Differentiating Perpetrators of Intimate Partner Violence From Other Violent Offenders Using a Statistical Learning Model: The Role of Cognition and Life History Variables

Abstract

Intimate partner violence (IPV) is a widespread crime that victimizes over 4-million women per year in the United States and results in significant monetary cost and unmeasured physical and psychological consequences for victims. Specialized IPV offender treatment programs demonstrate limited effectiveness, which may be due to an insufficient understanding of the factors that differentiate between IPV perpetrators and non-IPV violent offenders. In this study, we utilized classification and regression tree (CART) analysis to identify combinations of factors that best discriminate IPV perpetrators from non-IPV violent offenders. We also compared cognitive abilities between IPV perpetrators and non-IPV violent offenders using standardized neurocognitive tests. CART analysis presented two pathways for identifying offenders as IPV perpetrators: (a) extensive nonviolent criminal history and (b) moderate-to-severe expression of interpersonal traits of psychopathy without attentional deficits. In addition, a third pathway identified non-IPV violent offenders: (c) low levels of interpersonal psychopathic traits and no history of neurodevelopmental diagnosis. IPV perpetrators demonstrated intact cognition relative to test norms, and study groups did not significantly differ on cognitive performance. These findings suggest that individuals with multiple arrests for nonviolent crime or individuals with interpersonal traits of psychopathy without attentional difficulties may be at enhanced risk for IPV perpetration.

Keywords

intimate partner violence neuropsychological testing psychopathy statistical learning model

Introduction

According to a report published by the World Health Organization, violence against women “has reached epidemic proportions in many societies” (World Health Organization, 2005). In the United States alone, over 4-million women are physically assaulted by an intimate partner every year, and approximately one third of all women report having experienced intimate partner violence (IPV) in their lifetime (Smith et al., 2017).

Beyond being a common crime, IPV results in physical and psychological difficulties for victims and significant monetary costs to societies. It is well documented that IPV is associated with subsequent injury, trauma, and death for victims from IPV-related injuries (Kernic et al., 2000; Plichta, 2004; Rennison & Welchans, 2000). Compared with women who never experienced IPV, victims report elevated rates of physical and mental health problems and are more likely to utilize health care services (Plichta, 2004; Smith et al., 2017). In the United States, IPV costs $5.8 billion per year in health services in addition to the costs of IPV-related absences from work (Centers for Disease Control and Prevention [CDC], 2003). Given its widespread prevalence, as well as the high physical, psychological, and monetary costs associated with IPV, systematic study of IPV perpetrators is imperative to enhance prevention efforts and optimize treatment for known offenders.

Current treatment methods for IPV include cognitive behavioral therapy (CBT) or a feminist approach (typically based on the Duluth model; Babcock et al., 2004; Pence & Paymar, 1993). However, these treatment programs have had limited effectiveness in curbing reoffending (Babcock et al., 2004; Day et al., 2009). One reason for this may be an insufficient understanding of the factors that differentiate between IPV perpetrators and generally violent offenders, which precludes the development of tailored programs for IPV offenders. There is some indication from prior research that IPV perpetrators may represent a distinct criminological group from non-IPV violent offenders, though findings in this area have been mixed. Some studies have found that IPV perpetrators and non-IPV violent offenders differ on several demographic and childhood variables, with IPV perpetrators demonstrating higher levels of education but also higher rates of learning disability (Olson & Stalans, 2001; Stewart & Power, 2014). Other studies have found that the groups also differ on psychiatric and substance use variables in that IPV perpetrators are more likely to have histories of childhood abuse and mental health treatment, have higher levels of psychopathic personality traits and depression, and use more illegal substances than non-IPV violent offenders (Felson & Lane, 2010; Stewart & Power, 2014; Theobald et al., 2016). Finally, one study found IPV perpetrators to have more extensive criminal histories than non-IPV violent offenders (Stewart & Power, 2014). In contrast, other studies found no statistically significant differences between IPV perpetrators and non-IPV violent offenders in demographic, life history, psychiatric, substance use, and criminological variables (Mowat-Leger, 2003; Olson & Stalans, 2001).

Prior research has suggested that cognitive ability is another factor that may distinguish IPV perpetrators from non-IPV violent offenders (Hanlon et al., 2015). One study compared executive functioning in incarcerated IPV perpetrators and other (violent and nonviolent) offenders, and found reduced mental flexibility, but better response inhibition in the IPV group (Bueso-Izquierdo et al., 2016). Several other studies compared community-based samples of IPV perpetrators to noncriminal controls and found that IPV perpetrators performed more poorly on measures of attention, working memory, executive functions, and verbal skills (Cohen et al., 2003; Holtzworth-Munroe et al., 2000; Teichner et al., 2001). However, no prior research to our knowledge has directly compared IPV and non-IPV violent offenders on performance-based cognitive measures.

Prior studies have linked several other individual characteristics with IPV. Specifically, a significantly higher proportion of IPV perpetrators have low socioeconomic status compared with nonviolent community controls (Reichel, 2017). In terms of neurologic characteristics, several studies have found that IPV perpetrators exhibit elevated rates of head injury compared with samples of healthy controls (Rosenbaum & Hoge, 1989; Rosenbaum et al., 1994). However, to our knowledge, no prior studies have directly compared IPV perpetrators to non-IPV violent offenders on these variables.

In summary, there is preliminary indication from prior research that IPV perpetrators as a group may differ from non-IPV violent offenders on certain life history characteristics, as well as in cognitive ability. However, findings remain mixed and suffer from poor replicability. The variability in findings may be explained by two factors. First, many studies have utilized nonviolent or mixed violent/nonviolent offender control groups, which makes it difficult to parse individual differences unique to IPV perpetrators from those of violent offenders in general. Also, some studies that did utilize violent offenders as controls used the index offense to classify group membership. Group assignment based on the index offense without regard to prior offenses can result in contamination of the comparison group by including offenders with prior histories of IPV. Second, prior studies have utilized linear models to differentiate between offender groups, which assume that all variables contribute to the distinction between groups in an equally additive fashion despite the well-accepted notion that individual differences do not always contribute to real-world behavioral outcomes in an equally additive fashion (Steadman et al., 2000).

