Decision-making under risk and theory of mind in adolescent offenders in provisional deprivation of liberty

Abstract Introduction Delinquent behaviors are risky behaviors that increase during puberty and reach their highest peak in late adolescence. It has been proposed that poor decision-making and theory of mind (ToM) are key cognitive processes implicated with delinquency during adolescence, affecting evaluation of risks and impairing appreciation of social norms. Nevertheless, it is not yet clear whether adolescent offenders who are subjected to provisional deprivation of liberty due to conflict with the law (adolescents in conflict with the law [ACL]) might, in fact, present a specific profile with regard to these cognitive processes. Objectives To assess deliberative decision-making and ToM among adolescents in conflict with the law and adolescents not in conflict with the law. Methods The sample comprised 62 participants: ACL (n = 29) and a control group (CG) (n = 33). ToM was assessed with the Reading the Mind in the Eyes Test (RMET) and decision-making was assessed with the Columbia Card Task (CCT). Substance use, callous-unemotional traits, childhood maltreatment, and intelligence quotient (IQ) were also assessed. Results ACL had more ToM errors for negative mental states in comparison to CG, but not for error rates concerning neutral and positive mental states. With regards to decision-making, our results suggest that ACL group members did not vary their behavior based on the available information and that the risk information had an opposite effect on the number of cards chosen (risk-taking behavior) when compared to CG. Conclusion These findings have important implications for development of interventions for these adolescents, suggesting that they tend to learn little from negative outcomes and have reduced capacity to process negative emotions.


Introduction
It has been exhaustively shown that risk-taking behaviors such as use of licit and illicit substances, unsafe sexual practices, aggressive behaviors, and risky recreational sports increase during puberty, reaching their highest peak in adolescence, and decrease along adulthood. 1 Nevertheless, for many individuals, these behaviors might not fade during development, but turn into recurrent delinquency that leads them into problems with the law. The number of child delinquents entering the juvenile justice system is a serious worldwide problem, raising fiery political debates and challenging researchers in the social sciences, law, and health. Recent data showed that there were more than 22,000 adolescents in detention in Brazil, with approximately 18,300 already sentenced, corresponding to an incidence of 8.8/100,000 per inhabitants. 2 Several social, economic, cultural and biological factors have been identified as risk factors for many emotional and behavioral problems during the transition from adolescence into adulthood, 3 increasing adolescents' vulnerability to delinquent behavior and infraction. Therefore, beyond social vulnerability, which covers, not exclusively, marginalization, structural disadvantage, economic inequality, and childhood maltreatment, 4 cognitive developmental factors related to decision-making and inadequate reason about our own thoughts, feelings, and intentions have been widely related to criminal behavior among teenagers. 5 In this regard, proponents of dual-processing models of information processing have paid particular attention to a recently proposed framework of criminal decisionmaking, explaining, at least in part, how decisions could be triggered by an imbalance between an emotional "hot" neural system and a deliberative "cool" cognitive neural system. 5 In a broad sense, all decisions we make result, at some level, from the individual's capacity to properly identify possible alternatives and to evaluate them based on environmental contingencies, finally choosing the one with the highest utility. 6 Value-based normative decision-making approaches suggest that individuals should choose the alternative in accordance with their beliefs about the expected value of that alternative, in which an optimal strategy would take into account gains, losses, and probabilities (i.e., risk). 7 Thus, deliberative decision-making under risk, in which all the necessary information is available, refers to those situations in which it is possible to evaluate losses and gains for each alternative identified, as well as the risks involved without social contingencies. Deliberative decision-making is highly dependent on individual factors such as sex, 8 neuropsychiatric conditions, developmental stages 1,9 and early life experiences, such as parenting and stress. 10 For example, it has been suggested that adults with high levels of impulsivity and executive function impairments present increased risky behavior in unfavorable deliberative decisionmaking scenarios. 11 Additionally, aspects related to social cognition (e.g., morality, empathy, and how people care and handle social environmental stimuli) have always been guiding factors in social balance and in the construction of legal norms and may be compromised in adolescents in conflict with the law (ACL). A major process of social cognition is theory of mind (ToM), which refers to the ability to infer and understand the emotions, intentions, thoughts, and actions of others. 12,13 Hence, similar to what is proposed by value-based decision approaches, in which environmental contingencies such as the magnitude of losses and gains contingent on each of the alternatives identified might directly influence one's decision, ToM is highly dependent on the dynamic perception of facial cues during group interaction that might guide social value orientation and behaviors.
In this regard, social cognition is also an important correlate of behavioral impulsivity and violence among adults and neuropsychiatric patients. 14,15 Although there is evidence that adults with criminal records with and without associated mental illnesses might perform worse on ToM tasks, 16,17 some studies are inconclusive 14,18 and few studies have explored this ability among adolescents accused of committing a crime.
Altogether, considering that (i) social vulnerability related to adverse developmental experiences is associated with crime initiation and maintenance, with a peak rate of criminal justice reports among those in the late years of adolescence 19 ; that (ii) risky behaviors might reflect, at least in part, impairments to the capacity to properly evaluate losses, gains, and risks for each alternative identified; that (iii) ToM impairments might be associated with behavioral impulsivity, violence, and involvement in criminal activity; and (iv) that few studies have focused on deliberative decision-making under risk and social cognitive features in adolescent offenders; the primary aim of this study was to assess deliberative decision-making and ToM among adolescent offenders who are subject to provisional deprivation of liberty (temporary detention) due to conflict with the law and a control group (CG) of adolescents who are not. Moreover, although value-based decision-making and social cognition are distinct cognitive processes and ontogenetic behaviors, their relationship has not yet been fully explored.

