Abstract
This study evaluates the short- and long-term effects of a novel, immersive training programme designed to increase police officers’ willingness and ability to prevent ethnic profiling. The programme combines virtual reality (VR) simulations with group-based dialogue, grounded in constructive controversy methodology. Using a longitudinal design, 519 Dutch police officers participated in one of three conditions: no training, one-time training, or repetitive (three-times) training. To assess effects, participants completed questionnaires measuring individual-level variables (ethnic reform attitude, knowledge, self-efficacy) and team-level variables (group efficacy, psychological safety, intragroup conflict). These measures were collected at five time points: baseline (a month before training), directly after each training session, and 6 months after the final session. Results from linear mixed models indicate that training significantly affected both individual and team-level outcomes. One-time training was primarily effective in increasing knowledge and improving psychological safety. Repetitive training led to a temporary decline in self-efficacy and group efficacy immediately after training, but only through repetition were sustained long-term improvements observed in attitudes, self-efficacy, and group efficacy. Psychological safety improved following one-time training and remained stable over time, with no additional benefits from repeated sessions. These findings suggest that while single-session training may produce short-term gains in knowledge and safety perceptions, repeated training is essential to achieve lasting changes in attitudes, efficacy beliefs, and team dynamics. For police organisations aiming to reduce ethnic profiling, training must go beyond one-off interventions and instead focus on sustained, repeated efforts to embed change at both individual and organisational levels.
Keywords
Over the past decades, law enforcement agencies in Europe and North America have faced mounting criticism from citizens, politicians, and NGOs for the disproportionate targeting and harsher treatment of ethnic-minority citizens (e.g., Amnesty International, 2020; Narrillos, 2020). These practices are commonly referred to as ethnic profiling, defined here as ‘the use by the police, with no objective and reasonable justification, of grounds such as race, colour, language, religion, nationality, or national or ethnic origin in control, surveillance, or investigation activities’ (Council of Europe, 2019). In response to the criticism, the issue of ethnic profiling reform is high on the agenda (Henninger, 2022; Narrillos, 2020), and many police departments are implementing reforms to address ethnic profiling. These include bias-awareness training (Burke, 2019; Worden et al., 2024) and the development of ethical guidelines (Bradford et al., 2015; Rosenbaum & Lawrence, 2017). In the Netherlands, the police adopted a Code of Practice (CoP) for proactive police stops, aligning with international reform efforts (Nationale Politie, 2021). This policy states that the use of racial indicators, such as ethnicity and skin color, in risk profiles and selections is permitted only when officers are acting on a specific suspect description.
Despite these efforts, reform initiatives often meet internal resistance. Many officers deny the existence of bias or view concerns about ethnic profiling as exaggerated (Hine, 2015; White & Schafer, 2024). Some have protested against reform efforts, especially after high-profile incidents like the death of George Floyd (Knowles & Stanley-Becker, 2020). Research also suggests that continued public scrutiny can lead to emotional fatigue among officers, causing disengagement from reform processes (Cooper, 2020; Shiner, 2010). Within police teams, the fear of being judged by colleagues can make it difficult to engage in open dialogue. For example, a recent study in the Netherlands highlighted how reform-minded officers fear being labelled ‘woke’ or ‘racist’ (Böing et al., 2025a). This internal tension is a significant barrier to organisational change.
The current study evaluates a scientifically grounded training programme designed to improve officers’ willingness and ability to prevent ethnic profiling. To determine the importance of training frequency, we compared the outcomes of repeated training, one-time training, and no training. Both short- and long-term effects were examined to assess the durability of the intervention’s impact.
The training consists of two parts: an interactive, 360°-VR simulation simultaneously administered to small groups of police officers, followed by a group dialogue in which all officers take part (Böing et al., 2025b). The latter employs the method of constructive controversy (Johnson et al., 2000) to stimulate officers to reflect on the choices made in the VR simulation.
Building on earlier research into antecedents of police officers’ willingness to participate in reform attempts (e.g., Böing et al., 2025a), we assessed the effect of participation in this training programme on both individual-level variables (ethnic reform attitude, knowledge, self-efficacy) and group-level variables (group efficacy and psychological safety). Given the presumed importance of constructive controversy dialogue following the intervention, we included intragroup conflict about ethnic profiling as a group-level process variable.
This has led to the following research question: to what extent does participation in repeated, one-time, or no training influence police officers’ attitudes towards ethnic profiling, knowledge, self-efficacy, and group-level outcomes such as group efficacy, psychological safety, and intragroup conflict over time?
While reform training is often framed as a tool for improving attitudes and competencies, it may also cause discomfort, resistance, or conflict – especially when it addresses sensitive topics like ethnic profiling and racial bias. Rather than viewing such reactions as failures, this study explores whether temporary discomfort may, in fact, be a necessary phase in the process of sustained reform engagement.
Resistance to Reform Participation
Resistance is widely recognised as a barrier to police reform (Skogan, 2008; Walker, 2012). As highlighted in prior research, reform initiatives around ethnic profiling can evoke defensive responses among police officers, particularly when framed or experienced as corrective or accusatory (Cooper, 2020; Shiner, 2010). Officers may perceive such efforts as offensive and undermining their professional integrity, which can elicit anger, resistance, or disengagement (Landman & Sollie, 2018; Murrar & Brauer, 2019).
Resistance to reform in the context of ethnic profiling can be further understood through the framework developed in earlier work by Böing et al., 2025a reflecting underlying variation in key psychological and social factors, both at the individual and group level. Specifically, Böing et al., 2025a distinguish four categories of psychological and cultural barriers to reform engagement.
First, the “Don’t Know” category includes officers who do not perceive the relevance of the issue to their own work. These individuals often view themselves as neutral and unaffected by bias, suggesting limited knowledge and little perceived need to improve. Low self-efficacy may also play a role, as some may doubt their ability to meaningfully address such complex topics.
The “Don’t Care” category refers to officers who are aware of ethnic profiling but minimize its significance, viewing it as exaggerated or politically motivated. Their attitudes toward reform tend to be dismissive, and they may believe current practices are justified by operational realities, such as crime statistics.
The “Don’t Want” group includes officers who actively oppose reform, often due to identity-related or moral objections. They may feel accused by public discourse and perceive reform efforts as threatening to their professional values. In such cases, group-level factors such as low group efficacy or high intra-group conflict may reinforce their resistance, particularly if reform is seen as imposed from outside rather than arising from within the team.
The “Don’t Dare” category captures officers who may be open to reform or hold critical questions, but hesitate to speak up due to a lack of psychological safety within their team or unit. Fear of judgment or social exclusion may lead to silence, even when individuals privately support change. What becomes clear from the previous, is that police officers’ reform engagement depends on factors both on individual as well as the team level.
Enhancing Reform Participation
Drawing on Böing et al., 2025a, we identify six variables that may shape officers’ willingness and ability to participate meaningfully in reform: attitude, knowledge, and self-efficacy at the individual level, and group efficacy, psychological safety, and intragroup conflict at the group level.
