This study used nonprobabilistic sampling to recruit 39 male AOs aged between 14 and 18 years. The participants were residents of a reform center for young male offenders in Barranquilla, Colombia, where they fulfilled sentences for various punishable crimes such as sexual abuse, intentional homicide, aggravated crimes against freedom, integrity, theft, and drug possession. Furthermore, the study recruited 39 nonoffender adolescents from schools in the same city as control. The inclusion criteria for this group were ages ranging from 14 to 18 years, being male, level of education of less than 15 years of education, and not having committed a crime. Both groups were free from psychiatric or neurological diseases and were not under pharmacological treatment during the assessment. Although the majority of AOs displayed histories of drug and/or alcohol use, none were diagnosed with addiction or treated for this reason. They were given an informational form that describes the study and asked to sign an assent along with parental consent. The Ethics Committee of the Autonomous University of the Caribbean reviewed and approved the study.

Both groups, as well as their parents, signed informed assent following the guidelines of the Helsinki Declaration. The application of the different instruments were conducted at the reform center for AOs and completed at schools for nonoffender adolescents. Agreement was reached on the schedule with the directors of both centers to avoid interrupting various academic and recreational activities. The participants first performed the emotional flanker task followed by the emotional screening task. Both tasks were performed in one session, and the INECO and MOCA were completed in a second session to avoid fatigue. An important aspect to mention is that, as the members of each group were selected, the following sociodemographic data were recorded: age, years of education, and socioeconomic status. For the AOs sample, these data were obtained from admission and follow-up records from the center in which they are institutionalized. Information corresponding to the control group was derived from the demographic data section of the informed consent form.

The study used the Statistical Package for the Social Sciences version 26.0. Sociodemographic data, IFS, MOCA, and responses to the emotional screening test were compared between groups using Student’s t-test for independent means. Data from the emotional flanker task were analyzed using a general linear mixed model (ANOVA) that included two within-subject (factors and emotion; neutral and threat) and position (central and peripheral). The between-subject factor was group (control and offenders). This scheme led to a 2 × 2 × 2 design comprising four repeated measures (i.e., target neutral/flanker object, target object/flanker neutral, target threat/flanker object, and target object/flanker threat) and the two groups. We ran two mixed models: one for accuracy, and the other for RT. The significance threshold was set to alpha = 0.05 for all analyses.

In the present sample, AOs produced lower scores on MOCA compared with nonoffenders (although the scores in general were high and unlikely to indicate significant cognitive impairment). Notably, such differences persisted after omitting group differences in education. A specific assessment of frontal lobe functions using the IFS revealed that AOs presented with executive function impairments. Scores below 25 were reported as suitable for the detection of executive dysfunction in patients who present alterations related to the frontostriatal circuits; scores equal or close to 30 indicate preserved executive functions (Torralva et al., 2009). Gonzalez-Gadea et al. (2014) previously reported executive function impairments in AOs from similar populations (for a meta-analysis, see Gil-Fenoy et al., 2018 and Jiang et al., 2020). Such cognitive disorders share biological underpinnings with the behavioral problems displayed by these young individuals in which abnormalities in the maturation and expression of the frontal lobe seem to account for both (Borrani et al., 2015; Borrás et al., 2017). A potential explanation for these results is that chronic exposure to negative environments (e.g., violent, drug use, and childhood abuse) may impede the normal developmental trajectories of frontal lobe functions, which leads to maladaptive behaviors. For instance, previous studies demonstrated that institutionalized adolescents raised under social deprivation conditions displayed delayed development in relation to theory of mind (ToM) relative to adolescents raised with their biological families (Baez et al., 2018). This finding implied that the development of moral cognition is vulnerable to the impact of institutionalization. Taken together, previous studies and the current results reveal that the discrepancies observed between the groups may be associated with potential abnormalities in the frontal lobe. However, further research is required to identify whether or not these anomalies are a result of inadequate parenting and/or adverse rearing conditions, which would then increase the risks of maladaptive behaviors.

Based on previous studies (Baez et al., 2018), we expected that AOs would be less capable of identifying emotional valences from IAPS images than would adolescent nonoffenders. However, this prediction was not the case. Not only were AOs equally accurate compared with nonoffenders in detecting emotional valence, but they were also faster. A possible explanation for such a departure from our predictions remains unclear. One possibility is the task themselves. Other studies (Gonzalez-Gadea et al., 2014) that reported impaired social cognition (emotion recognition and ToM) in AOs used tasks in which participants conduct an emotional assessment of presented stimuli (e.g., positive, negative, angry, and happy). The emotional flanker task requires participants to decide if the IAPS images presented are violent (i.e., threatening condition) or neutral. Deciding whether or not an event is violent or if it conveys a particular type of emotion may rely on different underlying mechanisms.