Method Participants and procedure
A total sample of 100 participants (58 ACL, and  20,21 Clinical and cognitive assessment IQ was estimated using both the Vocabulary and the Matrix Reasoning tests from the WASI, 22 while general cognitive functioning (orientation, attention, memory, language, and visual-spatial skills) was assessed with the MMSE. 22,23 Substance consumption was assessed with the CRAFFT Screening Test. 24,25 The CRAFFT screens for alcohol, cannabis, and other drugs and its questions were designed to be developmentally appropriate for adolescents. In addition to assessing the prevalence of substance use (Criteria A score), the questionnaire also allowed us to assess whether the participants already had alcohol or other drug-related problems (Criteria B score). Moreover, a score of 2 has been identified as the optimal threshold for identifying a possible risk for substance use disorders. 26 The Juvenile Victimization Questionnaire second revision (JVQ-R2) was used to assess crime, child maltreatment, and other kinds of victimization experiences during childhood. 27 The JVQ-R2 contains screening questions about 34 offenses against youth that cover five general areas of concern, i.e., conventional crime, child maltreatment, peer and sibling victimization, sexual victimization, and witnessing and indirect victimization. Additionally, the Inventory of Callous-Unemotional Traits (ICU) was used to investigate possible antisocial and/or aggressive youth. 28 This 24-item questionnaire has three subscales: callousness, uncaring, and unemotional.