Attitude toward reform plays a foundational role in shaping behavioural engagement. Officers are more likely to participate in reform efforts when they perceive them as legitimate, relevant, and congruent with their professional values (Eby et al., 2000; Kotter, 2007). Conversely, where reform is perceived as politically driven or accusatory, officers may adopt a defensive stance that undermines engagement (Cooper, 2020; Shiner, 2010). For those who feel unfairly targeted, reform initiatives may be rejected outright. Shifting attitudes, therefore, may be critical for enabling reform uptake, particularly among those who occupy the “don’t want” or “don’t care” categories (cf. Dunbar & Hanink, 2023).
Knowledge of legal standards and ethical codes is essential for professional decision making and for preventing biased practices such as ethnic profiling. Officers who understand these frameworks are better equipped to interpret and apply them in complex operational contexts (Fridell & Lim, 2016). In contrast, reliance on heuristics or “gut feelings” can inadvertently lead to racialised assumptions (Eliasson, 2021). Improving officers’ legal and normative knowledge may be particularly important for those who fall into the “don’t know” category.
Self-efficacy, or the belief in one’s ability to make a difference, can also influence reform participation. Officers who feel confident in their capacity to act are more likely to take initiative and apply new insights in practice (cf. Bandura, 1977). However, self-efficacy may be constrained by peer norms or fear of social sanction. Officers in the “don’t dare” category, for example, may possess the motivation and understanding to change but hesitate to act due to perceived group pressures (Böing et al., 2025a). This shows that reform participation is also embedded within broader social dynamics.
At the team level, group efficacy – the shared belief that a team can achieve meaningful change – may foster collective motivation and resilience (van Zomeren et al., 2008). When reform is seen as a shared endeavour, rather than an individual burden, officers may be more likely to support each other, persist through setbacks, and model new behaviours.
Psychological safety refers to the perception that one can speak openly, ask questions, or admit uncertainty without fear of judgment or repercussion (Edmondson, 1999). This sense of safety is critical for officers to engage with potentially sensitive topics such as bias, profiling, or community relations. In team environments where officers fear being seen as “soft” or disloyal, silence may not indicate apathy, but fear (Skogan, 2008). Psychological safety thus creates the interpersonal conditions that enable open participation in reform.
Finally, intragroup conflict can stimulate meaningful engagement with reform when it remains focused on the task rather than personal dynamics. Disagreements about how to interpret or implement reform measures may prompt officers to articulate their reasoning, examine assumptions, and consider alternative perspectives (Johnson & Johnson, 2009). This type of task-related conflict can be a productive force, particularly when managed within a psychologically safe environment. Research suggests that moderate levels of task conflict are most conducive to innovation and learning, while too little or too much can reduce effectiveness (De Dreu, 2006). However, when task conflict becomes entangled with relational tensions, it can undermine trust and hinder participation (De Dreu & Weingart, 2003). Supporting teams in navigating these tensions constructively can help ensure that disagreement serves as a catalyst for growth, rather than a barrier to reform.
Together, these six variables provide a framework for diagnosing why officers may or may not engage with reform. They also suggest a set of leverage points change. The following section outlines intervention strategies that target these underlying factors at both individual and group level.
Strengthening the factors that underlie reform engagement requires training strategies that go beyond didactic instruction and instead incorporate experiential, immersive, and socially grounded learning approaches. A growing body of socio-psychological research supports the use of such interventions to promote deeper engagement and behavioural transfer. This research also offers strategies for strengthening these mechanisms through targeted learning designs. Rather than relying on abstract instruction, effective training incorporates experiential, immersive, and socially grounded methods that activate both individual-level competencies and group-level processes.
Experiential learning approaches such as role-play, scenario-based exercises, and guided reflection may be particularly effective in building knowledge, self-efficacy, and attitudinal openness. These methods provide a low-risk environment in which officers can test assumptions, explore ethical dilemmas, and receive feedback (Kolb & Kolb, 2005). When scenarios are grounded in real-life policing situations, they may also increase the perceived relevance of reform content, particularly among those who are initially skeptical (Murrar & Brauer, 2019).
Virtual reality (VR) expands on these principles by offering immersive simulations that replicate emotionally charged or ethically complex encounters. Officers can engage with realistic dilemmas, such as biased stops or community complaints, without real-world consequences (Hube et al., 2024; Kwon, 2019). Repeated exposure in these environments may strengthen self-efficacy and reduce defensiveness by allowing officers to practice and reflect in private. Moreover, VR’s novelty and realism may improve training uptake, especially among those with low initial motivation (Cornet & van Gelder, 2020).
To target group-level mechanisms, social and affective learning strategies play a central role. Among these, constructive controversy has shown particular promise. In this method, officers are encouraged to voice differing viewpoints, defend their reasoning, and collaboratively refine their perspectives (Johnson & Johnson, 2009). By making disagreement a structured part of the learning process, constructive controversy fosters psychological safety and may reduce the fear of social sanction. Over time, it may also strengthen group efficacy, as officers learn to coordinate around difficult topics and build mutual understanding.
Importantly, such dialogue-based approaches do not require full consensus. Rather, they aim to develop a team climate in which diverse viewpoints are respected, and disagreement is viewed as an opportunity for collective growth. In policing environments often characterised by strong occupational norms and hierarchical dynamics, this approach may be particularly useful in surfacing hidden resistance and facilitating reflective engagement with reform.
Together, these methods address both the internal drivers of reform participation and the interpersonal dynamics that support sustained behavioural change. By aligning training design with the mechanisms identified in the diagnostic model, reform initiatives can become not only more effective but more responsive to the real-world challenges officers face.
The Designed Training
In response to these conceptual insights, a group-based training programme was developed in the Netherlands, combining immersive VR simulations with structured group dialogue (Böing et al., 2025b). Officers begin with a 360° film VR scenario, taking on the role of a patrolling officer who must make real-time decisions in ambiguous situations. One example is a train station scenario in Amsterdam. Participants observe dozens of people passing through the station and must choose whom to approach and how to act. For instance, they may decide to engage with four young men who appear suspicious due to stereotypical cues, or focus on a man approaching passers-by who is later revealed to be selling drugs. Choices in the simulation are time-bound and follow legal and ethical principles, giving officers a sense of agency and responsibility for their decisions.
After the VR experience, participants engage in a structured group dialogue using the method of constructive controversy. Officers who made less appropriate choices are invited to explain their reasoning, while peers challenge and question these decisions. The discussion encourages participants to reflect critically on their assumptions, explore alternative perspectives, and link their actions to legal and ethical standards. This combination of immersive experience and guided dialogue aims to foster both cognitive and emotional engagement, helping officers recognize, prevent, and respond to ethnic profiling in practice.
