maintain the edge over citizens.” Trust in citizens varies greatly across officers, and this variation has strong links to officer behavior (Paoline Reference Paoline2003). For example, studies find that suspicion toward citizens is correlated with officer abuse and discrimination (Ingram, Terril, and Pauline Reference Ingram, Terril and Pauline2018), as well as lower productivity (Moon and Zager Reference Moon and Zager2007).

A third set of client-facing attitudes relates to officers’ empathy for citizens’ concerns, which we define as the belief that citizens’ concerns and complaints are warranted. While this captures only a narrow subset of the overarching concept of empathy, it comports with the mechanisms theorized to link officer empathy with improved policing outcomes. Inzunza (Reference Inzunza2015) outlines two fundamental components of empathy that are salient in the policing literature. The first is whether officers place themselves in citizens’ shoes when hearing their reports and complaints, which includes both affective (feeling what others feel) and cognitive (perspective-taking) dimensions of empathy (Todd and Galinsky Reference Todd and Galinsky2014; Zaki Reference Zaki2014). This leads to a second, behavioral manifestation of empathy, whereby officers outwardly convey that they are receptive to citizen-provided information (Barrett-Lennard Reference Barrett-Lennard1981). When officers convey that reports and complaints are justified and worth taking seriously, citizens—especially those from marginalized groups and those who experience sexual violence—are more likely to report positive contact with officers and follow through on initial reports (Birzer Reference Birzer2008; Maddox, Lee, and Barker Reference Maddox, Lee and Barker2011; Turgoose et al. Reference Turgoose, Glover, Barker and Maddox2017).

Finally, scholarship suggests that officer behavior is shaped by two aspects of their adherence to bureaucratic mission. The first regards officers’ perception of organizational accountability. When officers believe that police leadership can monitor citizen-facing misconduct, takes citizen complaints seriously, and would seriously reprimand misbehavior, officers may begin treating citizens as partners rather than adversaries (Cordner Reference Cordner1995; Crank Reference Crank2014). Accountability links to a second important concept, which is whether officers internalize that harming citizens constitutes serious misconduct. This includes officers’ views on the seriousness of activities like corruption. The perception of organizational accountability can make adversarial comments toward citizens a taboo topic, leading to a change in norms about the appropriateness of misconduct (Haarr Reference Haarr2001). The literature suggests that improvements in officer behavior are more sustainable when officers begin treating issues surrounding misconduct in accordance with a “logic of appropriateness,” independently of whether they are worried about being punished for bad behavior (March, Reference March1994; Sparrow, Moore, and Kennedy Reference Sparrow, Moore and Kennedy1990).

How Community Policing Influences Officer Attitudes

In the absence of regular, informal contact with citizens, a reliance on reactive policing gives officers a skewed view of the community by disproportionately bringing them into contact with citizens when a crime has occurred. Officer attitudes tend to depend heavily on rough indicators of “neighborhood context,” such as the demographic makeup, homocide rate, or reputation for citizen resistance to the police (Terrill and Reisig Reference Terrill and Reisig2003). Prior research suggests that officers commonly underestimate the proportion of the community that supports their efforts to improve public safety (Skolnick Reference Skolnick2011). These misperceptions can lead to alienation from the community and a feeling that citizens do not share the officer’s values (Twersky-Glasner Reference Twersky-Glasner2005). In short, it is easy for officers to resort to harmful stereotypes of the typical citizen’s experiences, intentions, and behaviors.

A core tenet of community policing is that positive, informal interactions with citizens can overcome these misperceptions (Paoline, Myers, and Worden Reference Paoline, Myers and Worden2000). When it comes to client assessments, community policing should first and foremost provide officers with information that leads them to share citizens’ concerns. Officers’ priorities may shift as they understand that issues appearing relatively minor to them are actually significant problems for citizens. This should translate to improvements in officer trust by giving them a more accurate view of the prevalence of threats to officer safety and the proportion of citizens who support police efforts. These interactions should also allow officers to understand citizen motivations when they choose to report issues to the police, generating greater empathy and leading officers to take citizen reports more seriously (Schuck and Rosenbaum Reference Schuck and Rosenbaum2005).Footnote ³

A second key goal of community policing is to signal increased organizational accountability and change officers’ beliefs about the seriousness of misconduct like corruption. By implementing community policing, police leadership signals that they care about the quality of citizen engagement (Cordner Reference Cordner1995). Community policing also typically involves strengthening procedures by which officers can be held accountable for misconduct, including efforts to emphasize the importance of reporting fellow officers. Moreover, when officers are placed in more frequent contact with law-abiding citizens, the hope is that, over time, they will internalize the seriousness of misconduct and the importance of serving people rather than on “doing policing” (Bracey Reference Bracey1992).

