Research into the process of understanding another person’s intentions or experiences as a facilitator of daily encounters can refer to related processes (Baron-Cohen et al., 2015; Launay et al., 2015). Cognitive empathy, perspective-taking, or ToM are terms used to describe these processes (Baron-Cohen et al., 2015; Launay et al., 2015). However, empathy has two dimensions, one cognitive and one affective (Decety and Jackson, 2006; Jolliffe and Farrington, 2006). The cognitive dimension involves understanding another person’s intentions or experiences and beliefs of others (Singer, 2006). The affective dimension includes the bodily sensation such as being able to mimic the expression of others’ emotional states (Decety and Jackson, 2006). While rapport has been suggested as an essential component of interpersonal relations (Jakobsen, 2021), it is not clear whether this involves cognitive or affective components.

This means that cognitive intellectual abilities involved in empathy may not be sufficient to explain why or how empathy operates. The number of studies focusing on the relation between empathy and cognitive ability is limited. An early study used the revised version of the Reading the Mind in the Eyes test (RMET), which measures how a test-taker can understand the mental state of another person (Baron-Cohen et al., 2001). The study reported no statistically significant correlation between the RMET and IQ (r = 0.09, p = 0.60). However, a later meta-analysis investigating how the RMET correlated to intelligence found a positive significant correlation (r = 0.24) with a robust effect size (Baker et al., 2014).

Launay et al. (2015) found that there was a positive association between higher forms of mentalizing and emotion recognition in the RMET (r = 0.21, p < 0.001, n = 279), but they did not find any relation between mentalizing scores and self-reported empathy, measured with the Empathy quotient (EQ) instrument (r = 0.022, p = 0.33, n = 282). Guo et al.’s (2019) study with undergraduate students (n = 518) reported a significant correlation between perspective-taking in the Interpersonal Reactivity Index (IRI) and fluid intelligence measured with Raven’s Standard Progressive Matrices, but the correlation was stronger between affective empathic concern and fluid intelligence (r = 0.17 vs. r = 0.27). In a multiple mediation analysis, they reported cognitive ability had a positive effect on perspective-taking (β = 0.22, SE = 0.03, p < 0.001) and also on empathic concern (β = 0.26, SE = 0.03, p < 0.001).

From this research, it would appear that the affective aspect of empathy is dependent on overall intelligence, but self-reported cognitive perspective-taking may be quite independent. However, further research with normal populations is needed to establish the nature of empathy outside clinical groups (Richell et al., 2003).

A related capacity to intelligence that may be related to empathy is executive functioning (EF). Yan et al. (2020) found a significant correlation between cognitive empathy and subcomponents of EF such as inhibitory control (r = 0.23, p < 0.001) and working memory (r = 0.20, p < 0.001) was found, while the affective dimension of empathy had a weaker relation. Another recent meta-analysis (Martingano and Konrath, 2022) found a positive association between the cognitive part of empathy and rational thinking as expected with both empirical data and meta-analysis.

Another issue related to the measurement of empathy is the difficulty individuals have in providing valid information on their empathic ability (Murphy and Lilienfeld, 2019). Murphy and Lilienfeld (2019) suggest that the weak relations between empathy-related self-reports and task performance might be a consequence of measures that do not have acceptable measurement quality. In line with these reflections Baker et al. (2014) argue that the correlation would probably be even stronger if tests measuring these types of constructs had levels of reliability that were as high as intelligence tests.

Empathy has been identified as an important ability when working as a police officer (Perez, 2010; Inzunza, 2015a). It is particularly relevant for professions where interaction with the public often involves difficult circumstances (Skogan, 2006; Rahr and Rice, 2015; Inzunza, 2015a). In the field, a police officer needs to understand how others perceive a complex critical situation, and at the same time have the capacity to regulate their own perceptions in order to act professionally. When developing models that include these types of processes they are often multi-dimensional models that include both a cognitive and an affective dimension (Davis, 1983; Inzunza, 2015b). The importance, and indeed the necessity, of including other elements such as regulating processes has also been discussed and incorporated in some of these models. This is especially relevant when applied to professions such as policing, where the complexity of some daily encounters may require robust self-regulatory processes (Inzunza, 2015a). In this context, the major goal is to end encounters without harm for any party. This requires the ability to predict how a situation may develop based on the various actors’ intentions in a social situation (Abanonu, 2018).

