Exploring the Correlates of Arrest for Violent and Serious Crimes in Children and Adolescents: Mental Health Implications

Abstract

Existing studies point to relationships between child abuse and crime perpetration in adulthood. Child abuse and trauma are strongly connected to mental health. In this study, the relationship between child abuse and neglect (AN), out-of-home placement (OP), lead poisoning (LP), and arrest for serious offenses including murder and homicide AC, prior to adulthood was explored. Findings showed that AN significantly predicted arrest for violent and serious crimes. OP did not have a significant relationship with arrest for serious and violent crimes AC. LP was not significantly associated with arrest for serious crime AC. Implications on child protection, mental health, and social work education are discussed.

1. Introduction

Some children and adolescents do engage in serious crimes ranging from animal cruelty [1, 2] to homicide [3]. Existing evidence also highlights associations between mental health problems such as psychosis and conduct disorder, and criminal involvement [4]. Recently, in a study of adult male rats exposed to lead poisoning, Ghaderi, Komaki, Salehi, Basir, and Rashno [5] documented evidences of links between lead exposure and cognitive decline, which can be a correlate of aggression [5]. White and Gala [5] have called attention to the need to mitigate lead exposure. Lead poisoning occurs when a child (for the purpose of this study) who is under the age of 6 has “had blood levels of 10 micrograms per deciliter or above”. A case of lead poisoning is confirmed with venous blood test ≥5 mcg/dl or 2 capillary tests within 72 days both ≥5 mcg/dl [6]. Lead poisoning exposure is well documented as an environmental justice issue [7] with numerous mental health and criminal justice implications. According to Mayo Clinic ([8], para 1–2), lead poisoning occurs when there is a buildup of lead in the body. It takes several months or years for the build-up to occur. According to Mayo Clinic [8], even minimal levels of lead may result to serious health challenges. Particularly children below the age of 6 years are vulnerable to lead poisoning, which can severely affect their mental and physical development. Death could occur where the level of lead poisoning is very high. Sources of lead poisoning include lead-based paint and lead-contaminated dust in older buildings. These are the main sources of lead exposure and poisoning in children [8]. According to Mayo Clinic [8], additional sources of lead exposure and poisoning include water, soil, and contaminated air.

This study adds to the current conversation on environmental and policy factors that are related to mental health and criminal offending. The study applied an exploratory design and used a county-level data because: Counties play significant roles in administering and addressing issues related to the study’s key variables. Second, counties provide a unit of authority from which careful comparisons could be made regarding the variables under study.

2. Background

In 2012, 25,839 cases of child maltreatment were assessed by Minnesota’s counties and tribal agencies. Of these reports, 30% of the children were offered services [9]. About 43% of all cases of maltreatment reports in the state involved children aged 5 years or younger, 22% involved children aged 3 years or younger, while 9% involved children under the age of 1. Of the children who were abused, 171 were in out-of-home facilities such as foster homes [9]. In the same year, 8 children died from maltreatment, while 52 children suffered life-threatening injuries. Of the total maltreatment reports in the State in 2012, 60% related to non-medical neglect, 35% alleged physical abuse, and 10% alleged sexual abuse while medical neglect and emotional or mental abuse were alleged by 1%. Families and caregivers accounted for perpetrators of 4309 cases of abuse and neglect in 2012 with more than 50% of the cases requiring protective services, which are provided at the county level. Also, birth parents accounted for 77% of child victimization while other relatives such as adoptive parents, stepparents, grandparents, and siblings were responsible for 12% of the total child maltreatment, with parental companions accounting for 7% of the offenses. According to Minnesota Department of Human Services (DHS) [9], families in which child maltreatment occurred most were characterized with mental illness, domestic violence, substance use, poverty, housing needs, and parenting issues. According to Minnesota DHS [10], child protection agencies in Minnesota received a total of 85,917 reports of child abuse and neglect. About 43% of the investigated maltreatment cases were assessed to have occurred. Of the maltreatment reports, in 2019, 17 child deaths occurred, while 13 life-threatening physical injuries were determined to have occurred because of child maltreatment [10].