The Current Study

The primary goal of this study was to identify a set of demographic, developmental, psychiatric, neurologic, substance use, criminological, and cognitive ability variables unique to IPV perpetrators, while addressing the methodological limitations of prior research. We classified group membership based on any history of IPV as opposed to focusing solely on the index offense to ensure that the comparison group included only offenders with histories of violent crime other than IPV. To avoid the reliance on linear models to assess group differences, we utilized Classification and Regression Tree (CART) analysis—a statistical learning technique designed to build more accurate models for real world classification of human behavior (Steadman et al., 2000). Several prior studies have utilized both CART and logistic regression to predict IPV reoffending and have found that CART models outperform logistic regression models in predicting reoffending (Brewer, 2012; Neuilly et al., 2011; Sledjeski et al., 2008). No prior study to our knowledge has utilized CART to determine the combination of variables that best differentiates between IPV and non-IPV violent offenders.

Three sets of analyses were carried out to accomplish study goals. First, we examined cognitive performance of IPV offenders relative to published test norms. Based on prior research, we hypothesized that incarcerated IPV perpetrators should score in the impaired range on measures of attention, working memory, executive functions, and verbal skills (Cohen et al., 2003; Holtzworth-Munroe et al., 2000; Teichner et al., 2001). Second, we compared the cognitive profile of IPV perpetrators to that of non-IPV violent offenders using independent t-tests or Wilcoxon Rank–Sum tests. Based on findings from prior research, we hypothesized that IPV perpetrators would demonstrate poorer mental flexibility than non-IPV violent offenders (Bueso-Izquierdo et al., 2016). Third, we utilized a statistical learning model, a CART analysis, to identify the combination of life history and cognitive variables that best discriminate between IPV perpetrators and non-IPV violent offenders. We included demographic, psychiatric, substance use, neurologic, and criminological variables that may differentiate between IPV perpetrators and generally violent offenders in the model (Felson & Lane, 2010; Olson & Stalans, 2001; Stewart & Power, 2014; Theobald et al., 2016). We did not have a priori hypotheses regarding the combination of variables that best discriminate between groups as this analysis was exploratory in nature.

Method

Participants and Procedures

Participants were 57 males with prior histories of IPV and 42 males with prior histories of only non-IPV violent crimes, aged 18 or older. All participants were inmates at a county jail near a major Midwestern city. Inclusion criteria were the following: (a) no history of stroke, brain tumor, or epilepsy; (b) no active psychiatric symptoms; and (c) English as first and primary language. Participants were chosen from a larger, randomly selected pool of inmates who had previously completed a baseline interview including a psychopathy assessment. Following the administration of written consent, participants completed a life history interview and a battery of neuropsychological tests administered by a trained graduate student. History of violent crime was collected from official files and life history interviews. For the file review, we recorded the number of charges and convictions for any violent crime from participants’ pretrial investigation files. Crimes were classified as violent using the Illinois criminal code definition (“Illinois Criminal Code,” 2012) and included kidnapping, sex offenses, bodily harm, intimate partner violence, robbery, burglary, aggravated arson, deadly weapons, and mob action. In addition to collecting arrest and conviction data, we asked each participant how many times he engaged in any of the above crimes, to account for violent offenses that did not result in a criminal charge or conviction. All participants were also asked if they ever physically attacked or made a death threat involving a weapon to a female partner to aid in determining group status. To control for reporter bias, participants were provided with simplified written definitions for each violent crime, per the Illinois criminal code.

Participants were classified as IPV offenders if they were ever charged with or convicted of an intimate partner violence offense per criminal records or if they self-reported a physical incident involving a current spouse, current nonmarital partner, former marital partner, or former nonmarital partner during the life history interview. This definition of IPV is empirically validated and used in criminology research and by the Federal Bureau of Investigation (FBI; Saltzman et al., 1999; United States Department of Justice [DOJ], 2017). Participants were classified as non-IPV violent offenders if they had a history of a violent crime other than IPV either by criminal record review or self-report. The institutional review board at Northwestern University Feinberg School of Medicine approved the study procedures, and all participants provided written informed consent.

Measures

The life history interview was designed to collect information regarding the demographic, developmental, psychiatric, neurologic, and criminological variables which have demonstrated significant associations with violence in the literature (see Table 1 for the complete list).

Table 1.

Demographic, Childhood, Psychiatric, Neurologic, Alcohol Use, and Criminal Characteristics of IPV Perpetrators and Non-IPV Violent Offenders.