Preliminary studies using the training prototype suggest that the programme has potential to enhance officers’ engagement with reform efforts and improve their ability to recognise instances of ethnic profiling. Participants in an initial pilot training responded positively to the VR component, showing high levels of immersion and engagement. Following the simulation, these officers were also observed engaging in critical reflection on the decisions they had made, indicating a degree of self-awareness and openness to examining their own assumptions and biases (Böing et al., 2025b). An experimental study by De Vries et al. (2023) further examined the training’s effects and found that participating officers demonstrated increased knowledge of ethnic profiling. However, the study did not find significant effects on participants’ attitudes toward reform or their willingness to discuss the topic with colleagues. One possible explanation for this limited impact is that the training was delivered only once, rather than being reinforced through repeated sessions. Repetition may be necessary to influence more deeply rooted beliefs and conversational norms.
The Importance of Duration and Repetition in Sustained Change
Changing deeply ingrained attitudes and fostering confidence in discussing sensitive issues—such as countering ethnic profiling—may require more than a single training session. Research consistently supports the idea that one-time interventions are insufficient for creating lasting change. For example, Lai and Lisnek (2023) found that a one-time training session on countering implicit bias led to only a temporary increase in knowledge and awareness, with the effects dissipating over time. Similarly, van der Meulen et al. (2018) conducted a quasi-experimental study examining psychological resilience. Although a three-day training session, supplemented by a follow-up day, produced short-term improvements in resilience and mental health, these effects were not sustained over time. Other studies have highlighted the importance of repeated training, for instance to improve attitudes towards training (Givron & Desseilles, 2021), performance (Franiatte et al., 2024; Kievik, 2017) and self-efficacy; (Odgers et al., 2022).
Taken together, these findings suggest that improving individual attitudes, knowledge, and self-efficacy in sensitive or difficult-to-change areas likely requires a sustained, iterative approach. Multiple sessions, spaced over time, could support deeper reflection and engagement in reform, enabling the progressive reinforcement of learning. This suggests that retention and behavioural change are more likely when learners revisit material periodically and engage actively through reflection and application. Moreover, repetition may foster the conditions for ongoing reflection and the gradual recognition of unexamined or ineffective group norms, such as those that may underlie ethnic profiling.
In short, lasting impact depends not only on what is taught, but also on how often and under what conditions. Regular training, combined with both short-term and long-term evaluation, may be essential to ensure that positive outcomes are reinforced, internalised, and sustained (Moreno et al., 2024).
The Present Study
Based on the considerations and the research question outlined in the introduction, we formulated and preregistered the following hypotheses: • • • •
Method
Transparency and Openness Statement
We preregistered the study and the analysis plan at AsPredicted (https://aspredicted.org/58tq-zgbm.pdf and https://aspredicted.org/4f6n-75sf.pdf). Analyses were conducted using R version 4.4.3 and R Studio, 2026.01.0; in conjunction with the Center for Open Science (Nosek et al., 2015), all data and R code have been published in OSF, https://osf.io/wy9h6/?view_only=3a03911b39c54761909c5a09815a1184. The study was approved by the Ethics Committee of the authors’ university (request nr. 221390).
Design
The study uses a 3 (Training: repetitive vs. one-time vs. no) x 5 (Time of Measurement: T0 vs. T1 vs. T2 vs. T3 vs. T4) mixed design. Training is varied between participants, whereas Time of Measurement (ToM) constituted a within-participants independent variable. The study also distinguishes between immediate (short-term) and durable effects (long-term) of training (T1-T3 for immediate and T0-T4 for long-term effects). An overview of the design is illustrated in Figure A1 in Appendix A.
Participants were nested within nine distinct police stations (Team). We corrected for potential diffences between these police teams, for example because of differences in team culture, or shared experiences and incidents.
Sampling
For context, since 2013, the Netherlands has operated as a single, centralized national police organization (Nationale Politie), consisting of approximately 51,000 sworn officers and 14,000 civilian staff. Operational policing is organized geographically into local police stations serving defined service areas.
A call for participation in this longitudinal study was distributed via LinkedIn among Dutch police chiefs. Approximately twenty chiefs indicated that they were able to facilitate the entire research project within their police stations.
Nine of these stations, comprising between 108 and 270 officers, were purposively selected to ensure substantial contextual heterogeneity. The selected stations were located in different parts of the Netherlands and served municipalities varying in ethnic composition and urbanicity. The aim was not statistical representativeness, but to avoid overrepresentation of a specific type of policing environment and to enhance external validity.
Stations were then assigned to the three experimental conditions based on the level of ethnic diversity in the municipalities they served. To reduce contextual confounding, each condition (repetitive training, one-time training, and no training) included one station from a municipality characterized by high (>35%), moderate (20–35 %), and low (<20%) levels of ethnic diversity. This decision was informed by prior research suggesting that contextual ethnic diversity may influence officers’ experiences during police–citizen interactions and their receptivity to antibias training (Pickett & Ryon, 2017; Wolfe & Nix, 2016).
Station commanders selected 100 participants from their staff, taking into account officer availability over the 10-month study period to ensure continuity. As a result, some officers, such as cadets, or those on leave for pregnancy or vacation, were excluded from selection.
In accordance with the preregistration, accidental duplicate entries and entry errors that could not be resolved (16 cases) were excluded from the analyses, as were participants showing indications of malintent (as evidenced by filling in multiple questionnaires, completion times of under 1 minute, identical-answer patterns, and remarks in open-ended questions; three cases). Finally, 23 cases were removed because they dit not give their consent.
Some 536 participants started the pre-measurement (T0) one month before the first training; 519 of those actually completed it. On average, 27 % were female, and 69 % were male (4 % did not disclose) 1 . Ethnically, 53 % identified as native Dutch and 37 % as non-native Dutch (9 % did not disclose). In terms of work experience, 31 % had less than ten years of service, 32 % had between ten and 20 years, and 37 % had more than 20 years experience 2 .
Participation in the post measurement (T4) dropped to 245. In the no-training condition 44% of the participants remained; here attrition was likely attributable to participant fatigue associated with completing multiple surveys over time, particularly given the absence of an active or engaging training component. In the repetitive-training condition, of which 37% remained, drop-out was largely concentrated on a single police station. This followed a high-profile incident in which several team members were accused of racist behaviour (NL Times, 2023). In response to the incident, police leadership decided to terminate participation of the entire police station to protect officers’ well-being and prevent further discomfort. Consequently, attrition was driven by external, organizational circumstances rather than individual withdrawal. In the one-time condition 71% remained (see Figure A2 in the Appendix).
Importantly, attrition did not appear to have led lead to shifts in demographic distributions. Of the remaining 245 officers, 28% were female and 70% male (2% did not disclose); 56% identified as native Dutch and 36 % as non-native Dutch (7% did not disclose); 34% had been in service for less than ten years, 28% between ten and 20 years, and 39% had more than 20 years experience; 28% were female, 70% male, and 2% did not disclose). In addition, to examine whether the police that dropped out differed from the other two stations in the Repetitive condition, we conducted a multivariate analysis of variance with attitude, group efficacy, self-efficacy, and psychological safety, measured at T0, as dependent variables. This showed there was no significant difference between the police stations, F (8, 496) = 0.97, p = .54, Wilks’ Lambda = .97. The police that dropped out therefore do not seem to be different from the remaining police, when it comes to the features relevant for this study. We therefore found no evidence that the attrition introduced a bias.
