Andreas Bauer

Research on legitimacy and crime is dominated by cross-sectional evidence from high-income countries. Furthermore, less is known about how legitimacy and deterrence jointly shape adolescent violence in fragile institutional contexts. Using longitudinal data from the 2004 Pelotas Birth Cohort Study in Brazil (n = 1,510), a population-based, prospective longitudinal study, we examine whether perceived punitive consequences and three legitimacy dimensions — regarding the police, the law, and the social system — are associated with violent crime at ages 15 and 18. Higher perceived punitive consequences and more positive perceptions of legitimacy across all domains were associated with lower odds of violent crime at age 15, whereas only police legitimacy and perceived punitive consequences were associated with lower odds of violent crime at age 18. Legitimacy of the law and of the social system moderated the association between perceived punitive consequences and violent crime at age 15, with negative associations strongest among adolescents who viewed institutions as less legitimate. These findings suggest that legitimacy is multidimensional and that both normative and instrumental processes jointly shape adolescent violence in high-violence contexts.

Depression is a global public health challenge, with 332 million people experiencing depression. Given that more than half of depression patients experience relapse, researchers have argued that investigating the causes of depression at a symptom (rather than syndrome) level is required to progress understanding. This study investigated the causal structure of depressive symptoms using data from a study that followed a 57-year-old individual over 239 consecutive days. The study followed the analytical steps: (1) descriptive analysis; (2) regime identification; (3) regime-dependent causal discovery; (4) causal model estimation; and (5) centrality analysis. The findings indicated the existence of two regimes, aligned with pre- and post-transition to depressive symptoms. In the second regime (where symptom severity increased), deficits in concentration and low self-esteem constituted intervention targets. The current study showed that state-of-the-art causal methods can be used to gain new insights into causal processes underlying depression and inform personalised mental health interventions.

In contemporary society, prediction models are extensively used to predict psychological outcomes, such as conduct problems, suicide, sexual reoffending, among others. Despite their widespread implementation, to the best of our knowledge, no studies have investigated model fairness from a causal perspective, elucidating discrimination pathways. The current case study aims to conduct a causal fairness analysis of a prediction model of conduct problems at age 14 years. Data were from the Longitudinal Study of Australian children, and the sample included adolescents (n=6,442) who participated in two follow-ups (ages 12 and 14). Conduct problems were measured with the Strengths and Difficulties Questionnaire. The prediction model was developed with Minimax Convex Penalty logistic regression and discrimination was evaluated with the Area Under the Receiver Operating Characteristic Curve (AUROC) and the Area Under the Precision-Recall Curve (AUPRC). Fairness metrics such as demographic parity (DP) were used to detect potential violations. In the causal fairness analysis, the causal structure was discovered with the Direct Linear Non-Gaussian Acyclic Model algorithm, while fair adaptation with quantile preservation was used to generate fair data and train the fair prediction models. The findings indicate that, while the prediction model was discriminatory according to Indigeneity (DP=15%) and sex (DP=11%), a causally fair prediction model retained good discrimination (AUROC=0.74, AUPRC=0.43) while mitigating unfairness. In conclusion, this case study showed that prediction models can be discriminatory, reproducing structural inequalities captured in the data they were trained on, so careful fairness consideration and adaptation are required before societal implementation.

Background: Callous‐unemotional (CU) traits in early childhood are risk markers for persistent antisocial behaviour, with marked adverse life outcomes. However, associations of CU traits with broader psychosocial characteristics are not well understood, particularly in early childhood. Objective: To examine developmental and psychosocial profiles of pre‐school children according to their levels of CU traits and conduct problems (CPs) in a large, population‐based cohort from southern Brazil. Methods: The 2015 Pelotas Birth Cohort includes 4,275 children followed from birth. At age 4 years, CU traits and CP were assessed via parent‐report questionnaires, and 19 other psychosocial characteristics were examined mostly in task‐based assessments regarding: neurodevelopment, language, executive functioning, emotion recognition, theory of mind, social information processing and empathy/prosocial behaviours. These 19 psychosocial characteristics were compared across four groups of children with varying levels of CU/CP, and regression models were used to test for independent associations with CU traits and CP, adjusting for sociodemographic factors at birth. Results: At age 4 years, 14.8% of children had elevated CU traits only, 14.2% had CP only, and 16.3% had both high CU + CP. Compared to children without CU traits or CP, children with higher CU traits had poorer performance across multiple domains of development, including language, motor development, emotion recognition and prosocial behaviour. In contrast, children with CP‐only showed more circumscribed deficits, mainly in self‐control and emotional functioning. In regression analyses, CU traits were associated with broader neurodevelopmental impairments, whereas CP was more specifically linked to behaviour regulation. Conclusions: Elevated CU traits in early childhood are associated with broad developmental difficulties, whereas CPs relate more specifically to emotional and self‐regulation challenges. These findings underscore the need for holistic approaches to understanding early manifestations of CU traits and interventions in the context of CP behaviours.