Characteristics	Violent Offenders
	Whole Sample (n = 99)	IPV (n = 57)	Non-IPV (n = 42)	IPV vs. Non-IPV
	M (SD)	M (SD)	M (SD)	W Statistic
Demographics
Age	32.4 (8.2)	34.3 (8.1)	29.8 (7.8)	—
Years of education	11.2 (2.0)	11.3 (1.8)	11.1 (2.1)	1,148.00
Number of children	2.4 (2.0)	3.0 (2.2)	1.7 (1.6)	760.00**
Parental socioeconomic status	53.6 (19.5)	51.3 (16.0)	56.6 (23.4)	1,302.00
	N (%)	N (%)	N (%)	Fisher’s Exact p
Race
Caucasian	18 (18.2)	10 (17.5)	8 (19.0)	.09⁺
African American	64 (64.5)	41 (71.9)	23 (54.8)
Hispanic	12 (12.1)	3 (5.3)	9 (21.4)
Other	5 (5.1)	3 (5.3)	2 (4.8)
Ever married/common law partner	86 (86.9)	54 (94.7)	32 (76.1)	.01*
	N (%)	N (%)	N (%)	χ² Statistic
Childhood characteristics
Special education	55 (55.6)	26 (45.6)	19 (45.2)	<0.01
Childhood aggression	65 (65.7)	38 (66.7)	27 (64.3)	<0.01
Arrested under age of 16	64 (64.6)	36 (63.2)	28 (66.7)	0.02
Lived with both bio parents to 16	24 (24.2)	15 (26.3)	9 (21.4)	0.11
Elementary school maladjustment	67 (67.7)	37 (64.9)	30 (71.4)	0.22
	N (%)	N (%)	N (%)	Fisher’s Exact p
Suspended/expelled from school	90 (90.9)	54 (94.7)	36 (85.7)	.16
	N (%)	N (%)	N (%)	χ² Statistic
Psychiatric history
Neurodevelopmental disorder	34 (34.3)	22 (38.6)	12 (28.6)	0.68
Intellectual disability	0 (0.0)	0 (0.0)	0 (0.0)	—
ADHD	24 (24.2)	14 (24.6)	10 (23.8)	<0.01
	N (%)	N (%)	N (%)	Fisher’s Exact p
Learning disorder	16 (16.2)	12 (21.1)	4 (9.5)	.17
Schizophrenia	3 (3.0)	3 (5.3)	0 (0.0)	.26
	N (%)	N (%)	N (%)	χ² Statistic
History of conduct disorder	74 (74.7)	39 (68.4)	35 (83.3)	2.11
Antisocial personality disorder	70 (70.7)	40 (70.2)	30 (71.4)	<0.01
Psychiatric treatment	61 (61.6)	34 (59.6)	27 (64.3)	0.07
	M (SD)	M (SD)	M (SD)	W Statistic
Psychopathy (PCL-R total)	26.0 (5.8)	27.0 (5.8)	24.6 (5.5)	850.00*
Interpersonal psychopathy traits	4.7 (1.8)	5.0 (1.7)	4.2 (1.8)	827.00**
Affective psychopathy traits	5.5 (1.7)	5.7 (1.8)	5.2 (1.6)	920.50⁺
Lifestyle psychopathy traits	6.5 (1.9)	6.7 (1.8)	6.3 (1.9)	1,034.50
Antisocial psychopathy traits	7.3 (1.9)	7.5 (1.8)	7.1 (2.1)	1,057.00
	N (%)	N (%)	N (%)	χ² Statistic
Alcohol use
Teenage alcohol use disorder	31 (31.3)	14 (24.6)	17 (40.5)	2.16
Adult alcohol use disorder	57 (57.6)	36 (63.2)	21 (50.0)	1.22
Parental alcoholism	38 (38.4)	21 (36.8)	17 (40.5)	<0.01
	N (%)	N (%)	N (%)	χ² Statistic
History of abuse (sexual or physical)	28 (28.3)	17 (29.8)	11 (26.2)	0.03
Physical	27 (27.2)	17 (29.8)	10 (23.8)	0.19
	N (%)	N (%)	N (%)	Fisher’s Exact p
Sexual	5 (5.1)	4 (7.0)	1 (2.4)	.39
	N (%)	N (%)	N (%)	χ² Statistic
Neurologic history
History of head injury	68 (68.7)	39 (68.4)	29 (69.0)	<0.01
	M (SD)	M (SD)	M (SD)	W Statistic
Number of TBI	3.4 (10.5)	3.6 (13.1)	3.1 (5.4)	1,252.00
	N (%)	N (%)	N (%)	χ² Statistic
Severity of TBI
Mild	60 (60.6)	34 (59.6)	26 (61.9)	<0.01
	N (%)	N (%)	N (%)	Fisher’s Exact p
Moderate	13 (13.1)	8 (14.0)	5 (11.9)	>.99
Severe	3 (3.0)	2 (3.5)	1 (2.4)	>.99
	M (SD)	M (SD)	M (SD)	W Statistic
Criminal characteristics
Time spent incarcerated (months)	76.3 (61.3)	83.4 (64.0)	66.8 (57.0)	1,021.00
Juvenile	10.4 (14.5)	10.2 (15.1)	10.7 (13.7)	1,270.00
Adult	65.9 (59.8)	73.1 (60.6)	56.1 (58.0)	965.50
	N (%)	N (%)	N (%)	χ² Statistic
Failure on prior conditional release	81 (81.8)	48 (84.2)	33 (78.6)	0.21
	M (SD)	M (SD)	M (SD)	W Statistic
Extent of crime (total)	90.6 (78.8)	97.1 (79.6)	81.7 (77.7)	972.00
Extent of violent crime	61.5 (79.5)	62.2 (79.0)	60.6 (81.1)	1,028.00
Extent of non-violent	29.1 (22.1)	34.9 (24.3)	21.2 (15.5)	784.00**

Note. PCL-R = Psychopathy Checklist–Revised; IPV = intimate partner violence; TBI = traumatic brain injury; ADHD = attention-deficit/hyperactivity disorder.

⁺p < .10. *p < .05. **p < .01.

Parental socioeconomic status rating

was calculated using Hollingshead’s two-factor model which uses a combination of educational and occupational achievement to arrive at a numerical rating of social status (Hollingshead & Redlich, 1958), with greater values indicating higher status.

Psychiatric history and alcohol use disorder

Diagnoses of schizophrenia, alcohol use disorder, conduct disorder, and antisocial personality disorder were rated according to Diagnostic and Statistical Manual of Mental Disorders (5th ed.; DSM-5; American Psychiatric Association, 2013) criteria. Participants were also asked if they had been diagnosed with a neurodevelopmental disorder including intellectual disability, attention deficit hyperactivity disorder, and learning disability.