Procedure
All participants were assigned a unique code to track their responses across multiple measurements. For each police station (Team A tm I) there were 100 unique codes available. This code was provided on a keychain distributed at the pre-measurement (T0) and used for subsequent surveys. Participants who underwent training (in the repetitive and one-time conditions) filled out a questionnaire 2 to 3 days post-training, while those in the no-training condition completed surveys in the third week of each month. A final measurement was done six months after the last training session.
In the repetitive-training condition, three VR simulations were administered over several months. On T1, participants experienced a “Parking Lot” scenario in a socioeconomically disadvantaged neighbourhood, where they encountered three young men from an ethnic minority near an expensive car. In T2, the “Parliament” scenario took place outside the Dutch House of Representatives, where participants decided between investigating a potentially suspicious individual or interacting with demonstrators. Finally, in T3, the “Train Station” scenario involved choosing whether to engage with a group of young men or an individual later revealed to be selling drugs. Whereas participants in the repetitive-training condition participated in all three training sessions, those in the one-time training condition only took part in the training on T3. Participants in the no-training condition did not participate in any of the training. See Figure A1 in Appendix A for an overview.
In each scenario, participants were prompted with multiple decision options, including both escalating (e.g., search, arrest) and de-escalating (e.g., cease interaction) actions. They had 5 seconds to make each decision, simulating real-world time pressure. After the VR session, participants participated in group dialogues facilitated by trained moderators, who led discussions on their choices, linking back to the Dutch CoP. A more detailed description of these three scenarios can be found in the Appendix B.
Measures
Police officers are a relatively difficult target group for studies involving surveys; they often exhibit low response rates, stemming from a general mistrust of researchers, survey fatigue, and limited perceived relevance of academic studies to their day-to-day work (cf. Khanizadeh et al., 2023). To minimise drop-out, the surveys were kept relatively short. Therefore, we included shortened scales for our measures, supported by systematic evidence that the reliability and validity of short and single item measures is often as good as of complete schales (Allen et al., 2022).
Unless stated differently, all items were measured on a Likert five-point scale (with 1 being ‘strongly disagree’, and 5 ‘strongly agree’) and included at all data collection moments.
On the individual level, Attitude was determined with a three-item measure, created for this study: ‘It is a good thing that the organisation focusses on professionalising police stops’, ‘It is a good thing that the organisation focuses on preventing ethnic profiling’, and ‘I do not find it necessary to increase public trust in the police (reversed)’. These items were developed based on face validity, ensuring that the content of the question would be recognised by the participating officers. The construct demonstrated moderate internal consistency (Cronbach’s α = .69).
Knowledge was measured in two ways. We first asked about participants’ awareness of the Code of Practice (CoP) covering the rules of engagement for proactive police stops with one item: ‘I know what the Code of Practice is about’ (with 0 being ‘no’, and 1 ‘yes’). Furthermore, we asked participants to indicate which of the provided grounds for proactive police stop were legitimate. The possible answers included two correct answers (e.g., ‘criminal behaviour’ and ‘behaviour fitting a Modus Operandi’) and three incorrect answers (e.g., ‘gut feeling’, ‘out-of-place’, and ‘crime overrepresentation’); an index was calculated, ranging from 0 to 5, where 5 meant all legitimate and illegitimate options were correctly recognised and 0 represented no correct answers.
This measure is directly relevant to bias-free policing in the Dutch context. The CoP (‘Handelingskader Professioneel Controleren’) provide clear guidelines on the grounds for a legitimate stop, aiming to prevent ethnic profiling. Respondents who correctly identify the legitimate grounds demonstrate knowledge that can help prevent bias-based stops, whereas those unable to identify these grounds are at greater risk of conducting stops in ways that could be perceived as discriminatory or profiling based on ethnicity.
Self-efficacy was measured with three-items, informed by work of Kievik et al. (2020), but tailored to this study, e.g., ‘I am able to apply the Code of Conduct on police practices’, ‘I am able to execute police stops in a correct manner’ and ‘I am able to discuss the quality of police stops with my direct colleagues’. Together, these items measure the participants’ confidence in their ability to prevent ethnic profiling. As with the attitude measure, these items were developed with an emphasis on face validity (Connell et al., 2018) and clarity. The construct demonstrated good internal consistency (α = .76).
On the group level, group efficacy was measured with three-items, informed by earlier research (Wu et al., 2010), but tailored to this study. Items were as follows; ‘As a group we are able to apply the Code of Conduct in police practices’, ‘As a group we are able to execute police stops in a right manner’, and ‘As a group we are able to discuss the quality of police stops with my direct colleagues’. The construct demonstrated a relatively high internal consistency (α = .88).
Psychological safety was measured with two-items, from Edmondson (1999): ‘Members on my unit are able to discuss controversial issues’, and ‘My contributions are appreciated in my unit’. We assumed that these questions would have a higher face valitidy among the participants than the other items from Edmondson’s scale to measure psychologiocal safety. The two questions demonstrated a reasonable internal consistency (α = .62). As we were primarily interested in the effects of the training on psychological safety, we included this measure during training sessions in T1, T2 and T3 and for all groups at T0 and T4.
Intragroup conflict about ethnic profiling was framed in a more neutral way by addressing conflict about the (correct) execution of police stops. It is measured with a single-item measure inspired by Curşeu’s (2011) research: ‘On my unit there are often differences of opinions about the execution of police stops’. As we were primarily interested in the effects of the training on intragroup conflict about ethic profiling, we included this measure during training sessions in T1, T2 and T3 and for all groups at T0 and T4.
The demographic variables measured only once at T0 and T4, and included gender (1 = ‘Male’, 2 = ‘Female’, and 3 = ‘Do not want to disclose’, ethnicity (1 = ‘Ethnic Dutch’, 2 = ‘Not Ethnic Dutch’, and 3 = ‘Do not want to disclose’), and years of service (‘<5 years’, ‘6–10’, ‘11–15’, ’16–20’, ’21–25’, ’26–30’, ’31–35’, ’36–40’, ‘>40’).
Analyses
Because the data are nested within participants and police stations, dependency is induced. Dependency that is not dealt with appropriately biases statistical inference (McCulloch & Searle, 2001; Scariano & Davenport, 1987; Van den Berg, 2024); to effectively deal with this, Linear Mixed Models accounting for the nested structure of the data were conducted, with fixed effects for ToM and Training, and random effects for participants and Team (to account for differences between police stations).