Evidence suggests that offenders regardless of age display attentional bias to violence-related and negative stimuli (Mellentin et al., 2015). Psychopathic traits only seemingly, partially explain such a bias; however, negative childhood experiences appears to account for it. Domes et al. (2013) cited that enhanced attention to violence-related stimuli in criminal offenders is associated with adverse developmental experiences and delinquency. Moreover, they put forward that this is carried into adulthood (Domes et al., 2013; Mellentin et al., 2015). Kimonis et al. (2008) observed that youth with high callous–unemotional traits and enhanced orientation toward distressing stimuli exhibited stronger histories of abuse. This result reinforced the notion that environmental factors seemingly influence the development of these traits. Such a bias appears to be very pronounced that even when presented with ambiguous situations, boys with conduct problems tend to interpret these stimuli as hostile regardless of levels of callous–unemotional traits (Hartmann et al., 2020). The latter author suggests that the interaction of attentional and attributional biases in children with conduct problems may contribute to increased aggressive behaviors. With respect to the previous point and in view of future research, exploring the attentional bias of AOs according to low or high levels of callous–unemotional traits is important. The reason is that they, as well as young people with behavioral problems, will be a very heterogeneous group and, therefore, would present diverse patterns of behavioral (reactive or proactive aggression) and neural responses to affective stimuli (Viding et al., 2012).

Finally, findings are in agreement with the literature. Classical attention bias to threat was evident not only in relation to RT (which is the variable revealing these effects) but also in the measurement of accuracy in target detection. Figure 2 depicts that AOs, relative to nonoffenders, were less accurate in identifying object targets when these were flanked by threating compared with neutral stimuli. This effect approached Bonferroni’s corrected threshold (p = 0.012). None of previous studies using the emotional flanker task has observed such an attention bias to threat in relation to accuracy (Tipples and Sharma, 1953; Fox and Damjanovic, 2006; Lakens et al., 2013; Parra et al., 2018; Trujillo et al., 2021). RT is known to be a more sensitive variable for revealing the subtle effects of emotion processing (normal or altered) on attention (Yiend and Mathews, 2005; Yiend, 2010; Petersen and Posner, 2012; Harrison et al., 2015). Such a reduced inhibition of task-irrelevant emotional information has been previously reported in batteries using an emotional Stroop task (Chan et al., 2010).

One may argued that the long-term exposure employed in the task should have enabled the endogenous processing of flankers. However, Parra et al. (2018) and Trujillo et al. (2021) previously illustrated that emotional bias to threat is observed when exposure time ranges from subliminal (200 ms) to supraliminal (>1,000 ms). This finding has led to the proposal of the adaptive nature of this response, which seemingly plays a key role in survival. The results presented further support for this notion, as if longer exposure time would have encouraged the endogenous processing of flanker images. Moreover, the current study should have observed changes in AOs similar to those observed for response-relevant (targets) threat-images. However, the opposite was true. On the contrary, threat-related bias remained uninfluenced. Given the previous studies and the evidence presented, the researcher feel confident in suggesting that the processing of endogenous stimuli of flankers should be omitted as drivers of the reported effects.

To the best of our knowledge, this study is the first to report on the presence of attention bias to threat for endogenous and exogenous stimuli in AOs. This finding calls for additional research on attention in these young people and raises further questions. For instance, are the perceptions of violence as a dichotomous decision and emotional valences dissociable in offenders? Which type of information would carry more value in mitigating the impact of chronic exposure to negative experiences via social cognition interventions? Could improving emotion recognition reduce the perception and perpetration of violence in such individuals (Tipples and Sharma, 1953; Hansen and Hansen, 1988; Desimone and Duncan, 1995; Psychology Software Tools Inc, 1996; Lang et al., 1999; Szekely et al., 2004; Yiend and Mathews, 2005; Calvo et al., 2006; Fox and Damjanovic, 2006; Horstmann et al., 2006; Fox et al., 2007; Frewen et al., 2008; Kimonis et al., 2008; Torralva et al., 2009; Chan et al., 2010; Yiend, 2010; Baluch and Itti, 2011; Petersen and Posner, 2012; Viding et al., 2012; Domes et al., 2013; Lakens et al., 2013; Zhou and Liu, 2013; Bowen et al., 2014; Carretié, 2014; Gonzalez-Gadea et al., 2014; Katsuki and Constantinidis, 2014; Borrani et al., 2015; Harrison et al., 2015; Mellentin et al., 2015; Tobón et al., 2015; Pool et al., 2016; Borrás et al., 2017; Pike et al., 2017; Trujillo et al., 2017; Baez et al., 2018; Carson et al., 2018; D’Hondt et al., 2018; Gil-Fenoy et al., 2018; Parra et al., 2018; Subra et al., 2018; Pino et al., 2019; Hartmann et al., 2020; Jiang et al., 2020; Nguyen et al., 2020; Trujillo et al., 2021)?

