Exploring predictive disciplinary exposure in police recruitment: A machine learning approach

Introduction

The overlap between initial motivations for joining law enforcement and subsequent conduct within the profession has long been a subject of academic and practical interest (Cox, 2011; Raganella and White, 2004; White et al., 2010). Understanding what drives individuals to pursue careers in law enforcement and how these motivations correlate with their professional behaviour has implications for designing and implementing recruitment strategies (Linos, 2018), training programmes and supporting informed policy development. Recent studies have highlighted the complexity of motivational factors, ranging from altruistic desires to serve the community [related to duty (Cox, 2011)] to more pragmatic considerations such as job security and benefits (Wilson et al., 2010).

In parallel, the rise of machine learning and predictive analytics has opened new avenues for exploring such behavioural patterns and outcomes in various fields, including law enforcement (Gibson and Stubbs, 2023; Halford and Gibson, 2025). The application of these technologies has the potential to enhance our understanding of the relationship between motivational factors and potential exposure to disciplinary action among police officers (Cubitt et al., 2022). By leveraging the collation of disparate datasets, along with advanced analytical techniques, researchers can now identify potential predictors of conduct. These insights may offer valuable opportunities for pre-emptive interventions and the introduction of relevant organisational support mechanisms.

Although George Herbert Mead’s concepts of the ‘I’ and the ‘me’ are not widely used in policing studies of misconduct, we justify their application here as a theory that explicitly links individual dispositions (the ‘I’) and pre-entry socialisation (the ‘me’) to later behavioural outcomes. In this study the ‘I’ maps to personal motives (authority/status, excitement/variability, service, stability/economic) and the ‘me’ maps to social exposure (family member officer, friend/colleague officer, perceived media influence), which guided variable selection, feature engineering and interpretation.

This study examines the gap between the initial motivation to join the police, existing social connections and exposure to misconduct investigations. It employs a machine learning model to assess data-driven linkages between officers’ initial motivations and subsequent disciplinary records. In doing so, it builds upon foundational literature that examines the spectrum of motivations for embarking on a career in law enforcement. It aims to contribute to a nuanced understanding of the factors that influence police officer behaviour and evaluates the potential of predictive modelling to address disciplinary challenges within the force.

Research question. Among pre-entry factors, which specific motivations and pre-existing social ties, measured at entry on seven-point scales, are associated with later exposure (yes/no) to a misconduct investigation during service?

The findings of this research have implications for policymakers, policy-training academies and law enforcement governance agencies. It provides a data-informed rationale for enhancing recruitment processes, developing targeted training programmes and assisting in fostering an organisational culture that aligns with the ethical and professional standards of police officers. It also offers some promise for developing such models for use within law enforcement as a matter of course, foreshadowing viable future research threads. This study not only advances the academic discourse on police motivation and conduct, yet offers suggestions for practical tools that may improve the legitimacy and effectiveness of law enforcement institutions.

Literature review

The motivation to join the police is a complex area that has been the subject of sustained research. Studies have identified various factors that drive individuals to pursue careers in law enforcement. Some motives, potentially based on altruism such as the desire to help others or uphold the system of law, have been consistently discussed as significant reasons for entering law enforcement (Wallace, 2021; White et al., 2010). The specific nature of police work is also cited as a drawing influence, including the opportunity to serve the community (Wu et al., 2009) and maintain public safety (Wallace, 2021). In addition, authority, power, position (Elntib and Milincic, 2021), protection/guardianship (Koslicki, 2021) and more exciting variability and risk-taking (Gudjonsson and Adlam, 1983; Lorr and Strack, 1994) have all been identified as contributing factors on the decision to become an officer.

Although these variables suggest a complex tapestry of motivational factors, more tangible factors, such as individual characteristics of gender, sex, race and ethnicity, can also be influential. Previous studies have shown that women, black officers and Hispanic officers are motivated by career opportunities available in police organisations (Vermeer et al., 2020). A study of police cadets in China revealed that job security is an important motivating factor in joining law enforcement (Wu et al., 2009). This indicates that motivations can be context-specific, such as the general need for employment, which has been cited as a significant reason for individuals to pursue careers in law enforcement in regions where jobs are scarce (Wozniak, 2017).