Person-centred approaches assume homogenous subgroups and are limited in capturing the nuanced relationships among bullying behaviours. We applied network analysis – a variable-centred approach – to examine bullying using the self-reported Zurich Brief Bullying Scales (measuring social exclusion, property destruction, verbal and physical aggression, and sexual bullying) in a representative sample of 2,598 15-year-olds (48.0% female) in São Paulo, Brazil. To identify unique item combinations categorised into bullying roles, we created an UpSet plot. We estimated a mixed graphical model using a regularised nodewise regression approach via the Extended Bayesian Information Criterion, and subsequently examined node predictability, clustering, centrality, and network stability. There was substantial heterogeneity in bullying roles, particularly for bully-victims. The network model revealed a relatively high density, with 33 out of 45 edges, and two distinct but deeply intertwined dimensions of bullying perpetration and victimisation, which were primarily linked by victimisation through physical aggression. Perpetration through verbal aggression was most strongly connected to individual bullying behaviours. Node predictability was modest, suggesting greater influence from factors outside the network. Verbal aggression showed higher predictability and could be reduced by targeting other forms of bullying, whereas physical aggression with lower predictability may be best addressed directly or by targeting external risk factors. The network approach represents an alternative approach to mapping the complexity of bullying, conceptualising bullying as mutually reinforcing behaviours. Researchers may want to integrate the network approach and leverage its decade of progress to uncover new insights into bullying involvement, leading to more effective prevention and treatment strategies.

Objective: This study aimed to evaluate the configural and metric structure, as well as internal consistency, of the Brazilian Portuguese adaptation of an instrument designed to assess mistreatment during childbirth. Methods: Data were drawn from the “Birth in Brazil” II, carried out in 2023/2024, comprising 1,881 women hospitalized for childbirth in the state of Rio de Janeiro, Brazil. Initially, confirmatory factor analysis was conducted to test the original dimensional structure. Due to an inadequate model, exploratory factor analysis was performed to reassess the factorial structure. Applying predefined criteria, items presenting cross-loadings or possible residual correlations were removed. The remaining items were subsequently evaluated using confirmatory factor analysis, and reliability was assessed using Cronbach’s α coefficients. Results: The final solution comprised 17 items distributed across four dimensions: verbal abuse; neglect and disrespect; stigma, discrimination and lack of privacy; and innappropriate vaginal examinations. The final model presented acceptable fit indices, satisfactory factorial definition, and no item redundancy. Most dimensions showed reasonable internal consistency, except the stigma, discrimination, and lack of privacydimension. Conclusion: The findings suggest that further psychometric evaluation is necessary to establish the instrument’s validity and reliability. Its use for assessing mistreatment during childbirth or obstetric violence in Brazil should be done with caution. Future psychometric studies using this instrument are recommended.