To test our hypotheses, planned contrasts are specified. While the first contrast (C1) assessed overall effects of any training versus no training, the following four contrasts (C2–C5) examined more specific comparisons in the short-term; all testing H1. Contrast 6 (C6) tested H2 by examining whether repetitive training had a greater short-term effect than one-time training. Contrast 7 (C7) tested H3, and Contrast 8 (C8) tested H4.
Contrasts C 1 – C 6 are visualised in Figure 1. Elbow connectors visualising planned contrasts for short-term effects (as well as the overall effect); X marks groups participating in any training and O groups with no training
C1: This contrast tests H 1 by assessing the overall effect of any training versus no training. It compares the grand mean of all training conditions (denoted as X) with the grand mean of all no-training conditions (denoted as O). C2: This contrast compares repetitive training at T1 (repetitiveT1) to the grand mean of no training (NoT1) and one-time training (one-timeT1). (Black elbow connector in Figure 1) C3: This contrast compares repetitive training at T2 to the grand mean of no training at T2 and one-time at T2. (Solid blue elbow connector) C4: This contrast compares one-time training at T3 to no training at T3. (Solid green elbow connector) C5: This contrast compares one-time training at T3 to the grand mean of one-time training at T1 and one-time training at T2. (Red elbow connector) C6: This contrast tests H 2 by comparing repetitive and one-time training conditions at T3. (Dashed blue elbow connector)
The two contrasts aimed to assess the long-term effects of training are listed below and are visualised in Figure 2: Elbow connectors visualising planned contrasts for long-term effects (as well as the overall effect); X marks groups participating in any training and O groups with no training
C7: The difference between the repetitive-training condition at T4 and T0 is contrasted with the same difference in the no-training condition. This contrast is illustrated with the dashed green elbow connector. It quantifies to what extent the change over time differs between repetitive training versus no training. C8: The contrast between the difference between T4 and T0 in the repetitive Taining condition is contrasted with the same difference in the one-time training condition. This is illustrated with the black elbow connector. It quantifies the extent to which the change over time differs between repetitive training versus one-time training.
The contrasts to test the hypotheses were performed in R with the lme package, using the method described by Van den Berg (2024).
Results
The following section presents the study’s findings across all outcome measures. For each measure—attitude, knowledge, self-efficacy, group efficacy, psychological safety, and intragroup conflict—we first report short-term effects immediately following training, and then long-term effects over the 10-month study period. Where relevant, we also highlight differences between the one-time and repetitive training conditions, as well as comparisons to the control group. Tables and figures complement the text by summarizing statistical contrasts and trends over time.
Attitude
Planned Contrasts Examining the Interaction of Training and ToM on Attitude
Note. SE = Standard error
*p < .05; **p < .01.
At T3, attitudes reported in the repetitive-training condition contrasted significantly with those in the one-time condition (Contrast 6). However, contrary to expectations this relationship was negative, indicating that directly after the third training session, participants in the repetitive-training condition had less favourable attitudes toward reform and ethnic profiling prevention efforts than those who received only one-time training. This suggests that (the anticipation of) repeated exposure may not strengthen and even weaken support for such interventions in first instance. Based on these findings, we partly accepted H 1 but rejected H 2.
Long-term effects showed a different picture. Over the full study period, from T0 to T4, only participants in the repetitive-training condition showed a sustained improvement in attitudes compared to the no-training condition. This suggests that repeated exposure is more effective in promoting durable attitude change. Table 1 confirms this with Contrast 7, showing a more pronounced increase in the repetitive group than in the control group over time. Notably, there was no significant long-term difference between the one-time and repetitive groups (contrast 8). Figure 3 further illustrates this trend: while the one-time group’s attitude gains taper off after T3, the repetitive group surpasses both other groups at T4. Average attitude ratings (and standard errors) as a function of training and ToM
To conclude, while the repetitive-training condition was associated with more negative attitudes immediately after training, it led to a positive increase over time. This suggests that although repeated training may trigger short-term discomfort, it supports more durable attitude change in the long run.
The random effect of Team showed a variance of .01, compared to .22 of individual effects within teams and a residual variance of .15. This suggests that only a small proportion of the total variance in attitude scores could be explained by team membership.
Knowledge: Awareness of CoP
Planned Contrasts Examining the Interaction of Training and ToM on Awareness of the Code of Practice (CoP)
Note. SE = Standard error
*p < .05; **p < .01.
Long-term effects showed that the differences in awareness between the groups were not sustained over time. No significant differences were found between participants who trained repetitively and those who did no training (Contrast 7), and participants who trained repetitively and those who trained only once (Contrast 8). These planned contrasts suggest that the impact of training on CoP awareness was not sustained over time. As a result, both H 3 and H 4 were rejected.
Hence, while training boosted CoP awareness in the short term, with no added benefit from repetitive training, long-term effects were not sustained. Figure 4 however illustrates that the lack of long-term differences was not caused by declining CoP awareness among the trained participants. Instead, it remained on the same level after training while the awareness in the control group gradually increased. Average ratings of awareness CoP (and standard errors) as a function of training and ToM
Considering random effects, the variance of Team was .01, while the individual variance within teams was .09, and the residual variance was .08, suggesting that only a small proportion of the total variance in these scores could be explained by team membership.
Knowledge About the Legitimate Grounds for Police Stops (Index)
Planned Contrasts Examining the Interaction of Training and ToM on Knowledge About the Legitimate Grounds for a Police Stop
Note. SE = Standard error
*p < .05; **p < .01.
At T3, knowledge reported in the repetitive-training condition contrasted significantly with that in the one-time condition (Contrast 6), indicating that participants who underwent repetitive training performed significantly better than those who trained only once. Consistent with H 2, repeated exposure to the material enhanced retention and understanding. Based on these results, we accepted both H 1 and H 2.
Long-term effects showed that repeated training led to significant higher test scores of the correct grounds for a police stop. Confirming hypothesis 3, Contrast 7 revealed a significant positive effect for participants who received repetitive training compared to those who did not received training; more specifically, the increase in knowledge at T4 compared to T0 was stronger in the repetitive-training condition than in the no-training condition. Contrast 8, testing H 4, also showed a similar difference between participants in the repetitive and one-time training conditions (see Figure 5). Average scores on the knowledge index (and standard errors) as a function of training and ToM
The line plots in Figure 5 further supports this interpretation. They show a clear upward trend in the proportion of correct answers among participants who received training, especially those who trained repetitively, compared to those who did not. Notably, the performance of the repetitively trained group remained consistently higher over time, illustrating the sustained impact of repeated training sessions. This visual pattern aligns closely with the statistical contrasts reported above, reinforcing the conclusion that training, particularly when repeated, has a lasting positive effect on knowledge retention.
Hence, the training improved knowledge in the short term, with repetitive training providing additional benefits. These effects were maintained in the long term, with repetitive training continuing to outperform one-time training.
The variance associated with the random factor Team was .01, and was minimal compared to the individual effects within teams (.27) and the residual variance (.30).