Policing, therefore, requires an adequate level of intelligence and resilience (Romosiou et al., 2019), as well as both cognitive and affective empathy. This perspective-taking and empathy capability is important for police, especially when conducting interrogations. There are several ways offenders act to make a story sound truthful (i.e., offender lie-telling strategies; Strömwall and Willén, 2011). Such strategies can involve impression management strategies or information management strategies, both of which require close attention from the police interrogator. Interrogators need to pay attention to minute details and use their verbal communication skills to retrieve information so that they can understand the intentions of the offender. Here the primary goal is to increase the probability of solving complex crime events. All parts are necessary when approaching a crime within the framework of the choices and decisions made by alleged offenders before, during, and after committing a specific type of crime (Leclerc, 2017).

Police need to demonstrate empathy also when working with crime victims (e.g., Maddox et al., 2011; Inzunza, 2022). This seems to be a somewhat challenging ability to learn for police students because the identity formation process that takes place within the profession struggles to treat vulnerable citizens appropriately (Charman, 2019).

A small-scale (n = 60) study with police students found no statistically significant relations when studying the total scores between a subjective self-report measure of empathy (i.e., an adapted version of the Empathy Assessment Index; EAI) and objective tasks aimed at measuring similar constructs (Inzunza et al., 2019). However, those self-reporting highly on perspective-taking also performed better on the objective tasks. Other studies have noted the importance of understanding the relation of subconstructs of empathy to other constructs.

The practical value of researching this area can be seen both in the context of selection and the education and training of future police officers. The construct of empathy has been identified as crucial for general interactions in terms of social skills but also when conducting training in specific units within police organizations such as de-escalating conflict situations (Abanonu, 2018). In a recent Danish study of how official political goals have been translated to measurable criteria, the construct of empathy played a crucial role (Bloksgaard and Prieur, 2021). The study concluded that, within Danish police, police officers need to express emotions in a controlled way so that these communicative skills counterbalance their authority and ability to act; indicating the importance of empathy in modern policing.

Training of police is an important tool to their development of de-escalating strategies. Performance associated with empathy (e.g., understanding the situation of the other) is at focus when implementing specific programs aimed at improving police officers’ interviews with victims (Darwinkel et al., 2013).

A more detailed understanding of how aspects of empathy relate to other cognitive abilities is important to the design of police education programs. Hence, this study focuses on empathy and its subconstructs, how they can be measured and how they relate to other selected measures within police education populations. Two hypotheses were investigated:

The participants in this study were police students enrolled in three of the Swedish national police programs during fall of 2017 (i.e., Umeå, n = 54; Södertörn, n = 48; and Linnaeus, n = 55). The approximate percentage of males was 72–73%, and their mean age in years was 25.7 (Umeå), 26.2 (Södertörn), and 25.4 (Linnaeus). Compared to the total group of police students in Sweden, this group of 157 students is relatively representative in terms of age and gender distribution (polistidningen.se). Through voluntary participation requirements, the obtained response rate varied by location (Umeå, 87%; Södertörn, 22%; and Linnaeus, 22%). The amount of police students differs between the three locations where Södertörn and Linnaeus have more students enrolled in their programs and it is more difficult to reach all students. Since the allocation of Swedish National Police students is centrally organized, the different police programs have a similar profile regarding individuals recruited, and they experience similar content throughout their training. It is noteworthy that Swedish police education is a two-year undergraduate university qualification plus 6 months on-the-job apprenticeship, rather than a brief sub-baccalaureate qualification as seen in some other jurisdictions. Of the 157 participants, complete data on all constructs ranged between 154 and 157, suggesting a very high rate of data recovery. Note that no pressure was put on the respondents to participate or to perform well on the different measures, meaning score means and relations might be different if the scores were used to select candidates for entry to the police.

A battery of instruments was deployed to measure subconstructs of empathy and cognitive ability. The three instruments used to measure empathy and ToM (i.e., Swe-EAI, Eyes test, and Swe-IMT) have all been used in the police context in previous studies (see Inzunza, 2015b; Inzunza et al., 2019). Intellectual ability was measured with the Raven APM Set II measures of fluid intelligence (also referred to as cognitive ability).