Drawing from the literatures discussed above, we propose five hypotheses for how community policing should shape officer attitudes toward citizens:Footnote ⁴

A small set of existing studies aims to empirically test a similar set of hypotheses regarding the effects of community policing on police attitudes. A first group of observational studies provides useful evidence that is nevertheless susceptible to various threats to causal inference. Paoline, Myers, and Worden (Reference Paoline, Myers and Worden2000) compare attitudes between specialized community policing officers and regular beat officers in two US police departments. Hayeslip and Cordner (Reference Hayeslip and Cordner1987) and Rosenbaum, Yeh, and Wilkinson (Reference Rosenbaum, Yeh and Wilkinson1994) both use longitudinal research designs, tracking officer attitudes before and after they implemented community policing. All three studies find that community policing corresponded with improved officer attitudes toward citizens. However, these results may be biased if the same factors that drove officer and departmental decisions to adopt community policing are correlated with officer attitudes, or if other changes that occurred concurrently with community policing programs shaped officer attitudes.

Another relevant study is Blair et al. (Reference Blair, Christia, Weinstein, Arias, Badran, Blair and Cheema2021), who implemented a set of community policing experiments in the Global South and report null results of community policing on officer attitudes, including empathy, corruption, and accountability.Footnote ⁵ We see our work as building on (rather than contradicting) the results of Blair et al. (Reference Blair, Christia, Weinstein, Arias, Badran, Blair and Cheema2021), which foregrounds the effects on crime and citizen-facing outcomes. At a basic level, we contribute to the advancement of research on this topic by elucidating the theory behind the predicted effects on officers and incorporating two additional outcomes (Shared Concerns and Trust) that are crucial to our interpretation of the results. More importantly, our inductive theory and the evidence we present in the latter half of the paper can help scholars and policymakers interpret why and under what conditions social contact with citizen may fail to translate into improved attitudes, both among police officers and bureaucrats more generally.

The “One Sorsogon” Community Policing Intervention

Sorsogon’s community policing intervention began as a reform initiative led by the Provincial Police Chief. He wanted to implement a community-centered approach to contrast with the violent “War on Drugs” happening elsewhere in the country. Even though virtually no drug-war related violence occurred in Sorsogon, the Police Chief expressed concern that the PNP’s local reputation was harmed by the national-level environment. Prior to the intervention, he implemented a number of small-scale programs aimed at bolstering the service-oriented side of policing, such as a program to engage with at-risk youth in schools. However, practically all community engagement activities prior to the intervention were under the domain of a designated “Police Community Relations” (PCR) Director at each municipal station, along with a handful of PCR officers at the Provincial Police Office, none of whom were eligible for our intervention. Regular beat officers in Sorsogon almost never engaged in the usual components of community policing, such as town halls or meetings with community leaders.

The new initiative, called “One Sorsogon,” spanned seven months and centered on directing rank-and-file officers to work closely with barangay leaders and ordinary citizens to identify the issues that most threatened public safety in their communities, and to devise locally appropriate solutions together. We randomized the selection of PNP officers who participated in the community policing activities, which is the design feature we use to draw inferences in the present study.Footnote ⁶ Control officers did not participate in any of the activities described below, and experienced policing-as-usual for the duration of the study.

The program was a bundled treatment that kicked off with a two-day training on best practices for community policing led by a professional consultant.Footnote ⁷ At the training, treatment officers were paired with the barangay leaders in their assigned barangay (who were also in attendance), practiced using crime reports and tanod logbooks to identify the most pressing local public safety issues, learned techniques for empathetic communication, and focused on recognizing and reporting officer misconduct, including corruption.

Treatment officers then attended six monthly meetings with leaders of their assigned barangay, lasting just under two hours on average, and held at the municipal police station. The first three meetings focused on identifying the most pressing local public safety problems, devising a plan to address those issues, and proposing a budget of up to 5,000 pesos (about USD $100) to help with implementation. The next three meetings focused on implementing this plan and monitoring its progress. This aspect of community policing is known as “Problem-Oriented Policing” (POP) (Goldstein Reference Goldstein1990; Weisburd et al. Reference Weisburd, Telep, Hinkle and Eck2010). In our case, POP teams were composed of two PNP officers, the barangay captain, the barangay councilor in charge of peace and order, the chief tanod, and 2–4 regular tanods. The POP aspect of the intervention was designed to enhance interactions with barangay leaders, first and foremost. We conceive of these leaders as being comparable to an important class of “community leaders” focused on by prior research on community policing, who typically have a more complete understanding of crime in their community and can better assist the police in implementing reforms. From the perspective of the PNP, barangay leaders were discussed as “citizen representatives,” and they play a similar role to informal village leaders and community watch members in other parts of the Global South.