2.1 Child abuse and criminal offending

Child abuse and neglect can be defined as offensive parenting, such as beating a child, sexually molesting a child, and withholding necessities or care from a child. As demonstrated in the literature, child abuse and neglect (AN) remains a serious problem across the world. It is well-documented as a risk factor for poor health outcome, poor academic performance [11, 12], poor attachment with parents and caregivers [13, 14, 15], substance abuse [16], employment and relational problems, physical aggression, and future criminal offending [16, 17]. Several studies have assessed the impact of AN on criminal offending during adulthood [18, 19]. Of interest is that AN has remained a risk factor for future crime involvement, specifically, arrest for serious crimes [11, 20]. Widom and Maxwell [21] assessed the associations between child abuse and adult criminal involvement in a metropolitan area and found that victims of AN were more likely to be arrested for crimes in their adulthood (42%) compared to those who did not have such experience (about 33%). Graff, Chihuri, and Blow [22] found that the effects of adverse childhood experiences on a victim could lead to multiple criminal offending in adulthood. Widom et al. [23] also found that AN was a predictor of externalizing problems (mental health issues that could manifest in use of physical force) which is a strong correlate of criminal involvement [16, 17, 22, 24, 25, 26, 27, 28, 29] have also documented associations between physical abuse and externalizing behaviors in the male children. Sexual abuse was reported as a predictor of relational aggression in females. But despite that several studies have highlighted linkages between AN and future criminal offending during adulthood [14, 29, 30, 31] associations between AN and involvement in serious criminal offenses vis a vis lead poisoning prevalence, at the county level seem to have remained understudied.

2.2 Out-of-home placement, mental health, serious crime, and lead poisoning

According to the Minnesota Department of Health Services, DHS (2015), out-of-home placement (OP), is a form of family assessment program that is implemented by the child protective system as an intervention to protect a child from potential or further maltreatment. OPs may include foster homes, adoptive homes, and shelters. Children in OPs often carry emotional burdens such as severe poverty, dysfunctional family situations, mental health issues, abuse, and neglect [32]. Studies have suggested that these background legacies may hinder a child’s capacity to become successful [33, 34]. OPs, particularly foster care, have become a usual means used by child protective systems to prevent further maltreatment of children. Hence, nationally, use of foster care increased more than 50% (from 276,000 to 568,000) between 1985 and 1999 [35]. As of 2006, about 3.6 million children in the USA were in the child protective system, within which more than 300,000 were in OPs [36]. Some studies have examined the links between foster care placement and criminal involvement. In a study that examined correlations between exit from foster care typology and entry into the correctional system, among 10,000 Wisconsin youth, Font et al. [37] found that about 13% of the study sample experienced incarceration as young adults. The authors echoed a need to enhance efforts to reduce risks of imprisonment among foster care youth. Lindquist and Santavirta [32] assessed the associations between placement in foster care as a child and criminal involvement during adulthood. Findings showed that being placed in a foster care predicted higher criminal involvement as an adult among boys (aged 13–18).

It is important to note that all OPs do not result from child maltreatment and do not result to criminal involvement. Sometimes, children may be in OPs due to their parents being incarcerated, due to a crime of their own, or due to not having an adult to care for them. According to the National Conference of State Legislatures [38] as of 2007, there were about a total of 1.7 million children of incarcerated parents in the USA. One in 110 white non-Hispanic children (n = 484,100) had their parents incarcerated. One in 15 black non-Hispanic children (n = 767,400) had their parents incarcerated, while one in 41 Hispanic children (n = 362,800) had their parents incarcerated. Further, more than 50% of the incarcerated males across race (Black and Hispanic) are more likely to be a parent compared to white parents (45%). Surprisingly, mothers of minor children accounted for 62% of inmates in state prisons and 56% of inmates in federal prisons. Because OP is associated with important baseline risks [36], it is necessary to continue to explore its relationship with serious criminal offending, especially in maltreated children [39] from a prevention and care standpoint.

2.3 Conceptual considerations

The social learning theory integrates cognitive and behavioral explanations in examining human behavior. It posits that learning occurs within a social milieu. It further posits that human behavior is a result of the interplay between cognition and environmental exposures or inputs ([40], p. 155). Thus, people learn through observation, imitation, and modeling. Learning by modeling implies the existence of a role player (an actor) or a symbolic model. It is such modeling that leads to replication of behavior through mimicking or practice [40]. With this understanding criminal behaviors are not the results of hereditary factors ([40], p. 204) only. Instead, involvement in serious crime and violence is considered as a learning outcome that has been possible due to previous exposure to a similar situation in which there is a teacher and a learner. It is thus the prevalence of these individual exposures and replication of such inputs that account for high or low prevalence of serious crimes in a county or a setting. Social learning theorists posit that the first process through which the transmission of a learned behavior occurs is ‘observation’. They argue that such learned negative behavior or serious violence or crime can be maintained or eliminated with the use of positive or negative reinforcements ([40], p. 155). Positive and negative reinforcements can be deterring factors such as local law enforcement, policies, and other tools that are available at a county level including police presence and mental health services. Within this understanding, punishment decreases undesired behavior. Social learning theorists further argue that the type of reinforcements received by a learner, and in this case a child or youth (either by a positive influencer or a negative influencer) determines whether the learned behavior will be imitated, mimicked, or avoided by the learner. These assumptions have been empirically tested through the Bobo doll experiment ([40], pp. 206, 208) in which exposure to violence predicted secondary violent behaviors by the learner.