Decision-making: Columbia Card Task (CCT)
Due to our focus on understanding how participants deliberatively evaluate environmental information such as gains, losses, and risk to make decisions, all participants performed the no-feedback condition of the CCT. 1,11,[29][30][31] In this task, participants are shown a deck with 32 cards placed facedown and three explicit pieces of information: how many losing cards are hidden in the deck (i.e., risk 1 or 3), the amount associated with each losing card (i.e., a loss of −250 or −750 points), and the amount associated with each winning card (i.e., a gain of 10 or 30 points). In each round, participants choose how many cards the computer will randomly select and turn over, knowing that the round will end immediately if the computer selects one of the losing cards. The different combinations of gain, loss, and risk culminate in eight possible decision scenarios that can be sorted from the most favorable (i.e., risk = 1, loss = -250, gain = 30) to the least favorable (i.e., risk = 3, loss = -750, gain = 10), according to the expected value. The primary outcome of the CCT is the average number of cards chosen, which can be interpreted as a general proxy of risk-taking behavior, with a higher number of cards corresponding to greater risk-proneness.
Additionally, the CCT's normative decision-making approach suggests that participants should choose the number of cards to be turned over in accordance with their belief that the subjective value of that number of cards is maximal, in which an optimal strategy takes into account gain, loss, and risk. Therefore, the CCT also enables assessment of how much the information (i.e., gain, loss and risk) weighs on risk behaviors (i.e., number of cards) at an individual level. 11

ToM: Reading the Mind in the Eyes Test (RMET)
We used the RMET to assess the ability to read others' intentions/feelings/thoughts through affective environmental cues such as facial expressions, specifically in the region of the eyes. 32 In this computerized task, 36 black-and-white pictures of the same region of the face (midway along the nose to just above the eyebrows) are shown to the participants. Each image is surrounded by four words regarding mental states and participants are requested to choose the word that correctly depicts the mental state expressed in the picture. To avoid misunderstandings, a glossary is given to the participants that comprises a list of all of the words the test contains with an example of the use of each word in a phrase. 12 The main outcome of the RMET is the total score, which is the sum of correct answers. Nevertheless, it is also possible to investigate different subscales, splitting the items into positive, negative, and neutral emotions. 33

Statistical analysis
Frequency data were analyzed with Pearson's chisquare tests and quantitative data with Student's t tests and Wilcoxon signed-rank tests, when data were non-normally distributed, Wilcoxon rank sum tests (i.e., Shapiro-Wilk W < 0.001, and skew and kurtosis divided by 2 standard errors < 2). Concerning decision-making, after performing Student's t test to analyze potential group differences in overall risk-attitude independently of the decision scenario, we then extracted the riskseeking behavior separately for the most favorable and the least favorable decision scenarios in order to assess risk-taking in a more fine-grained fashion. Differences between groups for both extreme decision scenarios and their interactions were analyzed using a linear mixed effect model (LMM), in which a two-level variable "group" (ACL and CG) and the two-level variable "decision scenario" (most favorable and the least favorable) were included as fixed effects and a random intercept was modeled for each participant. The decision scenario was established by using the expected value for each of the eight combinations of the magnitude of gain, the magnitude of loss, and probability of drawing a loss card. 11 Finally, to investigate how each participant weighted gains, losses, and risk information when making decisions, an LMM was performed for each individual separately including a random intercept for the three blocks and the 24 rounds in the model.
Coefficients to denote how the participant weighted the gain, loss, and risk were extracted by including each one of these two-level factors as fixed-effects. This strategy has been used before and proved to be useful for differentiating people with cocaine use disorders from healthy controls. 11 Even though severe cognitive deficit (IQ < 65) and acute intoxication or withdrawal symptoms were used as exclusion criteria, all main analyses were performed with and without IQ, cannabis use, and CRAFFT criteria B scores as covariates. Years of school education was not included as a covariate in the models due to multicollinearity. The relationship between deliberative decision-making under risk and ToM was analyzed using Pearson's r correlation coefficient. All statistical analyses were performed using R open-source statistical software.