Between meetings, the officers assigned to One Sorsogon were directed to engage with a broader range of citizens in the communities where they were assigned. Officers regularly traveled to their assigned barangay to engage in citizen-facing foot patrols, gather citizen perspectives, and disseminate information about police activities. The implementation strategies of the POP teams also tended to directly involve officers, for example by building community gardens and passing out sports equipment with at-risk youth. Treatment officers met monthly with their station’s police chief and PCR director to discuss the progress of the intervention, and were encouraged to surface reports of police misconduct during these meetings. Finally, the intervention concluded with a community town-hall in each barangay, during which PNP officers received input from the broader community and communicated their plans going forward. On average, each culminating town hall was attended by eighty citizens and lasted more than three hours.

Our study is best interpreted as an evaluation of community policing as it is actually implemented in many real world contexts, especially those outside of the US, Australia, and Europe, where most studies of community policing take place. Similar to many countries with moderate or weak police capacity, the Sorsogon PNP faces pressures associated with difficult terrain, weak infrastructure, under-staffing, decentralization of resources, and leadership rotations that disrupt policy continuity. Under these circumstances, community policing programs (including those supported by international donors like the UN Department of Peacekeeping) tend to be localized, limited to several months of intensive implementation, and involve officers who are also assigned to other duties. We note that One Sorsogon was a homegrown initiative, and we did not alter the duration of the program originally specified by the Provincial Chief. In this context, the intervention was relatively intensive, and many PNP officers expressed that One Sorsogon represented a wholesale change in the scope of their community-facing duties. That said, we acknowledge that community policing may produce different effects if paired with an overhaul in resources, staffing procedures, or long-term commitments from national leadership.

Ethics

By the time we approached the Sorsogon PNP about partnering on an evaluation of One Sorsogon, we had spent approximately one year conducting observational research on policing in the province. We previously conducted a large survey on policing and developed an extensive network of local academics, citizens, community leaders, police officers, and individuals with connections to the NPA, all of which helped us to better understand the risks of the study and the ways we could mitigate those risks. To briefly summarize three key points, we learned from our contacts in the area that the Sorsogon Police Chief was a well-known reformer who opposed the violent tactics of President Duterte’s Drug War, which primarily took place in major urban areas. We confirmed the low rate of Drug War related violence in Sorsogon using surveys, conversations with local human rights activists, and data from ACLED’s special project on the Philippine Drug War. We also relied heavily on our local network in each municipality to identify and exclude areas where NPA presence was strong enough to make the intervention unsafe. Finally, our research staff attended every POP meeting and town hall, and conducted extensive audits in each barangay to detect potential abuse. We expand on these points, as well as other aspects of research ethics and human subjects protections, in the Supplementary Material.

Experimental Design

The Provincial PNP Chief agreed to randomize several aspects of the One Sorsogon intervention, including the barangays where it was implemented: of the 298 study barangays, 99 were randomly selected to receive the PNP’s community policing program. In the present study, we draw empirical leverage from the randomized assignment of officers to these barangays.Footnote ⁸ From the list of 705 eligible PNP officers serving in Sorsogon, we randomly selected 198 officers to participate in the community policing intervention, two for each participating barangay. Blocking took place at two levels. First, we blocked on police station (from among the 17 stations in the province) to assign treatment officers to a POP team in a barangay within their station’s jurisdiction. We also blocked on rank, assigning one senior and one junior officer to each team.

A number of officers initially assigned to treatment later had to withdraw from their POP team, mostly due to being reassigned to national-level training programs or a different station. When an officer was reassigned, they were replaced by the next officer on our list of randomly selected officers (in the same station and of the same rank). In total, 47 randomly selected replacement officers were assigned to POP teams during the course of the intervention. Most replacements occurred during the first week of the intervention due to a region-wide training and replacement cycle. As a result, the average replacement officer was assigned to attend 5.5 of the 7 community policing meetings. Our main results focus on the Intent to Treat effect (ITT), meaning that 245 officers (198 original plus 47 replacements) are considered treated. Treatment officers displayed a high degree of balance with control officers on their demographic characteristics, as well as their baseline responses to the survey outcomes (Table A.3, in the Supplementary Material).