Applying a social learning lens to this study will mean harnessing the components of effective learning, attention, which is the first step in learning ([41], p. 139), retention (the coding of the information and its transfer to conscious memory), reproduction of modeled behavior (the ability to reproduce a particular learned behavior), and motivation or reinforcement ([40] p. 245). Hence, policies and tools to prevent perpetration of crime during childhood would have to address all the dimensions mentioned above. Social learning theorists also postulate that reciprocal causation (or two-way communication) occurs at three different levels. These levels are the individual level, the behavioral level, and the environmental level. They also consider the construct of self-efficacy to be vital in social learning. For example, they argue that youth will replicate behaviors that they have learned only if they are convinced that they have self-efficacy to engage in such behavior ([40], pp. 49, 163). The explanation for arrest for serious offenses is that criminal behavior must have been previously modeled to the child ([41], p. 139; [40], p. 115). All these result to higher propensity to engage in serious crimes that were modeled by an abuser, moderated by the child’s attention, retention, and self-efficacy or self-reinforcement ([40, 42], pp. 316–317). Hence, participation in serious crimes may be a way that a child uses to test the child’s self-efficacy. Building on a social learning lens and the studies reviewed above, the author hypothesizes that prevalence of child abuse and neglect, and out-of-home placement, and child lead poisoning exposure, will predict serious crime perpetration [11, 37, 43, 44]. Hence, the goal of an intervention will be to ensure that children and adolescents do not get exposed to delinquency, and that there is enough tool to mitigate such delinquencies when they occur so that re-perpetration of violence and serious crime can be stopped.

3. Method

3.1 Sample

Data for the study were specifically derived from Minnesota indicators for 2011: Safety and risky behavior indicators (child abuse and neglect), juvenile justice (children arrested for a serious crime), and out-of-home placement totals for 2011, derived from the Annie E. Cassie Foundation [44]. The Kids Count data were derived from Minnesota Department of Human Services, Minnesota’s Child Welfare Report. This study considered the entire counties in the State of Minnesota, and the unit of the analysis is the county. There are a total of 87 counties in Minnesota, excluding tribal territories, and the dataset considered each county as a datapoint (see Figure 1).

Figure 1.

Map of Minnesota Counties. Minnesota Employment and Economic Development. Minnesota County Profiles. https://mn.gov/deed/data/data-tools/county-profiles/2023

3.2 Procedure

This cross-sectional study relied on secondary data from The Annie E. Casey Foundation (Kids Count Data Center). The study relied on 2011 data because it is the dataset with complete variables for the kind of analysis that was required. This study focused on the entire counties in the state of Minnesota. All the counties in the state (N = 87) excluding tribal territories were included in the analysis, hence, the county was the study’s unit of analysis.

3.2.1 Operationalization of study variables

According to Minnesota Department of Human Services ([6], para 1–5), child abuse comprises three categories: (1) physical abuse, comprising “any physical injury or threat of harm or substantial injury, inflicted by a caregiver upon a child other than by accidental means,” (2) mental injury: refers to harm to a child’s psychological capacity or emotional stability evidenced by an observable and substantial impairment of the child’s functioning, (3) sexual abuse: involves subjecting a child to a criminal sexual act or threatened act by a person responsible for the child’s care or by a person who has a significant relationship to the child or is in a position of authority. Neglect involves the failure of the child’s caregiver to: “Supply the child with necessary food, clothing, shelter, medical or mental health care, or appropriate supervision, protect the child from conditions or actions that endanger the child, take steps to ensure that a child is educated according to the law. Exposing a child to certain drugs during pregnancy and causing emotional harm to a child may also be considered neglect” [9]. For this study, the term, child abuse and neglect and child maltreatment are used interchangeably.