The police students that participated in this study were informed that their participation was voluntary in an oral introduction given at the police academies and also in a cover letter. The participants were assured that all data would be treated according to prevailing research practices regarding confidentiality and ethics. The same procedures were at work in each of the three universities where data were collected. The police students completed the battery of instruments as described earlier. Data were collected from groups at each academy, with the participants sitting at individual tables under invigilated examination conditions.

Consistent with the two-step protocol (Anderson and Gerbing, 1988), measurement characteristics of each instrument were determined before inter-construct modeling took place. A combination of confirmatory factor analysis (CFA) and scale reliability estimation was used to establish the best-fitting set of items for each scale. Once scales were validated, inter-relationships among scales were examined using multiple structural equation models. Tested models included (a) simple inter-correlated scales, and (b) structural equation models (SEM). Analyses were conducted with latent variable software Mplus version 8 (Muthén and Muthén, 1998-2017). STATA version 16.1 (2020) was also used for calculating Raykov’s factor reliability coefficient.

Swe-EAI. Because the Swe-EAI has four latent factors, CFA was used to evaluate its structure.

Swe-IMT, Eyes test, and Ravens. These scales use classical test theory scoring protocols, summing all items scored as correct according to the scoring key and thereafter calculating reliability coefficients. The test and sub-test scores are evaluated for reliability with scale reliability estimation, which allows calculation of the standard error of measurement (i.e., SEm = SD*√(1–α)).

Data preparation. Because the sample size is small relative to the total number of items (155,118), SEM analysis used parceled scale scores for the Swe-IMT intention and memory, Eyes test, and Ravens test to give a more appropriate cases-to-items ratio (155:24) (Costello and Osborne, 2005).

Model fit. The fit of CFA and SEM models was assessed with several fit indices using conventional thresholds (Browne and Cudeck, 1992; Hu and Bentler, 1999; Albright and Park, 2009). Because the Chi-square fit index is overly sensitive for large degrees of freedom (Wheaton et al., 1977), the ratio of χ²/df is accepted as fit evidence if it has a statistically non-significant p > 0.05. The comparative fit index (CFI) indicates acceptable fit if it is above 0.90 and good fit above 0.95. We also included the more stable gamma hat index, interpreted in the same way as the CFI (Fan and Sivo, 2007). Both the root mean square error of approximation (RMSEA) and standardized root mean residual (SRMR) indicate acceptable fit if they are below 0.08 and good fit when below 0.05 (Browne and Cudeck, 1992; Hu and Bentler, 1999; Albright and Park, 2009).

A four-factor structure with 20 items of the Swe-EAI showed acceptable fit to the data (χ²₍₁₆₄₎ = 262.30, χ²/df = 1.60, p = 0.21; CFI = 0.88; gamma hat index = 0.93, RMSEA = 0.070, 90% CI [0.056, 0.083]; SRMR = 0.08). The relatively low CFI value probably arises from model complexity (Fan and Sivo, 2007) and can be discounted in light of the more robust values for the more stable indices (i.e., gamma hat, SRMR). This administration of the Swe-EAI had comparable results to those seen in previous studies (Inzunza, 2015b; Inzunza et al., 2019).

Table 1 below shows descriptive scale statistics, as well as three estimates of reliability for the two empathy subconstructs Affective Response (AR) and Perspective-Taking (PT) showing similar information concerning scale characteristics.

The Swe-IMT worked as intended in its shorter version. The mean number of errors was almost twice as high in the items measuring intentions compared to items measuring memory, suggesting that the ToM component is a more demanding skill (Table 1). The correlation between the number of errors was similar to previous studies (Spearman’s ρ = 0.47, p < 0.01; Pearson r = 0.61, p < 0.01). The scale estimates of reliability were lower than normally expected, but high scale reliabilities may reflect item homogeneity (Cattell and Tsujioka, 1964), which is not the case for the IMT Intention and Memory scales that are more heterogeneous in content. These values are deemed sufficient for research purposes.

The Eyes test mean (Table 1) is within the mean ranges reported previously. The conventional estimates of scale reliability were low, but this is expected with binary-scored items. Nevertheless, the SEm was relatively small, indicating that score differences of ≥7 are statistically significant at a 95% confidence level.