Outcomes

We surveyed officers before and after the intervention and constructed indices of sharing citizen concerns (H1), trust in citizens (H2), empathy for citizen concerns (H3), perceptions of organizational accountability for citizen misconduct (H4), and perceptions of corruption as serious misconduct (H5).Footnote ⁹

Sharing citizens’ concerns (H1) was measured by asking officers what they thought were the three most important public safety concerns (from a list of 13 items) in the municipality where they were assigned, and then matching these responses with the four most common responses to the same question among community leaders who lived in that municipality.Footnote ¹⁰ Table 1 shows that, at baseline, citizens were far more likely than the police to identify minor crimes like public intoxication, theft, and illegal gambling. The police, on the other hand, were more likely to identify crimes like rape, illegal drug use, and murder. We expect that the Shared Concerns measure is correlated with broader factual knowledge about the community, though limited survey space precluded asking additional questions measuring factual knowledge. For example, officers expressed to our research team that participating in the intervention gave them a better sense of which barangay leaders were the most effective local partners for motivating community action.

Table 1. Most Important Public Safety Issues According to Officers and Citizens

Note: Citizens and officers were each presented with the above list of public safety issues and asked to choose which three were the most important concerns in their municipality. The listed percentages show how often each issue was included in a respondent’s top three.

To capture the attitudinal outcomes in H2–H5, we used indices of several questions on the officer survey. Answers were captured using a slide bar that translated to a 100-point scale. The Trust index is composed of three questions: (1) citizens care about officer well-being, (2) citizen-provided information is likely to be accurate, and (3) how safe for officers is the municipality where you are assigned. We included the question on safety in the trust index based on literature that links officers’ perceptions of safety to their suspicions of the citizenry. Acknowledging that other factors may shape officer perceptions of safety, we separately report results for the individual components of the Trust index. The Empathy index is composed of two questions: (1) when people complain about the police, it is because they have a good reason, and (2) most things people report to the police are worth taking seriously. With the recognition that empathy is a broader concept than what we measured here, these questions are intended to capture one theoretically important aspect of how empathy manifests in policing, described in our theory section.

The Accountability and Corruption indices are both composed of questions about two hypothetical scenarios involving police misconduct. In one scenario, an officer accepts a bribe in exchange for not issuing a citation; in the other, an officer routinely accepts unsolicited gifts while on patrol. After respondents are presented with these scenarios, they are asked whether (1) they would report this behavior themselves, (2) they think their peers would report this behavior, and (3) the extent to which police leadership would punish this behavior. Along with a separate question (unrelated to the hypothetical scenarios) on (4) whether police leadership takes citizen complaints about officers seriously, these questions compose the Accountability index. Officers were also asked whether (1) they consider the behavior in the hypothetical scenario to be serious misconduct and (2) most other police officers consider it serious misconduct. These questions compose the Corruption index. Our focus on corrupt misconduct aligns with concepts raised during the community policing training and monthly check-ins with station leadership. It also accords with the PNP’s nationwide focus on tackling corruption in the lead up to the intervention period.Footnote ¹¹

Finally, to capture officers’ overall attitudes toward citizens, we construct a composite Officer Attitude Index, which is an “index of indices” incorporating all the attitudinal outcomes described above. To create each index, we standardized and summed all items, then divided by the number of constituent terms.Footnote ¹² One major concern was that officers’ answers about citizens could be biased upwards through social desirability bias. To mitigate this concern, surveys were self-enumerated on an iPad in a private location, which has been shown to reduce sensitivity bias (Nanes and Haim Reference Nanes and Haim2021).

Estimation

We analyze the average treatment effects using the following linear regression model:

⁽¹⁾ $$ \begin{array}{l}{Y}_{i,endline}=\alpha +{\beta}_1TREA{T}_i+{\beta}_2{Y}_{i,baseline}+{\beta}_3Block{s}_i\\ {}\hskip4.5em +{\beta}_4MissBaselin{e}_i+{\epsilon}_i.& \end{array} $$