The variable, AN reflects the number of children “of whom a report of child abuse or neglect was substantiated by a county child protection worker” [6]. The variable, AC reflects individuals under the age of 18 who were incarcerated due to homicide, murder, robbery, rape, aggravated assault, vehicle theft, rape, larceny, and arson. The source of this data is Minnesota Bureau of Criminal Apprehension. The variable, OP is the number of children who had been in group homes, foster care, residential treatment facilities, foster care, including children who were placed with their relatives. The source of this data is Minnesota Department of Human Services, Minnesota Child Maltreatment Report [6]. The variable, LP, is the number of children under the age of 5 who tested and were confirmed to have “had blood levels of 10 micrograms per deciliter or above. Cases that were confirmed are those with venous blood test ≥5 mcg/dl or 2 capillary tests within 72 days both ≥5 mcg/dl.” [6]. The year 2011 was used as the focal point of the study because it contained complete data matching the rest of the variables of interest in this study. This study is unique in that it looked at child welfare from the macro-level perspective, which is an important lens for social work.

3.2.2 Data analyses

Data were cleaned and examined for suitability with the different analyses that were conducted. The analysis focused on prevalence or the sum of incidences of AN, OP, LP, and AC. The number of counties in Minnesota is finite and small, in comparison with other states in the United States. For this reason, the sample size for the analysis was small (given that each county represented one sample). However, that is the maximum number that the state has, at the county level. To address the challenge of sample size, bootstrapping method was applied, which is an acceptable method for dealing with studies with small sample size [45]. Data was analyzed using SPSS and intellectus statistics. Findings from the bootstrapping analysis were compared with findings from the raw sample, and they showed similar results. Logistic regression is very robust to violation of assumptions of multiple regression. Logistic regression was applied in exploring the multivariate relationships within the study variables. Descriptive analyses of the key variables were conducted, which showed the distribution of the key study variables, and the data was subjected to further probing which showed high levels of skewness and kurtosis. For point biserial correlation analysis, a preliminary analysis of the data using the predictor variables in their original form led to fitted probabilities that were numerically close to 0 or 1, suggesting that there was possible minimal overlap in the variability of some of the predictor variables between low and high prevalence of AC. The association between the variables was assessed using Cohen’s standard, where 0.1 represents a small effect size, 0.24 represents medium effect size, and 0.37 represents large effect sizes [46], given the assumption that both the low (0) and high (1) values in AC might equally occur [47, 48]. Findings from the analysis were reviewed using Holm corrections, which is applied in adjusting for numerous comparisons using an alpha value of 0.05. For this reason, the logarithm of the variables was taken. This solution seemed to address the problem of minimal overlap in the variability of some of the independent variables. The predicted variable, AC was re-coded into a binary form, using the mean value of the original distribution of the variable as the lower cut-off. Hence, values that felt below the mean were assigned 0 while those that went above the mean were assigned 1. For the logistic regression analysis, the reference category for prevalence of AC was (0, low arrest incidence). The most frequently observed category of AC was 0 (n = 44, 50.57%), while the least frequently observed category of AC was 1 (n = 43, 49.43). The data was assessed for multicollinearity by examining the variance inflation factors (VIFs). All predictors in the regression model had VIFs below 10. Prevalence of OP had a VIF od 1.40, prevalence of abuse and neglect had a VIF of 1.41, prevalence of LP had a VIF of 7.36, and the interaction between prevalence of abuse and neglect and prevalence of lead poisoning had a VIF of 7.40. High VIFs indicate increased effects of multicollinearity in the model. VIFs greater than 5 are cause for concern, whereas VIF of 10 is the maximum upper limit [45].

4. Results

Summary statistics were calculated for the key variables, in their untransformed form: LP, AN, AC, and OP. The observations for OP had a range of 6–2399 and a median score of 55. The observations for LP had a range of 0–162, with a median score of 2.00. The observations for AN had a range of 0–3058, with a median score of 21. The observations for AC had a range of 0–2949, with a median score of 13. Upon transformation, the observations for OP had an average of 1.76 (SD = 0.50, Min = 0.78, Max = 3.38). The observations for LP had an average of 1.13 (SD = 1.06, Min = 0.00, Max = 5.09). The observations for AN had an average of 1.26 (SD = 0.60, Min = 0.00, Max = 3.09).

4.1 Point biserial correlation

A point biserial correlation analysis was conducted for the variables—AC (converted to binary) of OP (log value), AN (log value), and LP (log value). A point biserial correlation is a special case of the Pearson correlation. The observed correlations, along with the bootstrapped results for the standard error and the 95% confidence interval of each correlation, are presented in Tables 1 and 2.