Organisational factors and societal perceptions of the police can also affect individuals’ motivations to join a force. The Policing Vision 2025 and 2030 in the UK outlined long-term strategies to ensure consistency and accreditation in those joining policing careers, attempting to standardise the qualifications obtained by all new recruits (Watkinson-Miley et al., 2022). This introduced new potential variables for consideration, such as gaining transferable qualifications, further complicated by studies of perceptions of trust in certain communities (Vermeer et al., 2020), influencing individuals’ decisions to pursue a career in law enforcement. Isolating causal variables within such a milieu is challenging, and any study addressing motivation should consider a wide range of potential measurement avenues.

Considering this complexity, a review of the philosophy behind the related research may contribute to understanding how this subject has been investigated in the past. A wide range of psychology-based studies have been conducted in this area, focusing upon understanding how personality may have contributed to the choice to become an officer (Burbeck and Furnham, 1984; Carpenter and Raza, 1987; Evans et al., 1992; Harper et al., 1999; James et al., 1984; Lorr and Strack, 1994; Weiss et al., 1999), attempts to properly define a personality ‘type’ amongst police officers (Balch, 1972; Fenster and Locke, 1973; Garbarino et al., 2012; Gudjonsson and Adlam, 1983; Skolnick, 2010; Twersky-Glasner, 2005; Vastola, 1978) and how that personality may contribute to police officers’ performance over time (Adlam, 1982; Beutler et al., 1988; Cottle and Ford, 2000; Fabricatore et al., 1978; Gould, 2000; Griffin and Ruiz, 1999; Hogan, 1971; Richardson et al., 2007). These three areas are important, yet fundamentally rely on the exploration of latent constructs in the mind at the point of measurement. This, as Mead would suggest (1913), represents a distinct part of the psychologically centred ‘I’ and not the socially centred ‘me’.

Mead believes that the ‘I’ is an expression of one’s own identity, the identity that lies within the personal psyche and is expressed through character traits and tendencies. He also believes that together with the ‘I’ lies a socially contingent ‘me’, which encompasses how the ‘I’ is moderated and constructed through social interaction. In modern academic parlance, there is the largely statistical science of psychology, complemented by cognitive and neuroscience (in-depth exploration of the ‘I’) and the pursuance of social constructivism in sociology (Berger and Luckmann, 2016) and social psychology (explorations of the ‘me’) (Tajfel, 1982; Tyler et al., 2014). Unpicking these disparate and connected areas of literature is difficult and to establish convergence is even more challenging. It is fair to say however, that motivation as a product is born out of this convoluted mixture as a mix of personal traits and their expression and the influence of a police recruit’s existing and prior social network. In a police recruit, there will be both the ‘I’ and the ‘me’, making it reasonable to suggest that both of these areas are important for a more complete approach to the research of police recruitment.

Strangely, this work is supported by cutting-edge biological research. Sapolsky (2017, 2023) argues that choices made in everyday life are already determined by a combination of social opportunity or the lack thereof, development of the ‘me’ and biologically created resource, development of the ‘I’. His research on animal behaviour over lengthy periods has repeatedly reinforced this argument, and his latest book (2023) produces a great deal of evidence to support the presence of determinism in these choices. Within the canon of police recruitment literature however, there is a distinct lean towards the ‘I’ over the ‘me’. There is some well-established mixed-method research that examines the development of working identity as recruits join the police academy (Bennett, 1984; Chan, 2001; Charman, 2017; Fielding, 1988; Van Maanen, 1973, 1975), yet very little that looks at the social ‘me’ prior to becoming a police officer (Gibson and Stubbs, 2023; Stubbs et al., 2023). This social ‘me’ will form a part of the whole that contributes to an individual’s police service over time and may therefore offer some explanatory reason for police performance, just as the psychology literature above contributed to predicting patterns in this area.

With this discussion in mind, it is worth asking which areas of police performance can be investigated. The Casey Review (2023) highlighted many issues with police performance in the Metropolitan Police in England and Wales. One of the areas of focus is the poor behaviour of existing police officers over time and the inability of the organisation to effectively address misconduct. Similarly, a report from the U.S. Department of Justice, Office of Community Oriented Policing Services (2015) offered several recommendations that encompassed the operation of transparent accountability procedures together with a review of current hiring practices. These reports were produced in the wake of public scandals that affected confidence and trust in the police in both countries.