We regressed each of our attitudinal outcome measures at endline ( $ {Y}_{i,endline} $ ) on an indicator for whether the officer was assigned to One Sorsogon ( $ TREA{T}_i $ ), controlled for the same attitudinal measure at baseline ( $ {Y}_{i,baseline} $ ), and included fixed effects for the blocking variables ( $ Block{s}_i $ ). If there are missing data in the baseline outcome used as a control, we add an indicator variable for missingness ( $ MissBaselin{e}_i $ ), and replace missing values with zeros (Gerber and Green Reference Gerber and Green2012). Standard errors are clustered at the station level using White clustered-robust standard errors. As a robustness check, we run the regression using only the endline cross-section, which does not rely on the imputation of baseline values:

(2) $$ \begin{array}{rl}{Y}_{i,endline}=\alpha +{\beta}_1TREA{T}_i+{\beta}_2Block{s}_i+{\epsilon}_i.& \end{array} $$

The structure of Equation 1 is preregistered; Equation 2 is not. To account for different treatment probability across blocks, which may introduce bias if treatment effects are not assumed to be the same across subjects (Gerber and Green Reference Gerber and Green2012), we include robustness checks using weighted regression in the Supplementary Material. The number of clusters in our study (17) is relatively small, though simulations suggest that White cluster-robust standard errors tend to perform well in experimental studies with a similar number of clusters.Footnote ¹³

Embeddedness

A first main takeaway from our field staff’s observations was that officers from within Sorsogon responded differently to community policing than officers from outside the province. Early on, officers local to Sorsogon appeared more comfortable during the POP meetings and were quicker to understand citizens’ perspectives when issues were raised. Yet some of these local officers expressed to us that they viewed community policing in similar terms to what Crank (Reference Crank2014, 311) characterizes as “departmental bullshit,” stating that they already had enough prior experience in the community to do their job well. On the other hand, our field staff indicated that as the program progressed, non-local officers seemed increasingly at ease, displayed greater effort, and improved their communication style during the POP meetings, despite starting at a lower baseline on these metrics.Footnote ¹⁶ Our conversations with non-local officers indicated that community policing helped dispel some of their suspicions of citizens that stemmed from Sorsogon’s reputation for being a hotbed of citizen resistance (manifesting in insurgency and drug trafficking).

To further explore these notions, we looked to the survey data, which confirmed that officers from outside Sorsogon had worse attitudes toward local citizens at the start of the intervention. Among non-local officers, there was 8% less overlap in Shared Concerns with issues deemed important by citizens. Average responses to the survey questions on Trust and Empathy were also eight points lower among officers from outside Sorsogon at baseline, though average answers of 75/100 among officers from Sorsogon left room for upwards movement for those officers as well. We then explored whether the patterns noted by our field staff were reflected in heterogeneous effects. Results for models interacting the treatment variable with an indicator of whether officers lived in Sorsogon at the start of the intervention are shown in Figure 2.

Figure 2. Heterogeneous Effects of Community Policing, by Officer Embeddedness

Note: All estimates are from the specification controlling for baseline outcomes (Equation 1). Marginal effects displayed in blue are for officers from outside Sorsogon; estimates displayed in tan are for officers local to Sorsogon. Confidence intervals are displayed at the 95% level (thin bars) and 90% level (thick bars). Full regression results are in Table A.6 in the Supplementary Material.

Corroborating our field team’s observations, the intervention had large positive effects on the overall Officer Attitude Index among officers from outside Sorsogon, but no effect among officers who lived inside the province. By the end of the program, treatment officers from outside Sorsogon had virtually caught up to officers from inside the province in terms of their attitudes toward citizens. Consistent with our inductive theory, outlined below, the heterogeneous effects are driven primarily by officers’ Trust and Empathy, the two outcomes most clearly linked to social contact with citizens. Our theory does not speak to whether top-down Accountability or views on the appropriateness of Corruption should be shaped by officer embeddedness, and we do not find significant heterogeneous effects for these outcomes.

We probe the robustness of these patterns in the Supplementary Material. Results hold when controlling for officer characteristics that may correlate with embeddedness, including rank, education, religion, age, gender, time in service, commute time, and public service motivation. They also hold when interacting control variables with the treatment indicator. The inclusion of station fixed effects mitigates concerns about the PNP disproportionately assigning non-embedded officers to municipalities where the program was more likely to succeed (for example, stations with higher leadership buy-in), because the model compares effects among embedded and non-embedded officers at the same station. Still, station assignments of officers from outside and inside Sorsogon are balanced on crime rate, NPA presence, and a range of socio-demographics, and the results hold when replacing the station effects with municipality-level controls. Finally, results are similar when using an alternative measure of embeddedness—whether officers live in the municipality where they are assigned.