Combination	r	95.00% CI	n	P
AC-LP	0.50	[0.32, 0.64]	87	<0.001
AC-AN	0.61	[0.46, 0.73]	86	<0.001
AC-OP	0.60	[0.45, 0.72]	86	<0.001

Table 1.

Point biserial correlations for AC and LP, AN, and OP.

Note. p-values were adjusted using the Holm correction.

Combination	r	SE	95.00% CI
AC-LP	0.50	0.07	[0.36, 0.63]
AC-AN	0.61	0.06	[0.48, 0.72]
AC-OP	0.60	0.06	[0.48, 0.71]

Table 2.

Observed correlations with bootstrapped results for the standard error and the confidence interval.

Point biserial correlations for AC (0,1) and OP (continuous), AN (continuous), LP (continuous), and observed correlations with bootstrapped results for the standard error and the confidence interval. Note. n = 87; Holm corrections used to adjust p-values for the calculation with the original sample. Table 1 = raw analysis (n = 86–87). Table 2 = bootstrapped analysis (n = 2000).

The biserial correlations indicated that there was a statistically significant correlation between AC and LP with a correlation coefficient of 0.50 (p < 0.001; 95.00 % CI [0.32, 0.64)) indicating a large effect size. The analysis for AC and AN found that there was also a significant positive correlation between AC and AN with a correlation coefficient of 0.61 (p < 0.001, 95.00% CI [0.46, 0.73]), indicating a large effect size. Dealing with AC and OP, it was found that there was a significant positive correlation between AC and OP with a correlation coefficient of 0.60 (p < 0.001, 95.00% CI [0.45, 0.72]), indicating a large effect size.

The analysis with the bootstrapped sample showed that there was a statistically significant correlation between AC and LP (N = 2000) with a correlation coefficient of 0.50 (SE = 0.07, 95.00% CI [0.36, 0.63]), indicating a moderate effect size. For AC and AN, it was found that there was a significant positive correlation between AC and AN (N = 2000) with a correlation coefficient of 0.61 (SE = 0.06, 95.00% CI [0.48, 0.72]), indicating a large effect size. Dealing with AC and OP, the analysis indicated that there was a significant positive correlation between AC and AN (N = 2000) with a correlation coefficient of 0.61 (SE = 0.06, 95.00% CI [0.48, 0.72]), indicating a large effect size. For AC and OP, the analysis pointed out that there was a statistically significant positive correlation between AC and OP (N = 2000) with a correlation coefficient of 0.60 (SE = 0.06, 95.00% CI [0.48, 0.71]), indicating a large effect size (Table 3).

Variable	B	SE	χ^2	p	OR	95.00% Cl
(Intercept)	−2.64	0.59	20.01	<0.001	—	—
OP	0.01	0.008	2.55	0.110	1.01	[1.00, 1.03]
LP	0.24	0.13	3.22	0.073	1.27	[0.98, 1.64]
AN	0.05	0.02	4.59	0.032	1.05	[1.00, 1.09]
LP:AN	−0.0006	0.0002	7.60	0.006	1.00	[1.00, 1.00]

Table 3.

Logistic regression results with OP, LP, and AN predicting AC.

Note. χ²(4) = 49.95. p < 0.001, McFadden R² = 0.41.

4.2 Logistic regression

Logistic regression was conducted to examine whether prevalence of OP, prevalence of abuse and neglect, and prevalence of LP had a significant effect on the odds of observing higher prevalence of AC. The following equations were applied:

$$\ln \left(\text{p}/1-\text{p}\right)=\beta 0+{\beta }_1{\text{x}}_1+{\beta }_2{\text{x}}_2+{\beta }_3{\text{x}}_3+{\beta }_1{\text{x}}_1\ast {\beta }_2{\text{x}}_2$$E1

$$\ln \left(\text{p}/1-\text{p}\right)=\beta 0+{\beta }_1{\text{x}}_1+{\beta }_2{\text{x}}_2+{\beta }_3{\text{x}}_3+{\beta }_2{\text{x}}_2\ast {\beta }_3{\text{x}}_3$$E2

Where ln(p/1−p) = outcome: arrest for serious crime, AC β0 = constant, β₁x₁ = lead poisoning, LP, β₂x₂ = out-of-home placement, OP, β₃x₃ = abuse and neglect, AN; β₁x₁ * β₂x₂, and β₂x₂ * β₃x₃ are interaction terms that were applied to see how they affected the models (in comparison).