We did not observe differences between AOs and nonoffenders in the ratings of the IAPS images after the emotional flanker task. As previously described, the participants rated the images in the task as violent or neutral using a Likert-type scale. Future studies could use a similar task to compare ratings on emotional valences (e.g., negative, unpleasant, and excited) and investigate whether or not they differ from violence/threatening ratings. This aspect is important, because we identified that low levels of education or general cognitive abilities (i.e., MOCA) do not account for such heightened responses in AOs (faster responses to IAPS images that reveal the target and flanker effects with the latter also appearing for accuracy). However, frontal lobe functions (i.e., executive functions as assessed by the IFS) seemingly accounted for these results. This study is not the first to report poor executive functions in AOs. Gonzalez-Gadea et al. (2014) also observed poor executive functions in AOs. However, their study did not observe associations between impairments and difficulties with emotion recognition and cognitive empathy.

The fact that we found such an association reinforces the notion that emotion, ToM, and perception of violence may be reliant on different neurocognitive mechanisms. Where the former two could more heavily rely on the anatomo-functional integrity of limbic structures, the latter may be more reliant on the influence of the cortical components of attention–action networks (Bick and Nelson, 2016). Recently, using ERP analysis, Trujillo et al. (2021) found that the target effect reported with the emotion flanker task is largely driven by the frontocentral components of the attention network involved in top-down regulation. Alternatively, threat-related bias is regulated by earlier centroparietal activity that supports top-down or bottom-up attention mechanisms according to task demands (Petersen and Posner, 2012). The findings presented demonstrated the need for further research to better characterize the biopsychosocial drivers of conduct disorders in male AOs. Such knowledge will exert a significant impact on intervention programs designed to prevent or reverse maladaptive behaviors in children and adolescent at risk.

This study has its limitations. First, we worked with a modest sample size of 100% male subjects. However, the number of participants in the study was similar to that reported in previous research (Tipples and Sharma, 1953; Hansen and Hansen, 1988; Desimone and Duncan, 1995; Psychology Software Tools Inc, 1996; Lang et al., 1999; Szekely et al., 2004; Yiend and Mathews, 2005; Calvo et al., 2006; Fox and Damjanovic, 2006; Horstmann et al., 2006; Fox et al., 2007; Frewen et al., 2008; Torralva et al., 2009; Yiend, 2010; Baluch and Itti, 2011; Petersen and Posner, 2012; Domes et al., 2013; Lakens et al., 2013; Zhou and Liu, 2013; Carretié, 2014; Gonzalez-Gadea et al., 2014; Katsuki and Constantinidis, 2014; Borrani et al., 2015; Harrison et al., 2015; Mellentin et al., 2015; Pool et al., 2016; Borrás et al., 2017; Pike et al., 2017; Baez et al., 2018; Carson et al., 2018; D’Hondt et al., 2018; Gil-Fenoy et al., 2018; Parra et al., 2018; Subra et al., 2018; Jiang et al., 2020; Nguyen et al., 2020; Trujillo et al., 2021) on AOs. Thus, obtaining a larger sample that includes the female population would be pertinent. Second, the groups exhibited significant differences in age, level of education, and socioeconomic status, which were not controlled for during analyses for two reasons. First, data from the emotional flanker task did not comply with ANCOVA assumptions that in as much such covariates were strongly correlated with the grouping variable and only mildly correlated with the dependent variables. Nevertheless, such departures are interesting as Volavka noted in a model (Volavka, 1999). Education and socioeconomic status (e.g., rearing environment) are risk factors for antisocial or violent behaviors. Therefore, eliminating the variance accounted for by such factors can indeed omit the variance of interest. To avoid such methodological shortcomings, actions should be taken during recruitment in which groups are matched as close as possible according to these variables, moreover, samples should be sufficiently large to enable the categorization of demographic variables. Future studies should take the case of these suggestions to explore the extent to which such factors could be mediators of the effects reported here. Third, the present study used only psychometric measures and systematized activities to examine attentional bias. In the future, using neurophysiological and neuroimaging recording techniques will help researchers gain a better understanding of the modulations and neurobiological underpinnings of these cognitive and emotional information processing tasks in the reference population. Furthermore, extensive sociodemographic data should be considered, because the reported effects could be mediated by personal (e.g., educational attainment) and familiar (e.g., rearing environment) factors (Volavka, 1999). The reported data should also be considered together with affective status and individual dispositions (e.g., motivational, trait-anxiety, and negative emotions), which are known to be common comorbidities of these adolescents and to affect social cognition (Ryan and Redding, 2004).