Self-Efficacy
Planned Contrasts Examining the Interaction of Training and ToM on Self-Efficacy
Note. SE = Standard error
*p < .05; **p < .01.
At T3, self-efficacy in the repetitive training group remained significantly lower than in the one-time training group (Contrast 6), suggesting that repeated exposure to the training content may not reinforce confidence. Notably, it may even temporarily undermine it. This suggests that repeated exposure may not strengthen and may even temporarily weaken self-efficacy. Based on these findings H 1 and H 2 must be rejected.
Long-term effects showed a different picture. Contrast 7, testing H 3, revealed a significant positive effect. Participants who received repetitive training showed a greater increase in self-efficacy ratings from T0 to T4, compared to those who received no training. Similarly, Contrast 8, testing H 4, revealed a significant positive effect for participants in the repetitive-training condition compared to those who received training only once. This indicates that the increase in self-efficacy ratings at T4 compared to T0 was again stronger in the repetitive-training condition than in the one-time training condition.
Figure 6 illustrates the effects of training on self-efficacy over time, and reveals a clear trend. In the repetitive-training condition, self-efficacy gradually increased across the study period, suggesting a potential cumulative benefit of continued training. In contrast, participants in the no-training condition showed a steady decline in self-efficacy, indicating a loss of confidence over time. Average self-efficacy ratings (and standard errors) as a function of training and ToM
Hence, short-term effects showed that training had no effect on self-efficacy, with repetitive training causing a temporary decline in confidence compared to both one-time and no training. In the long term, repetitive training led to a significant and lasting improvement in self-efficacy, suggesting that continued exposure to training can have a cumulative positive effect on confidence.
Team membership once more proved to have little explanatory value; the random effect of Team was .00; the random effect of indidividuals within teams was .10, and the residual variance was .10.
Group Efficacy
Planned Contrasts Examining the Interaction of Training and ToM on Group Efficacy
Note. SE = Standard error
*p < .05; **p < .01
Long-term effects showed that repeated training led to a stronger and lasting improvement in group efficacy. Contrast 7 revealed a significant positive effect for participants who received repetitive training compared to those in the no-training condition. This suggests that group efficacy ratings increased more from T0 to T4 for those in the repetitive-training condition than in the no-training condition. Contrast 8 also revealed a significant difference between participants who received repetitive and one-time training, indicating a long-term advantage of repetitive training over one-time training. These findings support both H 3 and H 4. Figure 7 further supports this, showing a consistent increase in group efficacy for participants in the repetitive-training condition, while a decline was observed in the no-training condition. Average group efficacy ratings (and standard errors) as a function of training and ToM
Hence, in the short term, repetitive training had no impact on group efficacy, but long-term effects showed that repetitive training did lead to a significant, sustained improvement in group efficacy compared to no training and one-time training. The pattern observed here mirrors the attitude and self-efficacy findings, suggesting that while group confidence may dip immediately post-training, it can recover and even strengthen in the longer term.
The random effects variance of Team was .01; for individuals within Team this was .19. The residual variance was .13.
Psychological Safety
Short-term effects showed that training had a positive impact on psychological safety, independent of training frequency. The overall contrast testing H 1 showed that training had a significant positive impact on psychological safety, with trained participants reporting higher levels of psychological safety compared to untrained participants (see Table 6). Trained participants reported more frequently that their input was valued and that their group members were open to discussing controversial topics. There was no significant difference between the one-time and repetitive-training conditions, however, suggesting that repeated training did not affect psychological safety outcomes. Based on these findings H 1 must be accepted, and H 2 rejected.
Long-term effects indicated that psychological safety improved most when participants trained repeatedly. Contrast 7 revealed a significant positive effect, with participants in the repetitive-training condition showing a stronger increase in psychological safety from T0 to T4 than those in the no-training condition. However, no significant difference was found between the repetitive and one-time training conditions (Contrast 8), indicating that while repetition had a long-term benefit over no training, the additional benefit of multiple training sessions over a single session was limited. Based on these findings H 3 must be accepted, and H 4 rejected. See Figure 8 for a visual representation. Average psychological-safety ratings (and standard errors) as a function of Training and ToM
While Figure 8 shows similar levels of psychological safety across conditions at T4, the significant contrast reflects differences in the change over time: psychological safety increased more strongly from baseline in the repetitive-training condition than in the no-training condition.
Planned Contrasts Examining the Interaction of Training and ToM on Psychological Safety
Note. SE = Standard error
*p < .05; **p < .01
Variance in the ratings of psychological safety showed a very small effect of Team membership. I.e., analysis of random effects showed that the variance of Team was .00, while that of individuals within teams was .08, and residual variance was .17.
Intragroup Conflict
Planned Contrasts Examining the Interaction of Training and ToM on Intragroup Conflict
Note. SE = Standard error
*p < .05.
Long-term effects showed that these heightened perceptions of intragroup conflict were not sustained over time: Contrast 7, assessing the difference between participants who trained repetitively and those who did not train, revealed no difference, and neither did Contrast 8, testing the difference between the repetitive and one-time training conditions.
Jointly, these findings suggest that training increases intragroup conflict about ethnic profiling. However, the intensity of these discussions seem less in subsequent sessions. See Figure 9 for a visual representation. Average intragroup conflict ratings (and standard errors) as a function of Training and ToM
Variance in intragroup conflict of Team (.00) was again minimal, compared that of individuals within teams (.15), and residual variance (.11).
In Sum
The findings of this study indicate that training had a significant impact on both individual and group level outcomes, though its effects varied depending on both the timing and repetitively of the training. One-time training was sufficient to improve knowledge and psychological safety. However, it was only through repetitive training that sustained long-term changes were observed, particularly in attitudes, self-efficacy, and group efficacy. While repetitive training showed lasting benefits, especially in knowledge retention and self-efficacy, it also led to a temporary decline in self-efficacy and group efficacy immediately after training. This decline coincided with an increase in conflict, suggesting that more intragroup conflict caused less confidence among participants to translate the knowledge to practice to prevent ethnic profiling. This suggests that sometimes things get worse before they get better. Overall, the study highlights that while repetitive training offers long-term advantages, its short-term effects can vary depending on the specific outcome being measured.
Overview of Results, Showing Contrasts in Hypothesised and Opposing Direction (Shaded in Medium and Dark Gray, Respectively)
Random-effects analyses consistently showed that most outcome variance was attributable to individual-level differences rather than team-level dynamics, suggesting that training effects operated predominantly at the individual rather than the collective team level.
Discussion
Ethnic profiling is a harmful form of discrimination that not only negatively impacts individuals and communities but also erodes trust in law enforcement, undermines the legitimacy of police institutions, and challenges democratic principles by weakening the ideal of equal citizenship and the state’s responsibility to ensure justice for all (Henry, 2017; Schilliger, 2020; Tyler & Wakslak, 2004). Given the urgency of preventing ethnic profiling, the present study examines the short- and long-term effects of anti-ethnic profiling training on police officers, with a focus on various psychological outcomes. The findings show mixed results in relation to the study’s hypotheses (see Table 8), showing significant improvements in some areas while revealing more complex dynamics in others.