Psychopathy

The Psychopathy Checklist–Revised (PCL-R; Hare, 1991) includes 20 items each rated on an ordinal scale (0–2), and is the most extensively validated clinical measure of psychopathic personality traits in institutionalized samples. Item ratings were made on the basis of a semi-structured interview and review of institutional files. The measure yields a total score and four facet scores measuring the interpersonal (e.g., glibness, conning), affective (e.g., shallow affect, lack of empathy), lifestyle (e.g., impulsivity, irresponsibility), and antisocial (e.g., serious juvenile delinquency, criminal versatility) features of psychopathy (Hare, 1991). Psychopathy ratings were made by clinical psychology graduate students who received extensive training in PCL-R assessment. Inter-rater reliability for PCL-R scores was high (average intraclass r = .83, one-way random effects model, n = 12 pairs of raters).

Traumatic brain injury (TBI)

We rated the number and severity of head injuries from the life history interview. Each head injury was classified as mild, moderate, or severe in accordance with DSM-5 criteria.

Cognitive assessments

As presented in Figure 1, participants completed a comprehensive battery of neuropsychological measures used to assess performance in five cognitive domains (Delis et al., 1987; Reitan, 1958; Strauss et al., 2006; Wechsler, 2008; Wilkinson, 2006). Per clinical guidelines (Lezak et al., 2012), we used age-adjusted standardized test scores to measure cognitive ability. To ease comparison between tests, we linearly transformed all original standardized test scores to z-scores (M = 0, SD = 1) and calculated the mean z-score for each cognitive test. To calculate domain scores, we averaged the z-scores for the tests within each domain. Mean scores were characterized as impaired if z < –1.33, according to established practice guidelines (Lezak et al., 2012). To screen for adequate test taking effort, we used two embedded measures of performance validity: Forced Choice Recognition on the California Verbal Learning Test-II (CVLT-II; Delis et al., 1987) and the Reliable Digit Span on the Wechsler Adult Intelligence Scale–Fourth Edition (WAIS-IV; Wechsler, 2008).

Figure 1.

Cognitive assessment battery.

Statistical Analyses

We approached the question of variables that differentiate between IPV perpetrators and non-IPV violent offenders comprehensively by conducting both bivariate analyses and a statistical learning model, CART. First, we examined between-group differences using t-tests or, when assumptions of t-tests were violated, using Wilcoxon Rank–Sum tests and the corresponding effect size statistics. In addition, we used CART analysis from the Recursive Partitioning and Regression Trees script in R to identify the combination of independent variables that best differentiated between IPV perpetrators and non-IPV violent offenders (Therneau et al., 2015). CART analysis grows a tree by utilizing Bayesian statistics to split the predictor variables into nodes that maximize homogeneity among the data (Breiman et al., 1984). When further splitting of the data can no longer be justified by the increase in the variance explained, the CART analysis produces a terminal node, which predicts the outcome variable. The major limitation of CART analysis is that a single CART tree can be unstable (Breiman, 1996). To minimize this instability, we followed up the CART analysis by growing a random forest to evaluate how the model performed on different subsets of the data through Out-of-the-Bag sampling for each tree grown (Breiman, 2001). See Berk (2008) for additional technical details on the random forest method. For analyses, we used multiple imputation to compute missing values in the data (Rubin, 1987). Multiple imputation is recommended over alternative methods, because it utilizes data distributions to estimate missing values (Vaden et al., 2012).

Results

Descriptive Characteristics of IPV Perpetrators Compared With Non-IPV Violent Offenders

As shown in Table 1, overall the sample was comprised of young, primarily African American men who had partial high school education and history of conduct disorder. IPV perpetrators had significantly more children (W = 760.0, p < .01, r = –0.32) and were more likely to have been in at least one marital-type relationship (Fisher’s exact p = .01, p < .05, Cohen’s w = 0.27) than non-IPV violent offenders. In addition, IPV perpetrators demonstrated significantly higher overall psychopathy scores (W = 850.0, p < .05, r = –0.25), and higher interpersonal facet scores (W = 827.00, p < .01, r = –0.27) but the groups did not significantly differ on levels of the affective, lifestyle, or antisocial traits of psychopathy. Finally, IPV perpetrators committed significantly more non-violent crimes than non-IPV violent offenders (W = 784.0, r = –0.29, p < .01). All between-groups effect sizes were small with the exception of number of children which was associated with a medium effect size. IPV perpetrators and non-IPV violent offenders did not significantly differ on any additional characteristics (see Table 1).

Cognitive Profiles of IPV Perpetrators and Non-IPV Violent Offenders

All participants demonstrated sufficient test taking effort on embedded indices of performance validity. As presented in Figure 2, both groups performed in the low average range across all tests compared with published normative data. The mean overall intelligence quotient was 84.14 for the IPV group, and 85.74 for non-IPV violent offenders—a nonsignificant difference. As presented in Table 2 and Figure 2, there were no significant differences in cognitive performance between IPV perpetrators and non-IPV violent offenders across measures in any cognitive domain.

Figure 2.

Intimate partner violence perpetrators versus non-IPV violent offenders: cognitive test performance.

Table 2.

Cognitive Test Performance: Differences Between IPV Perpetrators and Non-IPV Violent Offenders.