The first model was assessed based on an alpha of 0.05. The overall model was significant, χ2(4) = 49.82, p < 0.001, showed that LP, OP, and AN had a significant effect on the odds of observing high level of AC. McFadden’s R-squared was applied in assessing the model fit. The McFadden R-squared value calculated for this model was 0.41. The effect of the LP was not significant, B = 0.25, OR = 1.28, p = 0.067, which indicates that LP did not have a significant effect on the odds of observing the 1 category of AC. The effect of the OP was not significant, B = 0.01, OR = 1.01, p = 0.091, indicating that OP did not have a significant effect on the odds of observing the 1 category of AC. The effect of the AN was significant, B = 0.04, OR = 1.05, p = 0.037, indicating that a one-unit increase in AN increase the odds of observing the 1 category of AC by approximately 4.60%. The interaction between LP and OP was not significant, B = −0.0003, OR = 1.00, p = 0.077, indicating that a one-unit increase in LP does not change the effect of OP on the odds of observing the 1 category of AC. Table 4 summarizes the results of the model.

Variable	B	SE	χ^2	P	OR	95.00% CI
(Intercept)	−2.65	0.59	19.90	<0.001	—	—
LP	0.25	0.14	3.36	0.067	1.28	[0.98, 1.67]
OP	0.01	0.008	2.86	0.091	1.01	[1.00, 1.03]
AN	0.04	0.02	4.35	0.037	1.05	[1.00, 1.09]
LP: OP	−0.0003	0.0002	3.12	0.077	1.00	[1.00, 1.00]

Table 4.

Logistic regression results with LP, OP, and AN predicting AC.

Note. χ²(4) = 49.82, p < 0.001, McFadden R² = 0.41. Note: Some of the fitted probabilities for this model were near o or 1. For this reason, the interpretation of this analysis is exploratory.

For the second model (replication with different interaction terms), they mimicked that of the first model with an alpha of 0.05. The overall model was significant, χ2(4) = 49.95, p < 0.001, suggesting that OP, LP, and AN had a significant effect on the odds of observing the higher category (1) of AC. McFadden’s R-squared was used to assess the model fit. Values that are 0.2 are indicative of models with good fit [49]. The McFadden R-squared value calculated for this model was same as the original model, 0.41.

5. Discussion

The present study has explored how prevalence of AN, OP, and LP were linked to prevalence of arrest for serious offenses, including the most serious forms of violence (murder, rape, and aggravated assault). Bivariate and multivariate analyses were applied in exploring such relationships. In discussing the findings of this study, it is important to highlight that the findings for both the bivariate and multivariate analyses were re-examined through bootstrapping of 2000 samples (resampling). This approach was applied due to the limited number of cases that were involved in the original analysis. Findings from this study highlight important considerations and variables that should be looked further into, in terms of violence prevention.

Findings from the point biserial correlations were examined with Holm correction to adjust for multiple comparisons based on a probability value of 0.05 for the raw data without bootstrapping. There was a significant positive correlation between AC and LP with a correlation coefficient of 0.50 (p < 0.001, 95.00% CI [0.32, 0.64]), indicating a large effect size. This suggests that moving from the 0 category to the 1 category of AC is associated with an increase in LP. Therefore, the 1 category of AC tends to be associated with higher values of LP. There was also a significant positive correlation between AC and AN with a correlation coefficient of 0.61 (p < 0.001, 95.00% CI [0.46, 0.73]), indicating a large effect size. This suggests that moving from the 0 category to the 1 category of AC is associated with an increase in AN. Therefore, the 1 category of AC tends to be associated with higher values of LOG_AN. There was a significant positive correlation between AC and OP with a correlation coefficient of 0.60 (p < 0.001, 95.00% CI [0.45, 0.72]), indicating a large effect size. This suggests that moving from the 0 category to the 1 category of AC is associated with an increase in OP. Therefore, the 1 category of AC tends to be associated with higher values of OP. Table 4 presents the results of the correlations.

The analysis with the bootstrapped sample: The result of the bootstrapped point biserial correlations was examined based on a probability value of 0.05. There was a significant positive correlation between AC and LP (N = 2000) with a correlation coefficient of 0.50 (SE = 0.07, 95.00% CI [0.36, 0.63]), indicating a moderate effect size. This indicates that moving from the 0 to 1 category of AC is associated with an increase in LP. Therefore, the 1 category of AC tends to be associated with higher values of LP. There was a significant positive correlation between AC and AN (N = 2000) with a correlation coefficient of 0.61 (SE = 0.06, 95.00% CI [0.48, 0.72]), indicating a large effect size. This indicates that moving from the 0 to 1 category of AC is associated with an increase in AN. Therefore, the 1 category of AC tends to be associated with higher values of AN. There was a significant positive correlation between AC and OP (N = 2000) with a correlation coefficient of 0.60 (SE = 0.06, 95.00% CI [0.48, 0.71]), indicating a large effect size. This indicates that moving from the 0 to 1 category of AC is associated with an increase in OP. Therefore, the 1 category of AC tends to be associated with higher values of OP. Table 3 presents the observed correlations, along with the bootstrapped results for the standard error and the 95.00% confidence interval of each correlation.