The Effect of the Training
Short-Term Effects
Consistent with Hypothesis 1, the training had some immediate positive outcomes, particularly in improving officers’ knowledge about lawful police stops. Trained participants reported more awareness of the CoP, while also showing a clearer understanding of the legitimate stop criteria. This is consistent with preliminary findings from De Vries et al. (2023). Psychological safety also improved after training, with officers feeling more able to express concerns and engage in open dialogue, and these outcomes sustained in the long run for both one-time and Repetitve training. These outcomes suggest that training can help address key barriers to behavioural change, such as lack of knowledge (“don’t know”) and fear of speaking out (“don’t dare”).
Short-term effects on attitude, self-efficacy, and group efficacy were more mixed, particularly in relation to Hypothesis 2. While one-time training led to modest improvements in attitudes, repeated training unexpectedly resulted in more negative attitudes immediately after the sessions. Although the reasons behind this attitude drop were not analysed in a structured manner, trainers reported that some officers interpreted the announcement of three training sessions as a sign of mistrust, suggesting they were heavily prejudiced. This perception may have provoked defensive responses and resistance. Such reactions are in line with findings by Murrar and Brauer (2019), who highlight that top-down reform initiatives can be perceived as accusatory and may trigger defensive attitudes. Similarly, self-efficacy and group efficacy temporarily declined after repeated training, likely due to the discomfort of confronting sensitive topics such as ethnic profiling and the difficult conversations these sessions provoked. Together, these findings indicate that repetition does not always enhance immediate outcomes and may sometimes temporarily undermine them.
Long-Term Effects
In contrast, the long-term effects of repetitive training were more consistent and clearly beneficial, supporting Hypotheses 3 and 4. Officers who participated in repeated sessions demonstrated more durable improvements in knowledge, attitudes, self-efficacy, group efficacy. These participants retained a more accurate understanding of lawful stop procedures, expressed greater confidence in acting ethically, and reported stronger beliefs in their team’s collective capabilities. In contrast, the effects of one-time training tended to fade over time. In some areas – such as CoP awareness – control group participants eventually caught up, likely due to other reform implementation initiatives. It is important to keep in mind that the training evaluated here is not the only initiative (yet the most immersive) inside the Dutch police to prevent ethnic profiling.
Importantly, the initial negative effects on attitudes and self-efficacy observed after repeated training rebounded over time, suggesting that early discomfort gave way to deeper reflection and meaningful behavioural change in the 6 months period between T3 and T4. Group dynamics likely played a key role in this process. As participants continued to interact, they accumulated new shared experiences in their day-to-day work, which offered opportunities to revisit and apply the discussions from training in a practical context. These ongoing interactions may have reinforced learning and supported the gradual development of more constructive attitudes and a stronger sense of group efficacy.
Hence, these findings underscore that repetitive training, while potentially unsettling in the short term, ultimately enables officers to process challenging insights, rebuild confidence, and adopt more constructive attitudes over time.
Variation in Baseline Attitudes Across Teams
It is important to acknowledge that officers in the repetitive-training condition consistently scored lower on many attitudinal and efficacy measures, even at the start of the study. This may reflect contextual differences between police stations rather than the training itself. As noted in the Methods, teams in highly diverse cities may be more likely to encounter allegations of ethnic profiling or broader concerns related to racial bias. It is possible that such issues were more salient for some teams in the repetitive-training condition than for teams in other conditions. Additionally, we cannot rule out that specific incidents at certain police stations, comparable to what occurred at Team B, may have influenced officers’ attitudes prior to the start of the study. While these baseline differences were not quantitatively captured, they did not pose a problem for our analyses: the linear mixed models account for variability between teams and individuals. Nevertheless, this contextual variation highlights that starting attitudes can differ across police stations due to prior experiences, local culture, and perceived relevance of the training topic.
Implications
Overcoming Barriers to Lasting Change
Resistance is widely recognised as a barrier to police reform, yet it has received limited scholarly attention (Cooper, 2020; Landman & Sollie, 2018; Shiner, 2010). As a result, training programmes frequently struggle to secure meaningful officer participation, particularly when dealing with sensitive and polarised topics such as ethnic profiling. This study contributes valuable empirical evidence to existing research on ethnic profiling reform participation by providing evidence that barriers (e.g., don’t know, don’t care, don’t want, don’t dare) can be addressed. While previous studies have documented persistent challenges in altering police attitudes and behaviour (e.g., Lai & Lisnek, 2023; Worden et al., 2024), this study shows that even initial resistance can transform into longer term effects.
Importantly, improvements in attitude, knowledge, self- and group efficacy, and psychological safety serve both as drivers of change and as indicators of reform progress. A more constructive attitude signals a shift from defensiveness to engagement, reflecting growing acceptance of ethnic profiling reform as legitimate and relevant (Dunbar & Hanink, 2023). Increased knowledge enables officers to conduct stops in alignment with legal standards and organisational policies, reducing reliance on stereotypes. Higher self-efficacy empowers officers to apply this knowledge in practice, while group efficacy strengthens the belief that the team can collectively prevent ethnic profiling (Bandura, 1977; Malone, 2001). This fosters mutual accountability and encourages officers to reflect on their own actions and discuss negative behaviours. When supported by increased psychological safety, officers feel more comfortable openly discussing mistakes, sharing experiences, and learning from one another without fear of judgement. Together, these factors foster individual learning and contribute to a broader cultural shift toward ethical, reflective policing.
The Importance of Repetition and Reflection
These findings suggest that repeated training may initially provoke resistance or self-doubt, but ultimately enables officers to process difficult insights, rebuild confidence, and internalize more constructive attitudes. The finding that intragroup conflict on ethnic profiling issues increased after training, but returned to the pretraining level after six months also suggests that the group has reached a new shared understanding on the topic. This aligns with other research showing the importance of task-related conflict for innovation processes (De Dreu, 2006; Giebels et al., 2016), particularly when it stimulates critical thinking (e.g., Jehn, 1995; Tjosvold, 2008).
However, we do not know whether this effect lasts in the long run, for example, when group composition changes because new colleagues enter or leave the team. It might then be necessary to repeat the training. However, a 6-month follow-up is relatively long compared to most studies (e.g., O’Neill et al., 2019), which is indicative of a sustainable effect. Given that social safety also improved in the long run, indicating the team culture changed in a positive way, newcomers might be successfully socialized, and leavers may further extend positive effects. Regardless, follow up research would be needed to see whether the effects would still be visible after e.g. one or two years. Especially in a dynamic context like police work, sustained impact cannot be taken for granted.