The mean score for the odd Ravens items was slightly higher (d = 0.60, 90% CI = 0.27–0.92) than the mean score of the even items (Table 1). However, the estimates of reliability were robust, producing quite small SEm relative to means.

The factor inter-correlations for participants with complete data showed that few of the linear relations had statistical significance (Table 2). Of interest here is the weak positive relationship of the two EAI and two IMT scales to the Eyes test. In contrast, only the AR and IMT Intention scales had a similar strength of relationship with the Ravens scale. The within-construct correlations for the Swe-EAI were moderate and strong for the Swe-IMT scales.

The fully correlated model, with six items each for PT and AR, had poor fit to the data (χ²₍₉₃₎ = 178.34, χ²/df = 1.97, p < 0.01; CFI = 0.79; gamma hat = 0.93., RMSEA = 0.081, 90% CI [0.063, 0.098]; SRMR = 0.079). Trimming two items each from PT and AR for weak loadings generated a better fitting model (χ²₍₄₃₎ = 62.34, χ²/df = 1.82, p < 0.05; CFI = 0.933; gamma hat = 0.97, RMSEA = 0.058, 90% CI [0.024, 0.085]; SRMR = 0.055). Instead of treating each scale as an independent construct, the three measures reflecting processes within the cognitive dimension (i.e., Ravens, IMT Intention, and IMT Memory) were treated as indicators of a latent trait for cognitive processes measured as tasks. Given that the Eyes test is meant to reflect and thus indirectly measure the mental state of an individual, it was also positioned as an indicator of overall cognitive processes measured with cognitive tasks. Rather than simply correlate the scales, both PT and AR were situated as predictors of the three cognitive processes and the Eyes test to test the hypotheses. That model had good fit (χ²₍₅₁₎ = 78.35, χ²/df = 1.54, p = <0.05; CFI = 0.91; gamma hat = 0.97, RMSEA = 0.058, 90% CI [0.030, 0.083]; SRMR = 0.063.) and showed that PT had a modest standardized regression weight to the Eye test and AR had a weaker regression weight to Ravens (Figure 1).

Some of the limitations of this study stem from the small sample size of participants. It may be difficult to predict performance on different tasks. It would be valuable to conduct a similar study with a larger sample, including professionals from within the police organization, rather than just police students. Such a study might reveal different relations once police students gain experience in the field. More importantly, the current results have to be seen as exploratory and require replication to validate the observed relations. While the project team will be able to map cognitive and affective empathy to emotion recognition, intention detection, and working memory in future studies, obtaining information with fluid intelligence measures in similar contexts is more challenging. It may be possible to test these findings with a sample from the general population of university students but that is currently out of scope for the project. Furthermore, such a study would not provide robust information about the population of concern, that is the police.

Generalizability to other police educational organizations is limited by the fact of having collected only Swedish data; hence, replication studies in different international contexts are warranted. This study was conducted among police students, but findings of the structure of the EAI have been used in an evaluation with in-service police officers (Hansson et al., 2021). The current measures treat each of these interactive aspects of empathy and cognition separately. Perhaps, a dynamic integrated assessment, along the lines of the objective structured clinical examinations used in medical education (Sloan et al., 1995), would improve our understanding of how these constructs inter-relate.

Our conclusions are that abilities beyond fluid intelligence are needed to understand emotional states and the intentions of others. Policing involves navigating complex and highly charged situations, where culturally specific knowledge of how individuals communicate is also required (Günthner, 2007)—this is a matter of increasing importance in the rapidly changing European demographic context (Ilie, 2019). In policing situations, officers need to remember relevant event information, express empathy, and identify possible intentions from fleeting facial expressions; clearly, this is a complex and dynamic process, and police training needs to prepare recruits for it as fully as possible, including in terms of empathy. Our findings which also are in line with previous studies are inherently valuable since there is higher expectancy from society that the police will act empathically and treat citizens professionally: while good performance in contact situations can have a weak positive effect on police trustworthiness, poor performance has a strong negative effect, essentially meaning that trustworthiness is hard to achieve but easy to lose (Oliveira et al., 2020). Thus, this study adds to our understanding of how we might improve the empathy of police students as they prepare to enter a very demanding profession.