IMPORTANCE: Adverse childhood experience (ACE) scores mask substantial heterogeneity within the 2-, 3-, and ≥4-ACE exposure categories. The magnitude of this heterogeneity has not been sufficiently quantified. OBJECTIVE: To quantify and visualise the heterogeneity of ACE scores. DESIGN: Cross-sectional analysis of the 2023 National Survey of Children’s Health, an address-based, nationally representative survey. SETTING: US households. PARTICIPANTS: A national sample of 385,000 addresses was drawn. Between 23 June 2023 and 19 January 2024, caregivers of children aged 0–17 years completed screening and topical questionnaires; 55,162 of 70,187 eligible households completed the topical questionnaire. EXPOSURES: Lifetime exposure to 10 ACEs, assessed via caregiver report: economic hardship; parental divorce/separation; household mental illness; household substance use; parental incarceration; domestic violence; racial discrimination; neighbourhood violence; health- or disability-based discrimination; and parental death. MAIN OUTCOMES AND MEASURES: Heterogeneity of ACE co-occurrence within the 2-, 3-, and ≥4-ACE exposure categories, quantified using combinatorial coverage (CC; proportion of possible co-occurrence patterns observed) and visualised with an UpSet plot of item combinations with unweighted n ≥ 30. RESULTS: Analyses used complete ACE data ( N = 51,468; 93.3% of the total sample); co-occurrence analyses were restricted to children with ≥2 ACEs ( n = 7,897). Across the 2-, 3-, and ≥4-ACE categories, 582 of 1,013 possible combinations were observed: all 45/45 patterns for 2 ACEs (CC ₂ = 1.00), 110/120 for 3 ACEs (CC ₃ = 0.917) and 427/848 for ≥4 ACEs (CC≥4 = 0.504). Of the 582 distinct combinations, 50 met the display threshold, accounting for 5,204 (66%) of the 7,897 children with ≥2 ACEs: 23, 15 and 12 for 2, 3 and ≥4 ACEs, respectively. Certain ACEs recurred disproportionately among these intersections: parental divorce/separation appeared in 34 of 50 patterns, household substance use in 23 and household mental illness in 20. CONCLUSIONS AND RELEVANCE: In this nationally representative sample of US children, ACE scores masked substantial heterogeneity in exposure patterns within the 2-, 3-, and ≥4-ACE categories. Simple heterogeneity diagnostics, such as combinatorial coverage and UpSet plots, can make this heterogeneity explicit and may prompt pattern-specific analyses of outcomes. Key Points: Question: How heterogeneous are cumulative adverse childhood experience (ACE) scores, and how can this heterogeneity be quantified and visualised? Findings: In a nationally representative cross-sectional survey of 51,468 US children, 582 of 1,013 theoretically possible ACE combinations among those with ≥2 ACEs were observed: all 45/45 patterns for 2 ACEs, 110/120 for 3 ACEs and 427/848 for ≥4 ACEs. When visualised with an UpSet plot, some individual ACEs recurred disproportionately among the 50 most common co-occurrence patterns, which together accounted for 5,204 (66%) of the 7,897 children with ≥2 ACEs. Meaning: ACE scores mask substantial heterogeneity within the 2-, 3- and ≥4-ACE exposure categories, risking misleading inferences about which children are at highest risk; this heterogeneity can be made explicit using a simple metric and visualisation technique.

Introduction Person-centred approaches, such as classifying children as bullies, victims, and bully-victims, assume homogenous subgroups and are limited in capturing the nuanced relationships between different forms of bullying. The current study proposes network analysis, a variable-centred method, as an alternative approach to examining bullying involvement.Methods We used data from the São Paulo Project for the Social Development of Children and Adolescents (SP-PROSO), a representative cross-sectional survey of high school students aged 15 years in São Paulo, Brazil (N = 2,598; 48.0% female). The self-reported Zurich Brief Bullying Scales were used to measure social exclusion, property destruction, verbal and physical aggression, and sexual bullying. To identify unique item combinations categorised into bullying roles, we created an UpSet plot. We estimated a mixed graphical model using a regularised nodewise regression approach via the Extended Bayesian Information Criterion, and subsequently examined node predictability, clustering, centrality, and network stability.Results There was substantial heterogeneity in bullying roles, particularly for bully-victims. The network model revealed a relatively high density, with 33 out of 45 edges, and two distinct but deeply intertwined dimensions of bullying perpetration and victimisation, which were primarily linked by victimisation through physical aggression. Perpetration through verbal aggression was most strongly connected to individual bullying behaviours. Node predictability was modest, suggesting greater influence from factors outside the network. Verbal aggression showed higher predictability and could be reduced by targeting other forms of bullying, whereas physical aggression with lower predictability may be best addressed directly or by targeting external risk factors. Conclusions The network approach represents an alternative approach to mapping the complexity of bullying, conceptualising bullying as mutually reinforcing behaviours. Researchers may want to integrate the network approach and leverage its decade of progress to uncover new insights into bullying involvement, leading to more effective prevention and treatment strategies.

Background: Many studies have found that a small group of chronic offenders is responsible for the majority of crimes, and tend to be particularly violent. However, there is a major lack of evidence on chronic offending in low and middle-income countries; understanding these patterns is especially important in settings with very high levels of violence, like Brazil.Aims: To identify the extent that crime is concentrated in chronic offenders and linked to violence and homicide in a Brazilian cohort. Methods: The 1982 Pelotas Birth Cohort includes 5,623 participants followed from birth in 1982 with complete crime records up to age 30. Chronic offenders were defined as individuals who committed five or more offences and over 50% of all crimes. Key offending characteristics including age of onset, violent and homicide offending, and incarceration rates were compared across three offender groups (one-timers, below-chronic, and chronic offenders).Results: Approximately 27% of participants committed at least one crime by age 30, and 5% were chronic offenders who were responsible for 57% of all crimes, 54% of violent crimes, and 68% of homicides. Chronic offenders exhibited more severe patterns of offending on all characteristics examined, compared to non-chronic offenders. Homicides committed by chronic offenders were more likely to involve firearms than those by non-chronic offenders.Discussion: This is the first study to analyze chronic offending within a large, prospective cohort in a low and middle-income country. Chronic offending patterns in this Brazilian setting resemble those in high-income country studies, and are highly related to homicides.