Drawing from these observations, we posit that programs intended to enhance interactions with citizens will be more likely to improve attitudes among officers who are less embedded ex-ante, assuming that they hold more negative suspicions of citizens than local officers.Footnote ¹⁷ Our theory relies on two propositions that challenge common assumptions in the community policing literature. First, existing work largely assumes that, under a system of reactive policing, informal and non-threatening interactions with citizens are rare for all officers. We contend that embedded officers are already more likely to have these types of interactions with local citizens during their daily lives, outside of work. Research suggests that prior contact can increase the likelihood that people will seek out future contact (Pettigrew Reference Pettigrew1998), meaning that embedded officers may also already be more likely to engage with citizens in the course of their regular policing duties. Simply put, community policing is less likely to generate new experiences with citizens for embedded officers, preventing the social contact mechanisms from taking hold.

Second, the canonical community policing literature places outsized emphasis on information about citizens as being the key mechanism that can unlock improvements in officer trust and empathy. The results displayed in Figure 2 suggest that Shared Citizen Concerns increased for both embedded and non-embedded officers, meaning it is unlikely that non-embedded officers saw improved attitudes solely because they benefited more in terms of information gained through contact. This is not to say that the informational mechanism is altogether unimportant. Rather, our claim is that information may need to be activated together with other mechanisms in order for contact with citizens to result in improved officer attitudes, and that these mechanisms may be blunted for embedded officers.

Consistent with our field staff’s observations about local officers’ comfort and communication style during the POP meetings, we theorize that embedded officers are already less likely to be anxious about interacting with citizens in the absence of community policing, as well as better able to take the perspective of local citizens. This translates to reduced potential to see movement in their attitudes. Anxiety-reduction is a crucial moderator shaping outgroup attitudes (Pettigrew Reference Pettigrew1998; Pettigrew and Tropp Reference Pettigrew and Tropp2008), and information has a bigger effect on outgroup attitudes when paired with improved perspective-taking (Adida et al. Reference Adida, Lo, Platas, Prather and Williamson2024; Kalla and Broockman Reference Kalla and Broockman2023; Pettigrew and Tropp Reference Pettigrew and Tropp2008). It may also be the case that embedded officers already share aspects of their identity with local citizens, in this case, as a “Sorsogueño.” When individuals share identity features, their attitudes are less malleable because they tend to see new information as “situational,” rather than contributing to their views of an outgroup (Lyall Reference Lyall2010; Lyall, Blair, and Imai Reference Lyall, Blair and Imai2013; Paolini et al. Reference Paolini, Harwood, Rubin, Husnu, Joyce and Hewstone2014). In contrast, unembedded officers are more likely to develop a new common group identity shared with locals, a key mechanism through which social contact can trigger changes to outgroup attitudes (Gaertner et al. Reference Gaertner, Dovidio, Rust, Nier, Banker, Ward and Mottola1999; Hewstone, Rubin, and Willis Reference Hewstone, Rubin and Willis2002).

Our theory is important in light of the tremendous variation in police embeddedness around the world. This variation is driven by a debate about the tradeoff between officer effectiveness and partiality, highlighted by the literature on bureaucratic embeddedness (Pepinsky, Pierskalla, and Sacks Reference Pepinsky, Pierskalla and Sacks2017). Embedded bureaucrats often exhibit greater local knowledge, higher effort, and provide better services (Bhavnani and Lee Reference Bhavnani and Lee2018; Ricks Reference Ricks2016), but they are also prone to favoritism and further alienating marginalized populations (Fjeldstad Reference Fjeldstad2005; Haim, Nanes, and Davidson Reference Haim, Nanes and Davidson2021). Reflecting this tradeoff, police departments in the US differ in whether they require officers to live within the city boundaries where they serve (Kennedy et al. Reference Kennedy, Butz, Lajevardi and Nanes2017), and officer embeddedness varies from 7% of officers living locally in Miami to 88% of officers living locally in Chicago (see Figure 3). This discussion is also salient in countries with weaker police capacity and a high risk of corruption. In Mexico, for example, policymakers debate whether to assign local officers who are better at identifying key actors in the drug trade, or non-local “Federales” who are less likely to collude with traffickers.

Figure 3. Distribution of Police Embeddedness in US Cities

Note: Data from the U.S. Equal Employment Opportunity Commission (EEOC) and the Census Bureau. Displayed are the 25 largest US police forces—points are sized by the number of officers employed by the force. For an expanded description of these data, see https://fivethirtyeight.com/features/most-police-dont-live-in-the-cities-they-serve/.