While findings from the point biserial correlation analysis showed strong associations between OP and AC, indicating that while OP increased, arrest for serious crime also increased, a further analysis (logistic regression) showed that prevalence of OP did not have a significant effect on the odds of observing higher prevalence of AC. This finding is slightly consistent with previous findings. For example, [50] found that OP only had slight positive association between child abuse and criminal offending during adulthood. The authors had rather found that placement inconsistency was more influential than just out-of-home placement. Also, Font et al. [51] found that only 13% of criminal offenders had been placed in foster care in a comparison study the authors conducted, recently. In fact, [50] findings as well as that of Hall et al. [52] may expose some reasons OP in isolation, might not be a significant predictor of AC. In support of this, Hall et al. [52] found that adverse childhood experiences were associated with OP of individuals who had engaged in sexually violent behaviors. It is also important to note that a previous study has highlighted high prevalence of abuse among children placed in out-of-home situations [53], which may further exacerbate criminal offending and violence perpetration among the children. In short, it does seem that the pathway to serious crime through OP is through other variables which could include abuse that occurred while in OP.

While there was strong positive correlation between LP and AC, indicating that increase in prevalence of LP is linked to AC, further analysis through logistic regression showed that LP was not a predictor of AC. This finding is interesting because it is established in the literature that LP is linked with poor academic performance and delinquency [54] as well as neurological and behavioral challenges. In fact, a recent study [55] had found that lead concentrations in the human body (after age 6) had arrest consequences. Wright et al. [55] did not find a link between lead poisoning in childhood and violence, which is comparable to the findings of Beckley et al. [56] but contrary to Nevin [57].

The most important statistically significant finding was between AN and AC. The correlation analysis showed strong positive correlation between AN and AC. Specifically, the finding showed that increase in AN was associated with increase in AC. In other words, increase in prevalence of AN was associated with increase in prevalence of AC.

Regarding the logistic regression, for the non-bootstrapped model, the effect of the LP was not significant, B = 0.25, OR = 1.28, p = 0.067, which indicates that LP did not have a significant effect on the odds of observing the 1 category of AC. The effect of the OP was not significant, B = 0.01, OR = 1.01, p = 0.091, indicating that OP did not have a significant effect on the odds of observing the 1 category of AC. The effect of the AN was significant, B = 0.04, OR = 1.05, p = 0.037, indicating that a one-unit increase in AN increases the odds of observing the 1 category of AC by approximately 4.60%. The interaction between LP and OP was not significant, B = −0.0003, OR = 1.00, p = 0.077, indicating that a one-unit increase in LP does not change the effect of OP on the odds of observing the 1 category of AC. Table 4 summarizes the results of the model.

For the bootstrapped model of the logistic regression, the effect of the OP was not significant, B = 0.01, OR = 1.01, p = 0.110, indicating that OP did not have a significant effect on the odds of observing the 1 category of AC. The effect of the LP was not significant, B = 0.24, OR = 1.27, p = 0.073, indicating that LP did not have a significant effect on the odds of observing the 1 category of AC. The effect of the AN was significant, B = 0.05, OR = 1.05, p = 0.032, indicating that a one-unit increase in AN increase the odds of observing the 1 category of AC by approximately 4.83%. The interaction between LP and AN was significant, B = −0.0006, OR = 1.00, p = 0.006, indicating that a one-unit increase in LP changes the effect of AN on the odds of observing the 1 category of AC. Table 3 summarizes the results of the regression model.

Studies have documented links between different forms of child abuse and neglect and externalizing problems and other forms of crimes or correlates of criminal offending such as poor school performance and mental health and substance use issues [16, 17, 18, 24, 26, 27, 28, 58]. In fact, Grafet al.’s [18] recent findings uncovered causal links between child abuse and criminal offending. It is important to note that, while a few indications of causal links between child abuse and neglect and criminal offending are present, the non-direct pathways through which abuse can associate with criminal offending, for example, through externalizing problems [23] are also worth considering. In fact, that child abuse and neglect is a predictor of subsequent criminal offending as an adult is established in the literature [19, 29, 30, 31] and that it is the significant predictor of arrest for serious crime including violence, highlights the social learning consequences of aggression and maltreatment in the home setting [40], thereby justifying the theoretical reasoning for this study. In the end, one of three main hypotheses of the study were supported, thereby accentuating the need to lay emphasis on prevention of violence and maltreatment of children.