It also remains unclear whether repeating training once instead of twice is already sufficient to resort long-term effects. While repeated sessions still had a lasting impact after 6 months, future research should explore whether fewer repetitions could achieve similar outcomes, or, alternatively, whether more sessions produce even stronger effects. It is nevertheless to be expected that trainings need a form of consolidation to preserve effects, especially in light of staff turnover or new societal tensions, e.g. in the form of annual maintenance training. Given the costly nature of such initiatives and a higher risk of dropouts, this is also something to consider in follow-up research.
Policy Implications for Police Training Programmes
The present findings have several direct implications for the design and implementation of anti-ethnic profiling training in police organizations. First, the results suggest that one-off training sessions are insufficient to produce durable change in attitudes, self-efficacy, and group efficacy. While one-time training can improve knowledge and psychological safety in the short term, only repetitive training led to sustained improvements over time. Training programmes should therefore be designed as multi-session trajectories rather than isolated interventions.
Second, the short-term declines in attitudes and efficacy observed after repeated training indicate that early resistance or discomfort should not be interpreted as programme failure. On the contrary, such reactions may signal meaningful engagement with challenging content. Training programmes should explicitly anticipate and normalize this temporary discomfort, for example by communicating upfront that resistance and doubt are part of the learning process.
Third, the findings highlight the importance of structured reflection and group dialogue alongside immersive training methods. Increases in intragroup conflict immediately after training suggest that training activates critical discussion, which can be productive when supported by psychological safety and skilled facilitation. Police organizations should therefore invest not only in training content, but also in facilitation capacity and follow-up conversations within teams.
Finally, the observed flattening of psychological safety between T3 and T4 suggests that gains may erode when training and facilitated reflection cease. This points to the importance of maintenance mechanisms, such as periodic refresher sessions or integration of training principles into routine team meetings. Together, these implications suggest that effective anti-ethnic profiling training requires sustained investment, repetition, and organizational support rather than short-term, symbolic interventions.
Limitations
This study comes with several limitations. Longitudinal and experimental field studies are often affected by challenges such as participant dropouts and measurement issues, all of which can impact the generalisability of the results. Nevertheless, we aimed to minimise such concerns by employing robust statistical methods and conducting careful analysis.
Measurement Issues
This study relied on self-reported data. While self-reports are useful for capturing perceptions, they can be influenced by social-desirability bias (Van de Mortel, 2008) in that participants may have sought to present themselves in a favourable light. For example, items like “I am able to discuss the quality of police stops with my direct colleagues” might prompt inflated responses, especially early in the study. Yet, since all items were administered consistently across conditions and time points, any bias is likely to have been evenly distributed. However, future studies could benefit from incorporating behavioural measures, for example audio recording systems equipped with software capable of detecting turn-taking patterns. In the current study, we intentionally chose not to implement this technology, anticipating that it might provoke resistance or discomfort among participating police officers. Future studies could also examine whether trained officers actually make different choices in practice. For instance, researchers could investigate whether there is a measurable decline in citizen complaints related to alleged ethnic profiling for the teams that went through the training programme. Such outcome-based indicators would help assess whether changes in perceived efficacy translate into meaningful behavioural change.
Another measurement issue that is typical for field research is the tension between the number of constructs and the number of items for each that can be included. While measures with limited items are then often unavoidable from the viewpoint of participant burden and response rates, their use may also introduce limitations. In particular, single items are often argued to be more susceptible to measurement error than multi-item constructs, and hence are seen as less reliable (Hinkin, 1995).
However, research increasingly questions the problem of single-item measures. For example, research shows that shortened scales can yield results comparable to multi-item scales (Cheung & Lucas, 2014; Lucas & Donnellan, 2012). This is also supported by a recent systematic review (Allen et al., 2022). However, some single-item scales might have come with limitations. For example, while the single item measure for intragroup conflict allowed us to efficiently include the construct within a broader survey, it may have limited our ability to capture the full complexity of it, including aspects such as emotional tension, communication breakdowns, or deeper value-based disagreements. Future studies might orient towards using multi-dimensional item measures to more comprehensively assess such key constructs.
Attrition and Selection Effects
This longitudinal field study was characterised by substantial attrition, which warrants careful consideration. Although the distributions of gender, ethnicity, or years of service showed no substantial differences between original sample of police officers and those who remained at the final measurement, the possibility of selective attrition on unobserved characteristics cannot be excluded. For example, officers who experienced higher workload, stronger resistance to the training’s topic, or lower motivation to engage with the intervention may have been more likely to discontinue participation.
Such selective dropout could have influenced the observed effects, particularly for attitudinal and efficacy-related outcomes that are closely tied to motivation and engagement. While the use of linear mixed models allowed us to retain partially observed cases and to model change over time, these techniques cannot fully compensate for attrition that is systematically related to the intervention or its perceived relevance.
However, we would like to stress that the key results concern within-person changes over time and differential trajectories across conditions, rather than cross-sectional differences at a single time point. Moreover, the convergence of findings across multiple related outcomes, and the theoretically coherent pattern in which short-term declines were followed by longer-term improvements in the repetitive-training condition, suggest that the results reflect substantive change processes rather than artefacts of dropout alone.
Nevertheless, future studies would benefit from strategies to reduce attrition and to more directly assess reasons for dropout, for example through brief exit surveys or administrative data on workload and deployment. This would allow for a more precise assessment of how participation dynamics interact with training effects in operational police settings.
To Conclude
Many training programmes to reduce ethic profiling meet resistance and their effectiveness remains uncertain. Our study shows that an immersive and scientifically grounded training session can resort positive effects on both the individual and group level. Importantly, repeated sessions are necessary to gain sustainable organizational level effects. At the same time, it is important to realize that repeated training implies that sometimes things get worse before they get better. As such, early discomfort can signal the start of profound change.
Footnotes
Acknowledgement
We extend our sincere thanks to all station chiefs who facilitated this work within their units – particularly Ruben Boomsma, Koen Cammeraat, Pepeijn Bosboom, Ab van Vliet, Manfred Schipper, Gideon Zandstra, and Frank van Leeuwen – for their warm and committed involvement. We are also grateful to Justin Karten, Erik-Jan Bijvank, and the development team at Scopic Labs for their assistance in building the VR application and data collection interface. Special thanks go to our dedicated trainers, whose work in moderating difficult and often sensitive discussions was central to the success of this project. We especially acknowledge the contributions of Mattijs Oskam, Ruben Lunter, Richard van der Tillaard, Patrick van der Hout, René Vos, Ans Vermeulen, Mirjam Berends-Timmermans, Frank Arendsen, Eric van der Lingen, Theo Anrochte, Liane van de Velde, Bart Mijnster, Martijn van Heiningen, Jan Eigenraam, Dominic Heeres, Sander de Ruiter, Izequiel Eugenio, Leo de Haan, Patrick Kleijer, Patrick Fluyt, Angeline Driesprong, Tom Nauta, Koen Keijzers, Rutger Paters, Rik van Essen, and Vera Bout. Finally, we thank Jacqueline Evans for her careful and thoughtful academic editing.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The work was supported (in kind) by the Dutch National Police and The Police Academy of the Netherlands.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.