Cognition	IPV (n = 57)	Non-IPV (n = 42)	Test Statistic
Cognition	M (SD)	M (SD)	t Stat (Cohen’s d) or W Stat (r)
	Intellectual functions (WAIS)
Full scale IQ	–1.06 (0.63)	–0.95 (0.79)	–0.72 (–0.15)
General ability index	–0.99 (0.75)	–0.90 (0.81)	–0.52 (–0.11)
Verbal comprehension	–1.05 (0.75)	–1.07 (0.80)	0.09 (0.02)
Perceptual reasoning	–0.80 (0.79)	–0.54 (0.91)	–1.02 (–0.21)
Working memory	–0.90 (0.71)	–0.73 (0.87)	1,097.00 (–0.07)
Processing speed	–0.78 (0.64)	–0.72 (0.77)	1,175.50 (–0.02)
	Attention, working memory, and processing speed
Digit span total (WAIS)	–0.66 (0.76)	–0.52 (0.87)	1,119.00 (–0.06)
Arithmetic (WAIS)	–0.92 (0.65)	–0.78 (0.82)	1,099.00 (–0.07)
Coding (WAIS)	–0.59 (0.76)	–0.60 (0.85)	1,245.00 (–0.03)
Symbol search (WAIS)	–0.87 (0.67)	–0.73 (0.73)	1,077.00 (–0.09)
Trail making test Part A	–0.16 (1.13)	–0.23 (0.93)	1,285.00 (–0.06)
	Reasoning and executive functions
Similarities (WAIS)	–0.81 (0.76)	–0.77 (0.75)	1,158.00 (–0.03)
Matrix reasoning (WAIS)	–0.46 (1.00)	–0.44 (0.98)	1,157.50 (–0.03)
Block design (WAIS)	–0.79 (0.74)	–0.52 (0.99)	937.50 (–0.16)
Trail making test Part B	–0.38 (1.23)	–0.57 (1.59)	1,271.50 (–0.05)
	Anterograde memory functions
Word list (CVLT-II) encoding	–0.81 (1.04)	–0.67 (1.03)	–0.63 (–0.13)
CVLT-II long delayed recall	–0.83 (0.92)	–0.83 (1.00)	1,185.50 (–0.01)
CVLT-II recognition discrimination	–0.39 (1.03)	–0.38 (1.02)	1,168.00 (–0.02)
	Language functions
Vocabulary (WAIS)	–1.12 (0.75)	–1.18 (0.80)	1,286.00 (–0.06)
Word reading (WRAT)	–0.77 (0.65)	–0.57 (0.78)	–1.45 (–0.30)

Note. All scores are z-scores (M = 0, SD = 1). IPV = intimate partner violence; WAIS = Wechsler Adult Intelligence Scale; CVLT-II = California Verbal Learning Test-II; WRAT = Wide Range Achievement Test.

Combination of Factors That Differentiate IPV Perpetrators From Non-IPV Violent Offenders

As presented in Figure 3, the combination of variables and the cut-points that best differentiated between IPV perpetrators and non-IPV violent offenders were identified via CART analysis. In terms of clinical utilization, the tree presented two pathways for identifying offenders as IPV perpetrators. First, offenders with an extensive history of nonviolent charges or convictions (greater than 42 nonviolent crimes) were likely to be IPV perpetrators. Second, for offenders who do not have an extensive history of nonviolent crime, IPV perpetration was likely when an offender presented with a combination of medium to high levels of interpersonal traits of psychopathy (greater than 5.2 on interpersonal facet of the PCL-R) and low average or higher scores on the attention, working memory, and processing speed domain (scores greater than z = –0.86). As shown in Figure 4, the individual tree predicted classification groups in the current sample fairly well with a misclassification rate of 19.2%. Errors were biased toward false positives. Area under the curve (AUC) for the individual tree was strong (AUC = 0.85).

Figure 3.

Classification analysis and regression tree for IPV perpetrators and non-IPV violent offenders.

Figure 4.

Classification statistics and receiver operating characteristic curve for single classification analysis and regression tree.

Figure 5 depicts results from a random forest with 2,000 trees that help identify the most important variables in predicting classification group. Excluding scores on the interpersonal facet of psychopathy resulted in the largest mean decrease in model accuracy, followed by excluding the number of nonviolent crimes, then scores on the attention, working memory, and processing speed cognitive domain, and finally neurodevelopmental diagnosis. As shown in Figure 5, the average tree in the random forest also predicted classification groups in various subsets of the data fairly well with a misclassification rate of 28.3%. For the random forest tree, errors were fairly evenly split between false positives and negatives. The AUC for the random forest was still good (AUC = 0.74).

Figure 5.

Predictor accuracy contributions toward group status via random forest (2,000 trees) and classification statistics and receiver operating characteristic (ROC) curve.

Discussion

This study used a combination of group comparisons and Bayesian statistics (i.e., CART analysis) to investigate which life history and cognitive variables differentiated IPV perpetrators from non-IPV violent offenders. In addition, this study was the first to compare cognitive profiles between these two groups of offenders.

Differentiating IPV Perpetrators From Non-IPV Violent Offenders

According to the CART analysis, IPV perpetrators were distinguishable from non-IPV violent offenders in the current sample with high accuracy. Specifically, the CART analysis predicted IPV perpetrator status through two pathways: (a) the participant had extensive nonviolent criminal history and (b) the participant did not have extensive nonviolent criminal history but had medium to high levels of interpersonal traits of psychopathy and at least low average scores on the attention, working memory, and processing speed domain. The random forest analysis indicated that interpersonal traits of psychopathy and extent of nonviolent history contributed the largest amounts of accuracy to the model. Attention, working memory, and processing speed and neurodevelopmental diagnosis contributed relatively small amounts of accuracy to the model. In addition, random forest analysis indicated that the variables in the CART tree in this study continued to accurately differentiate between IPV perpetrators and non-IPV violent offenders in various subsets of the data.

Currently, the literature provides conflicting findings regarding whether IPV perpetrators represent a distinct criminological group from non-IPV violent offenders (Felson & Lane, 2010; Mowat-Leger, 2003; Olson & Stalans, 2001; Stewart & Power, 2014; Theobald et al., 2016). Through CART, this study provides evidence that IPV perpetrators are distinct from non-IPV violent offenders. Viewing IPV perpetrators as a distinct criminological group from non-IPV violent offenders has important implications for assessment of IPV risk. The tree from this study suggests that certain combinations of historical and cognitive characteristics, including an extensive (nonviolent) criminal career, the interpersonal features of psychopathy, and the absence of measurable attentional impairment may place an individual at risk of committing IPV as opposed to another violent crime. These findings replicate the previously reported associations between IPV and psychopathic personality traits (Harris et al., 2011; Holtzworth-Munroe & Stuart, 1994) and suggest that the impersonal traits of psychopathy such as glibness, grandiosity, deceitfulness, and manipulativeness may be particularly important to identifying IPV offenders. Research suggests that individuals with high scores on the interpersonal trait of psychopathy exhibit decreased physiological fear responses to stimuli, including interpersonal stressors (Babcock et al., 2005; Benning et al., 2005). This decreased fear response regarding interpersonal stressors may increase an individual’s likelihood of specifically committing IPV. This study did not measure IPV as an index crime which limits our ability to comment on directionality of our findings; however, research suggests that interpersonal traits of psychopathy are stable across the lifespan (Harpur & Hare, 1994; Lynam et al., 2009). Therefore, these traits were likely present prior to IPV perpetration and may represent a prospective risk factor for IPV. Future longitudinal studies should investigate whether the combinations of variables identified in this study represent prospective risk pathways specific to IPV.