Recent research has begun to connect these two areas. Cubitt et al. (2022) explored how gender influenced the prediction of serious police misconduct. This large-scale study in New York used big data and machine-learning methods. These findings illustrate that gender is a predictive variable for serious misconduct, leading to several suggestions on how policing could take advantage of this finding. This study covers a range of data incorporating aspects of the ‘I’ and the ‘me’, but it focuses on officers who had already been recruited and not on potential recruits. The research output may be viewed as highly useful, however the real opportunity for prevention may lie at the point of recruitment. It is more useful to prevent those with a propensity for misconduct from becoming police officers, rather than recruiting, training and deploying these officers, only to find out that they are unfit to wear the badge.

Operational framing (Mead-aligned). Which measurable aspects of the ‘I’ and the ‘me’ predict exposure to misconduct investigations during service?

This study explores this question using types of motivation as independent variables and exposure to misconduct investigations as the dependent variable over the period of an officer’s police service.

Method

Discussion

Any machine learning model is constrained by the data used to create it. In this study the dataset is exploratory and requires refinement and further development. We measured a defined portion of the social ‘me’ to assess whether it helps predict exposure to misconduct investigation. The model showed high apparent accuracy, with feature engineering contributing substantially. Because the topic is important and in the public interest, improved data access is essential, although difficult to secure given the long-noted ‘dialogue of the deaf’ between academia and police (Easton and De Vlieger, 2018; Engel and Whalen, 2010), even if there are contested signs of progress.

Some feature combinations show promise and warrant closer examination. For example, joining the police for economic reasons in combination with having a family member in the police emerged as a strong engineered predictor, as did the pairing of authority with excitement. The latter could plausibly relate to a propensity for involvement in incident responses that are both high-arousal and authority-expressive. Since these combinations were influential in the model, it is reasonable to ask whether any are explored explicitly at the point of recruitment.

The results rest on socially situated reasons and motivations which differ from much of the psychological literature used at recruitment. They point to particular forms of the social ‘me’ that may be associated with later exposure to misconduct investigation. If developed further, this line of work could inform recruitment design by providing a rationale for examining specific aspects of pre-entry socialisation in a structured and transparent way. Where particular social conditions and motivations are present for a potential recruit, they may justify additional supportive assessment and oversight during training and probation. Current evidence does not justify automated screening or exclusion of applicants on this basis, but it does justify expanding the method and setting a future research agenda. This study is hypothesis-generating and requires prospective replication, calibration assessment and external validation before any operational consideration.

Predictive modelling in recruitment is a high-stakes use case. If developed further, the model should be used only as an adjunct to structured assessment and professional judgement, not as a standalone decision tool. Prior to any operational deployment, external validation, calibration assessment and group fairness evaluation should be undertaken, with clear documentation of intended use, error costs and mitigation pathways for false positives and false negatives. Data handling should follow applicable ethical approvals and data protection requirements, with explicit safeguards against disparate impact and misuse.

Given the exploratory nature of the dataset, the reported accuracy cannot support broad generalisation at this stage, although the findings show promise. The New York big-data study (Cubitt et al., 2022) used administrative records and identified gender as a meaningful predictor. By contrast, the social ‘me’ variables in our study are not directly measured during recruitment, nor routinely recorded or retained by forces. They may surface indirectly in assessment centres or interviews (Hesketh and Stubbs, 2024a), yet are rarely collected in a way that supports research and may be constrained by data retention rules. This underscores the need to expand data collection with appropriate governance and transparency.

A further limitation is the heterogeneity in how respondents understand ‘misconduct’ investigation across jurisdictions. The Casey Review (2023) emphasises patterns of repeated exposure rather than a minor versus major distinction, which supports inclusive measurement that introduces variation. Some degree of self-selection may also exist among officer offenders, as in conventional crime (Roach and Pease, 2016). Future iterations should tighten terminology and harmonise outcome definitions across contexts.

Finally, psychological studies at recruitment are rarely retrospective. If more aspects of the ‘I’ can be assessed at career end, they may add predictive value over time. More and better data will improve prediction and help police services make evidence-informed recruitment decisions. The article therefore ends with recommendations for future research that prioritise prospective data collection, cross-jurisdictional harmonisation and rigorous validation.