5.1 Mental health implications

Findings from this study speak to the combined importance of addressing crime and violence from environmental [7], policy, criminogenic [1, 2], and mental health [4] standpoints. Findings from this study and existing studies on the topic are a call for action to the mental health community, especially at the county level. All aspects of child abuse and neglect victimization, some reasons for out-of- home placement, and some reasons for arrest for serious offenses [1, 2, 3] have been linked with mental health problems [4]. These are also connected to the wellbeing of children and adults in terms of their education and quality of life. Ghaderi et al.’s [59] study’s highlight of a link between cognitive decline and lead poisoning is a direct connection between lead poisoning and mental health issues because cognitive decline reduces quality of life. Moreso, the connection is also highlighted in Mbonane et al. [5] wherein such could lead to aggressive tendencies. As victims of abuse and neglect have been found to have a high propensity to re-enact abuse or be engaged in crime [18, 21, 23, 60], it is important to bear in mind that mental health issues could play out in the form of aggression and crime [16, 17, 18, 24, 26, 27, 28, 58] . It is therefore important to consider the milieu and environment where children and adolescents live in making policies that address safety and wellbeing. Given the transgenerational burden that may be associated with poverty (which can be a reason for lead poison exposure) and other negative outcomes, it is important to also engage and continue to highlight the person-in-environment perspective from which social workers operate, in enacting policies, and in allocating funds to address preventive measures such as crime prevention—to reduce the counts of abuse and neglect, lead poisoning, out-of-home placement, and arrest for serious crimes at the county levels. Such thinking, in addition to including other areas of the human ecology, will make way for a better future.

Again, findings from this study uncovered important considerations and variables that should be looked further into, in terms of violence prevention. Specifically, the relationships between child abuse and neglect and arrest for serious crimes are important. This finding speaks to the need to pay close attention to the criminogenic consequences of child abuse and violence in children. At the community and policy levels, findings from this study accentuate a need to examine and enhance child protective efforts at the county levels in terms of decreasing its prevalence by adapting evidence-based measures. In doing so, violence against children might be minimized as well as the current challenge of the childhood prison pipeline. Findings from this study emphasize the need for social workers in practice, research, and education, to, in concert with other disciplines and agencies, and come up with evidence-based means to decrease the prevalence of child abuse and neglect. It is also important to point out certain limitations that are associated with this study. First, the study is exploratory in nature, given its sample size. The data that was used is also somewhat dated but was used in this analysis because it is the available, complete data, given the variables that were of interest in this study. To attend to the small sample size, bootstrapping was applied. Yet, findings from the logistic regression, although with two models, ought to be interpreted with caution, given that some of the fitted probabilities were close to zero or 1. With these in mind, the current study underscores a need to continue to explore the combined relationships of the variables, especially among social work educators.

6. Conclusion

The links between mental illness, child abuse and neglect, including lead exposure, and involvement in serious criminal activities call for continuous attention. This is because the health and wellbeing of any society largely depends on its youth. This study explored uniquely, the combined relationships of key variables that are important to study at the county level, given that a county has many resources and some level of authority to administer interventions of safety within its jurisdiction. The study uncovered several interesting findings, prominent of which is that abuse and neglect was associated with crime involvement. This study provides an additional voice to the call to end child abuse and neglect and to begin to build new pathways for child and youth transition to adulthood. It is also a call to adopt policies that are capable of preventing and mitigating pathways to crime, mental illness, child abuse and neglect, and other preventable adverse childhood experiences. It is also a call for policy analyst to become more interested in using models that are comprehensive in doing such. One of the models that canvasses for a comprehensive approach is the multiphasic policy analysis model [61] which has been applied in evaluating an international program that is aimed at addressing the needs of vulnerable children. The call to action following this exploration is not only to policymaker: it is for educators, police officers, child protection workers, civil organizations, family systems, and other agencies of government. Simply put, efforts should continue to be made to ensure that children and adolescents are protected against all forms of abuse and neglect including exposure to environments that may have cognitive and other mental health consequences on them.

Conflict of interest

The author has declared no conflict of interest in this study.