Demographic characteristics
Between-group comparisons of sociodemographic and substance use data are presented in Table 1.
Despite our efforts to keep the matching procedure, the groups differed concerning years of school education and IQ, showing that the ACL group had lower IQ and fewer years of school education. Nevertheless, as mentioned before, the analyses were performed with and without IQ, cannabis use, and CRAFFT Criteria B score as covariates, while years of school education was not included as a covariate in the models due to multicollinearity.
We found that ACL performed better than controls in the MMSE calculation score. Taken together with the overall IQ, this observed difference in the MMSE calculation score (based on the median) could suggest that ACL may perform better in tasks that require more fluid intelligence than crystallized intelligence.
This would indicate that, despite their educational disadvantage, ACL were still able to perform basic math slightly better than controls. Nevertheless, this hypothesis extrapolates the data and does not seem to be in accordance with the literature, which has found that adopted juvenile delinquents scored lower in the arithmetic subscale of the Wechsler Intelligence Scale for Children compared to the population mean. 34 It should be noted, however, that both controls and ACL still performed within the expected average in our study. In general, ACL reported more cannabis use and more problems related to substance use when compared to controls, with most of the ACL group being at risk for substance use disorder. No differences were found for juvenile victimization or antisocial personality traits.

ToM
In general, our RMET data (Figure 1) suggest that the ACL differ from CG concerning ToM, as depicted by the group differences in the total score (t 58.09 = 2.14, p = 0.036). However, this finding is mainly explained by group differences in the ability to correctly infer negative emotional states (t 55.36 = 2.32, p = 0.023), since no significant group differences were found in the positive (t 58.64 = 1.35, p = 0.181) or neutral (t 59.57 = 1.28, p = 0.204) subscales. When IQ, cannabis use, and CRAFFT Criteria B were included in the model, neither group differences in total score nor in the negative subscale score remained. However, an effect was found for IQ in the total score (t 57 = 2.08, p = 0.041, r = 0.082), while an effect for cannabis use was found in the negative subscale (t 57 = -2.07, p = 0.042, r = 0.264).

Deliberative decision-making under risk.
No significant group differences were found regarding the overall number of cards chosen, suggesting that groups did not differ concerning general risk-taking behavior (t 53.73 = -0.08, p = 0.935, r = 0.011).
Nonetheless, a more detailed inspection applying LMM to the most favorable and the least favorable decision scenarios revealed an effect for the decision scenario, suggesting that, in general, participants chose less cards in the least favorable scenario when compared to the most favorable scenario (β = -0.51, t 60 = -3.14, p = 0.002, r = 0.376). An interaction between the ACL group and decision scenario was found (β = 0.50, Use of risk, loss, and gain information was extracted for all participants individually. Subsequently, group averages were compared and a group effect was found for risk information only (Figure 3). Due to the small sample size and lack of power, no significant difference between groups was found concerning use of loss and gain information, as shown in Figure 3. Remarkably, the groups showed distinct patterns of use of information.
While the risk and loss information had a negative effect on the number of cards chosen by the CG adolescents, as would be expected, we observed that the same information tends to increase the number of cards chosen by members of the ACL group. Moreover, as shown by the effect sizes, the most used information was risk, followed by loss and gain, successively. These  results remained significant even when including IQ, cannabis use, and CRAFFT Criteria B score.

Decision-making and ToM
Finally, concerning the relationship between deliberative decision-making under risk and affective inference of emotional states through ToM, no correlation was found between the main CCT and RMET outcomes.

Discussion
Here we examined the differences between ACL and a CG regarding deliberative decision-making under risk, using the CCT, and ToM, assessed using the RMET.
The key findings are that ACL differed from controls concerning their capacity to properly infer emotional states in others and to make decisions under risk scenarios. Specifically, we found that ACL committed more errors when inferring negative emotional states, but no difference was found in the error rate concerning or psycho-social well-being of themselves or others. 30 Moreover, when facing affective contexts, such as those that require social cognitive abilities, as in the RMET, the active affective neural system may be impaired, overloaded, or even blunted. 36,37 Together, our findings could suggest that ACL may have even later maturation years. 38 Similarly, it was also found to be a predictor of a lack of appreciation of social rules and norms, especially when associated with psychopathologies. 39,40 Nevertheless, whether ToM can explain an individual's propensity to delinquent behavior is under debate.
On one hand, Majorek et al. 39

Disclosure
No conflicts of interest declared concerning the publication of this article.