In light of this debate, we offer a few conjectures on where community policing might be most beneficial. Our results imply that community policing provides experiences similar to those embedded officers already hold, but this does not mean that community policing should be eschewed in favor of recruiting officers who are already embedded in their community. Optimistically, community engagement could help close the gap in trust and empathy between embedded and non-embedded officers, perhaps allowing communities to benefit from the upsides of police embeddedness without being subject to as many of the downsides. One idea would be to focus community policing efforts in places where police embeddedness is low, or to disproportionately assign unembedded officers to these programs, while focusing on other reforms to improve attitudes among officers with extensive local experience. That said, officer attitudes are not the only relevant policing outcomes; embedded officers may still be more effective at using the platform of community policing to improve citizen-facing outcomes. If this is the case, police departments may face a tradeoff, whereby they can either prioritize improving the attitudes of officers or citizens, but not both.

Safety Context and Threats to Personal Safety

A second major takeaway from our qualitative observations concerns how the local safety context shaped the content of information to which officers were exposed while participating in community policing. Learning more about the issues that citizens dealt with on a day-to-day basis shaped officers’ understanding of citizen-facing public safety issues, but it also affected assessments of their own personal safety. Specifically, community policing heightened the salience of threats to officer safety in areas that were proximate to the NPA insurgency.

Although none of the POP teams officially designated the NPA threat as their focus issue, we found that barangay leaders in NPA-affected municipalities used the POP meetings to raise instances where NPA personnel were sighted or suspected of recruiting at-risk youth.Footnote ¹⁸ Our field staff reported that when these issues were raised, officers tended to act uncomfortable and express outward frustration that citizens were not more forthcoming with real-time information about NPA activities. This reaction was likely tied to officers’ fears of NPA ambushes, which was the most common threat to officer safety mentioned in our interviews. Many officers told stories about someone they knew on the force who had been killed, injured, or kidnapped by the NPA. When ambushes occur, officers often blame the citizen population for alerting the NPA about police movements or, at the very least, failing to provide the police with information that would allow them to avoid unexpected encounters.

Notably, officers who conducted community policing in NPA-affected areas tended to express frustration with the broader citizenry, even though the individuals who attended community policing events were largely law-abiding, cooperative, and respectful (see Hanson, Kronick, and Slough Reference Hanson, Kronick and SloughForthcoming). We think that information about insurgent activity led officers to infer that a larger proportion of citizens must be sympathetic to the NPA, given that the group survives by hiding amongst the population. Officers focused on citizen behavior being shaped by NPA propaganda while minimizing citizens’ own fear of retaliation, a well-known factor that impedes citizen information-sharing about insurgents.

These observations led us to investigate whether the effect of community policing was conditioned by the baseline level of NPA presence in the municipality where officers were stationed. To measure NPA presence, we rely on internal military intelligence reports that code barangays on a 3-point scale of “cleared” (0), “influenced” (1), or “controlled” (2) by the NPA.Footnote ¹⁹ The latter two categories indicate that insurgents regularly take residence or have an active party organization in the barangay. Our measure proxies for NPA activity by calculating the mean value of the 3-point scale for all barangays in a given municipality. While our sample was restricted to “cleared” barangays, NPA personnel still interact with civilians and try to expand their presence into these barangays from their nearby areas of control.Footnote ²⁰ In Figure 4, we begin by focusing on the heterogeneous effect of NPA presence on officer Trust, the citizen-facing outcome most directly tied to officer safety.

Figure 4. Heterogeneous Effect of Community Policing on Officer Trust, by NPA Presence

Note: All plots display marginal effects of treatment by NPA (rebel) presence in the municipality, using the specification controlling for baseline outcomes (Equation 1). The top panel displays results for the main trust index and a separate outcome asking officers to rate civilians’ sympathy toward the NPA in their assigned municipality. The bottom panel displays results for the individual survey items that compose the trust index. Confidence intervals are displayed at the 95% level (light gray) and 90% level (dark gray). Full regression results are in Table A.7 in the Supplementary Material.

Participation in community policing in NPA-adjacent areas substantially decreased officers’ trust in civilians (see the Trust index in the top-left panel). On the other hand, in municipalities without an active NPA presence, community policing marginally increased trust in civilians. This pattern is especially pronounced for officers’ perception of their own safety (Trust–Safety) and of citizens’ concern for officer well-being (Trust–Intent), displayed in the bottom panel of Figure 4. In the Supplementary Material, we show that the interpretation of the marginal effects is consistent when using the Hainmueller, Mummolo, and Xu (Reference Hainmueller, Mummolo and Xu2019) binning estimator, and when including a battery of municipal-level controls.