Cognition in Incarcerated IPV Perpetrators

As a group, the current sample of incarcerated IPV perpetrators performed in the low average range across all cognitive domains and did not demonstrate impairment on any task when referenced against test norms. In addition, incarcerated IPV perpetrators and non-IPV violent offenders did not significantly differ on cognitive performance. Although there are no prior studies that have directly compared groups of IPV and non-IPV violent offenders on neurocognitive function, this finding appears to conflict with one prior study reporting differences between IPV perpetrators and other offenders on measures of executive functioning (Bueso-Izquierdo et al., 2016). However, that study included both violent and non-violent offenders in the non-IPV group, which may have inflated the magnitude of group differences. Clearly, further research comparing cognition in incarcerated IPV perpetrators and other types of offenders is warranted.

That mean intellectual ability of IPV offenders in our sample fell within the low average range relative to published test norms is consistent with intelligence quotient (IQ) estimates reported for incarcerated samples in general (Zamora, 2001). Relatively preserved cognitive ability in incarcerated IPV perpetrators means that, on average, they possess the intellectual capacity to engage in and respond to educational and cognitive restructuring programs as well as other inmates and that the poor treatment outcomes commonly reported for this population (Babcock et al., 2004; Day et al., 2009; Olver et al., 2011) are not due to a cognitive deficit.

Limitations

This study has several limitations. First, this study was based on a small sample of male inmates from one jail in the United States, limiting generalizability to the total population of incarcerated IPV perpetrators. Second, we relied solely on self-report information for many of the life history variables, so participants may have under or over-reported frequency regarding these variables. However, comparison between groups is thought to be accurate. In addition, we only focused on incidents of physical violence toward an intimate partner, but other forms of intimate partner violence exist, including sexual violence and psychological violence. The IPV and non-IPV violent offender groups were based on total history of offending as opposed to offense immediately prior to incarceration. Because of this, we cannot comment on the directionality of our findings, and we did not examine variables specific to relationship functioning, as these variables may have changed over time. Future prospective studies should examine whether relationship functioning variables also differentiate between these two groups.

Conclusion

This study identified combinations of factors that distinguished IPV perpetrators from other violent offenders and established a cognitive profile for incarcerated IPV perpetrators. CART analysis differentiated IPV perpetrators from non-IPV violent offenders with strong accuracy. The most important factors in differentiating between the two groups were interpersonal traits of psychopathy and the extent of an individual’s nonviolent history. Diagnosis of neurodevelopmental disorder and scores on performance-based measures of attention, working memory, and processing speed also contributed toward this differentiation. Follow-up analysis indicated that this model accurately differentiated between these two groups in various subsets of the data. In terms of cognition, bivariate analyses indicated that IPV perpetrators exhibited low average cognitive performance across cognitive domains relative to test norms and did not significantly differ from non-IPV violent offenders. Better understanding of the differences between IPV perpetrators and non-IPV violent offenders can help us identify risk factors specific to IPV and enable earlier enrollment in IPV treatment programs for high risk individuals. Specifically, individuals with extensive arrest history for nonviolent crime may be at enhanced risk for IPV perpetration. In those with less extensive histories of nonviolent offending, the combination of prominent interpersonal traits of psychopathy and preserved attentional functioning may confer increased risk.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Author Biographies

Jaclyn M. Fox, MS, is a graduate student in Northwestern University Feinberg School of Medicine’s clinical psychology PhD program and is completing her clinical internship in neuropsychology at University of California, Los Angeles. Her research focuses on forensic neuropsychology with a specific interest in cognition and intimate partner violence. She is interested in accurately predicting recidivism among IPV perpetrators.

James L. Reilly, PhD, is a clinical neuropsychologist and associate professor of Psychiatry and Behavioral Sciences at Northwestern University Feinberg School of Medicine. His research focuses on the evaluation of neurocognition in patients with or at risk for neuropsychiatric disorders and the effects of pharmacological and behavioral treatment on cognition, with an aim to bridge the basic cognitive and clinical neurosciences.

David S. Kosson, PhD, is professor of Psychology at Rosalind Franklin University of Medicine and Science. His research focuses on cognitive, emotional, interpersonal, developmental, and neural mechanisms underlying psychopathy and has been funded by the National Institute of Mental Health. He is a past president of Society for the Scientific Study of Psychopathy and president of Aftermath: Surviving Psychopathy Foundation.

Allison Brown, MS, is a graduate student in Rosalind Franklin University of Medicine and Science’s clinical psychology PhD program. She is interested in the relationship between emotional and behavioral correlates of externalizing disorders. In particular, her research has investigated the relationship between empathy and aggressive behaviors.

Robert E. Hanlon, PhD, ABPP, is professor of Neurology and Psychiatry and co-director of the Forensic Psychology Research Lab at Northwestern University Feinberg School of Medicine in Chicago. He specializes in traumatic brain injury, psychological effects of trauma, and brain disorders associated with impulsive aggression. He is also a consultant with the Division of Psychiatry and Law at Northwestern University Medical Center and was the director of the Inpatient Neuropsychology Service at Northwestern Memorial Hospital for 18 years. He has served as an expert witness in hundreds of civil and criminal cases.

Michael Brook, PhD, is assistant professor of Psychiatry and Behavioral Sciences at Northwestern University Feinberg School of Medicine, where he directs the Isaac Ray Research Program in Behavioral Sciences and the Law. His research investigates the bio-psycho-social determinants of violence. He is particularly interested in forensic neuropsychology and application of public health research to the study of community violence.

ORCID iD

Jaclyn M. Fox

References

American Psychiatric Association. (2013). Diagnostic and statistical manual of mental disorders (5th ed.). https://doi.org/10.1176/appi.books.9780890425596

Babcock

J. C.

Green

C. E.

Robie

(2004). Does batterers’ treatment work? A meta-analytic review of domestic violence treatment. Clinical Psychology Review, 23(8), 1023–1053. https://doi.org/doi.org/10.1016/j.cpr.2002.07.001

Babcock

J. C.

Green

C. E.

Webb

S. A.

Yerington

T. P.

(2005). Psychophysiological profiles of batterers: Autonomic emotional reactivity as it predicts the antisocial spectrum of behavior among intimate partner abusers. Journal of Abnormal Psychology, 114(3), 444–455.

Benning

S. D.

Patrick

C. J.

Iacono

W. G.

(2005). Psychopathy, startle blink modulation, and electrodermal reactivity in twin men. Psychophysiology, 42(6), 753–762. https://doi.org/10.1111/j.1469-8986.2005.00353.x

Berk

R. A.

(2008). Statistical learning from a regression perspective. Springer.

Breiman

(1996). Bagging predictors. Machine Learning, 24(2), 123–140. https://doi.org/10.1007/bf00058655

Breiman

(2001). Random forests. Machine Learning, 45(1), 5–32.

Breiman

Friedman

Stone

C. J.

Olshen

R. A.

(1984). Classification and regression trees. CRC press.

Brewer

S. L.

Jr. (2012). An empirical comparison of logistic regression to decision tree induction in the prediction of intimate partner violence reassault [Doctoral dissertation], Indiana University of Pennsylvania.

10.

Bueso-Izquierdo

Hidalgo-Ruzzante

Daugherty

J. C.

Burneo-Garcés

Pérez-García

(2016). Differences in executive function between batterers and other criminals. Journal of Forensic Psychology Practice, 16(5), 321–335. https://doi.org/10.1080/15228932.2016.1219216

11.

Centers for Disease Control and Prevention. (2003). Costs of intimate partner violence against women in the United States. National Center for Injury Prevention and Control.

12.

Cohen

R. A.

Brumm

Zawacki

T. M.

Paul

Sweet

Rosenbaum

(2003). Impulsivity and verbal deficits associated with domestic violence. Journal of the International Neuropsychological Society, 9(5), 760–770. https://doi.org/10.1017/S1355617703950090

13.

Day

Chung

O’Leary

Carson

(2009). Programs for men who perpetrate domestic violence: An examination of the issues underlying the effectiveness of intervention programs. Journal of Family Violence, 24(3), 203–212. https://doi.org/10.1007/s10896-008-9221-4

14.

Delis

D. C.

Kramer

J. H.

Kaplan

Ober

B. A.

(1987). CVLT: California Verbal Learning Test—Adult Version, manual. Psychological Corporation.

15.

Felson

R. B.

Lane

K. J.

(2010). Does violence involving women and intimate partners have a special etiology? Criminology, 48(1), 321–338. https://doi.org/10.1111/j.1745-9125.2010.00186.x

16.

Hanlon

R. E.

Brook

Demery

J. A.

Cunningham

M. D.

(2015). Domestic homicide: Neuropsychological profiles of murderers who kill family members and intimate partners. Journal of Forensic Sciences, 61, S163–S170. https://doi.org/10.1111/1556-4029.12908

17.

Hare

R. D.

(1991). The Hare Psychopathy Checklist–Revised: Manual. Multi-Health Systems, Incorporated.

18.

Harpur

T. J.

Hare

R. D.

(1994). Assessment of psychopathy as a function of age. Journal of Abnormal Psychology, 103(4), 604–609.

19.

Harris

G. T.

Hilton

Z. N.

Rice

M. E.

(2011). Explaining the frequency of intimate partner violence by male perpetrators: Do attitude, relationship, and neighborhood variables add to antisociality? Criminal Justice and Behavior, 38(4), 309–331. https://doi.org/10.1177/0093854810397449

20.

Hollingshead

A. B.

Redlich

F. C.

(1958). Social class and mental illness: A community study. John Wiley.

21.

Holtzworth-Munroe

Meehan

J. C.

Herron

Rehman

Stuart

G. L.

(2000). Testing the Holtzworth-Munroe and Stuart (1994) batterer typology. Journal of Consulting and Clinical Psychology, 68(6), 1000–1019. https://doi.org/10.1037/0022-006X.68.6.1000

22.

Holtzworth-Munroe

Stuart

G. L.

(1994). Typologies of male batterers: Three subtypes and the differences among them. Psychological Bulletin, 116(3), 476–497. https://doi.org/10.1037/0033-2909.116.3.476

23.

Illinois Criminal Code of 2012: Summary of Relevant Violent Crimes, § 720 ILCS 5/. (2012).

24.

Kernic

M. A.

Wolf

M. E.

Holt

V. L.

(2000). Rates and relative risk of hospital admission among women in violent intimate partner relationships. American Journal of Public Health, 90(9), 1416–1420. https://doi.org/10.2105/AJPH.90.9.1416

25.

Lezak

M. D.

Howieson

D. B.

Bigler

E. D.

Tranel

(2012). Neuropsychological assessment (5th ed.). Oxford University Press.

26.

Lynam

D. R.

Charnigo

Moffitt

T. E.

Raine

Loeber

Stouthamer-Loeber

(2009). The stability of psychopathy across adolescence. Development and Psychopathology, 21(4), 1133–1153. https://doi.org/10.1017/s0954579409990083

27.