The heterogeneous effects for the remaining outcome indices are shown in the Supplementary Material. There was no significant heterogeneous effect on Shared Citizen Concerns, for which the average treatment effect is positive across the full range of the NPA Presence moderator. This is consistent with the idea that officers learned about the community’s concerns regardless of where they were assigned, but the content of those concerns differed. To examine this mechanism, we looked to a separate survey item that asked officers to estimate the percentage of citizens in their municipality who are sympathetic to the NPA (top-right panel of Figure 4). Consistent with our qualitative observations, treatment officers in low-NPA areas perceived that citizens were less sympathetic to the NPA (relative to the control group), while treatment officers in high-NPA areas perceived that citizens were more sympathetic to the NPA.Footnote ²¹ We do not find significant heterogeneous effects on officer Empathy, the other citizen-facing outcome. This result is not surprising in light of the fact that we operationalized this concept to focus on the seriousness of crimes and complaints that citizens did report to the police. As described above, the most relevant aspect of empathy in the context of the NPA is whether officers understand citizens’ choice not to report rebel activity.Footnote ²²

Based on these patterns, we challenge the idea that the information officers attain through community policing will uniformly improve officers’ expectation of personal safety. Although this assumption may hold in “safe” areas, where citizens are likely to raise concerns that pose little threat to officers, we predict that additional information about citizen concerns in “dangerous” areas can increase officers’ perceived risks to personal safety. This can generate anxiety and worsen trust in citizens. Our hypothesis is again rooted in intergroup contact theory: “positive” contact can improve outgroup attitudes, while “negative” contact has the reverse effect (Lowe Reference Lowe2021; Paolini, Harwood, and Rubin Reference Paolini, Harwood and Rubin2010; Schäfer et al. Reference Schäfer, Kauff, Prati, Kros, Lang and Christ2021). Even when officers experience both positive and negative contact during community policing, their overall trust may decrease because negative contact tends to be a more powerful predictor of outgroup attitudes (Barlow et al. Reference Barlow, Paolini, Pedersen, Hornsey, Radke, Harwood and Rubin2012). Importantly, this process does not require that direct interactions with citizens during community policing are antagonistic. Rather, negative contact can occur from exposure to information that highlights threats posed by outgroup members, regardless of whether the threat comes from the individuals with whom one interacts (Schäfer et al. Reference Schäfer, Kauff, Prati, Kros, Lang and Christ2021). Negative information can raise intergroup anxiety and harden identity divides between officers and citizens, two of the strongest mediators linking negative contact to attitudinal change (Paolini, Harwood, and Rubin Reference Paolini, Harwood and Rubin2010; Pettigrew and Tropp Reference Pettigrew and Tropp2011).

Our theory is particularly powerful when combined with insights on how “neighborhood context” affects police officers (Paoline, Myers, and Worden Reference Paoline, Myers and Worden2000; Terrill and Reisig Reference Terrill and Reisig2003). The key insight is that rather than overcoming the effects of neighborhood context (as is commonly assumed), community policing can heighten the effects of neighborhood context on officers’ attitudes and behaviors. This mechanism corresponds to Nix, Pickett, and Wolfe’s (Reference Nix, Pickett and Wolfe2020) claim that working in violent communities may condition officers to expect more noncompliance, disrespect, and violence from citizens. We contend that the information officers acquire during community policing can worsen harmful stereotypes of the citizenry, even if officers primarily interact with cooperative, law-abiding citizens. We focused on the threat posed by insurgency due to its relevance in Sorsogon. The literature on neighborhood context highlights additional threats like organized crime, high gun ownership rates, or a history of citizen antagonism toward the police, which are less relevant in Sorsogon but may yield similar effects in settings where they are present.

One troubling implication of our theory is that community policing may be most likely to fail in places where improvements to officer–citizen relations are most needed. Officers assigned to areas with high safety threats may become more guarded and more likely to think force is necessary, which suggests the need to explore alternative reforms in high risk areas. We do not suggest that the police should become less engaged or more militaristic, which could have negative effects on citizen-facing outcomes. Instead, departments may consider alternative modalities of community policing that retain a citizen-centered approach without immediately ramping up contact on a large scale. The police could also incorporate perspective-taking interventions into officer trainings, which may serve to set appropriate expectations and help officers see the world from citizens’ perspectives.