Mowat-Leger

(2003). Risk factors for violence: A comparison of domestic batterers and other violent and non-violent offenders (Accession No. NQ67053) [Doctoral dissertation]. ProQuest Dissertations and Theses. Carleton University.

28.

Neuilly

M.-A.

Zgoba

K. M.

Tita

G. E.

Lee

S. S.

(2011). Predicting recidivism in homicide offenders using classification tree analysis. Homicide Studies, 15(2), 154–176. https://doi.org/10.1177/1088767911406867

29.

Olson

D. E.

Stalans

L. J.

(2001). Violent offenders on probation: Profile, sentence, and outcome differences among domestic violence and other violent probationers. Violence Against Women, 7(10), 1164–1185. https://doi.org/10.1177/10778010122183793

30.

Olver

M. E.

Stockdale

K. C.

Wormith

J. S.

(2011). A meta-analysis of predictors of offender treatment attrition and its relationship to recidivism. Journal of Consulting and Clinical Psychology, 79(1), 6–21. https://doi.org/10.1037/a0022200

31.

Pence

Paymar

(1993). Education groups for men who batter: The Duluth model. Springer.

32.

Plichta

S. B.

(2004). Intimate partner violence and physical health consequences: Policy and practice implications. Journal of Interpersonal Violence, 19(11), 1296–1323. https://doi.org/10.1177/0886260504269685

33.

Reichel

(2017). Determinants of intimate partner violence in Europe: The role of socioeconomic status, inequality, and partner behavior. Journal of Interpersonal Violence, 32(12), 1853–1873. https://doi.org/10.1177/0886260517698951

34.

Reitan

R. M.

(1958). Validity of the Trail Making Test as an indicator of organic brain damage. Perceptual and Motor Skills, 8(8), 271–276.

35.

Rennison

C. M.

Welchans

(2000). Intimate partner violence. United States Department of Justice, Bureau of Justice Statistics.

36.

Rosenbaum

Hoge

S. K.

(1989). Head injury and marital aggression. The American Journal of Psychiatry, 146(8), 1048–1051. https://doi.org/10.1176/ajp.146.8.1048

37.

Rosenbaum

Hoge

S. K.

Adelman

S. A.

Warnken

W. J.

Fletcher

K. E.

Kane

R. L.

(1994). Head injury in partner-abusive men. Journal of Consulting and Clinical Psychology, 62(6), 1187–1193. https://doi.org/10.1037/0022-006x.62.6.1187

38.

Rubin

D. B.

(1987). Multiple imputation for nonresponse in surveys (Vol. 81). John Wiley.

39.

Saltzman

L. E.

Fanslow

J. L.

McMahon

P. M.

Shelley

G. A.

(1999). Intimate partner violence surveillance: Uniform definitions and recommended data elements. Version 1.0. Centers for Disease Control and Prevention.

40.

Sledjeski

E. M.

Dierker

L. C.

Brigham

Breslin

(2008). The use of risk assessment to predict recurrent maltreatment: A classification and regression tree analysis (CART). Prevention Science, 9(1), 28–37. https://doi.org/10.1007/s11121-007-0079-0

41.

Smith

S. G.

Chen

Basile

K. C.

Gilbert

L. K.

Merrick

M. T.

Patel

, … Jain

(2017). The national intimate partner and sexual violence survey (NISVS): 2010-2012 State report. National Center for Injury Prevention and Control, Centers for Disease Control and Prevention.

42.

Steadman

H. J.

Silver

Monahan

Appelbaum

P. S.

Clark Robbins

Mulvey

E. P.

, … Banks

(2000). A classification tree approach to the development of actuarial violence risk assessment tools. Law and Human Behavior, 24(1), 83–100. https://doi.org/10.1023/A:1005478820425

43.

Stewart

L. A.

Power

(2014). Profile and programming needs of federal offenders with histories of intimate partner violence. Journal of Interpersonal Violence, 29(15), 2723–2747. https://doi.org/10.1177/0886260514526059

44.

Strauss

Sherman

E. M.

Spreen

(2006). Norms selection in neuropsychological assessment. In E. Strauss, E. M. Sherman, & O. Spreen (Eds.), A compendium of neuropsychological tests: Administration, norms, and commentary (3rd ed.). Oxford University Press.

45.

Teichner

Golden

C. J.

Van Hasselt

V. B.

Peterson

(2001). Assessment of cognitive functioning in men who batter. International Journal of Neuroscience, 111(3–4), 241–253. https://doi.org/10.3109/00207450108994235

46.

Theobald

Farrington

D. P.

Coid

J. W.

Piquero

A. R.

(2016). Are male perpetrators of intimate partner violence different from convicted violent offenders? Examination of psychopathic traits and life success in males from a community survey. Journal of Interpersonal Violence, 31(9), 1687–1718. https://doi.org/10.1177/0886260515569061

47.

Therneau

Atkinson

Ripley

(2015). rpart: Recursive partitioning and regression trees. R Package Version 4.

48.

United States Department of Justice. (2017). Domestic violence. https://www.justice.gov/ovw/domestic-violence

49.

Vaden

K. I.

Gebregziabher

Kuchinsky

S. E.

Eckert

M. A.

(2012). Multiple imputation of missing fMRI data in whole brain analysis. Neuroimage, 60(3), 1843–1855. https://doi.org/10.1016/j.neuroimage.2012.01.123

50.

Wechsler

(2008). Wechsler Adult Intelligence Scale—Fourth Edition (WAIS-IV). NCS Pearson.

51.

Wilkinson

G. S. R. G. J.

(2006). Wide Range Acheivement Test—Fourth Edition (WRAT-4). Psychological Assessment Resources, Lutz.

52.

World Health Organization. (2005). WHO multi-country study on women’s health and domestic violence against women: Summary report of initial results on prevalence, health outcomes and women’s responses.

53.

Zamora

(2001). Intellectual characteristics of incarcerated males: Does education and race matter? [Master’s thesis]. New York University.