Women Arrested for Domestic Violence:
Unintended Consequences of Pro and Mandatory Arrest Policies’
Peter S. Hovmand, Ph.D.”
George Warren Brown School of Social Work
Washington University in St. Louis
Campus Box 1196
One Brookings Drive
St. Louis, MO 63130, USA
(314) 935 7968
phovmand@ wustl.edu
David N. Ford, Ph.D.
Zachry Department of Civil Engineering
Texas A&M University
College Station, TX 77843-3136 USA
(979) 845 3759
davidford@ tamu.edu
Ingrid Flom
George Warren Brown School of Social Work
Washington University in St. Louis
Campus Box 1196
One Brookings Drive
St. Louis, MO 63130, USA
iflom@ wustl.edu
Stavroula Kyriakakis, MSW
George Warren Brown School of Social Work
Washington University in St. Louis
Campus Box 1196
One Brookings Drive
St. Louis, MO 63130, USA
skyriakakis@ wustl.edu
| Paper presented at the International System Dynamics Conference July 29-August 2, 2007, Boston, MA. Data
discussed in this paper was made possible through the generous participation of a rural Michigan County and partly
supported under a grant from the Centers for Disease Control and Prevention, Atlanta, GA.
? Investigator with the Center for Mental Health Services Research, George Warren Brown School of Social Work,
Washington University; through an award from the National Institute of Mental Health (P30 MH068579).
Domestic Violence 1
Abstract
Domestic violence is a major social problem worldwide. In the United States, the failure
of communities and police departments to intervene resulted in a push to adopt and implement
pro and mandatory arrest policies for domestic violence. These policies have led to an
unexpected increase in the number of arrests of women. Competing explanations have been
offered. This paper describes the development of a system dynamics model of women arrested
for domestic violence. Results suggest that these policies may have created or strengthened a
crossover mechanism that shifts the risk of arrests in domestic violence cases from men to
women. Model analysis demonstrates how the changing role of cooperation between advocates
and police can help explain the trends in women arrests. Implications for research and policy
are discussed.
Keywords: domestic violence, mandatory arrest, criminal justice response, women
1. Introduction
Domestic violence is a major global problem with women reporting a lifetime prevalence
rate of physical abuse by intimate partners between 10% and 52% (World Health Organization
2005). In the United States, 25% of women report having been raped or physically assaulted by
an intimate partner during their lifetime (Tjaden and Thoennes 1998). Battering’ involves a
constellation of tactics, including emotional abuse, isolating the victim from resources,
preventing her from getting or keeping a job, coercion and threats of violence, rape, and murder.
Abusers use battering tactics to keep or delay a woman from leaving an abusive relationship,
with violence continuing and often escalating after she leaves. In a Michigan study, 37.4% of
the women with one or more violent partners said that the violence continued after separation
and 45.6% indicated that the violence increased (Largo et al. 1999).
Prior to the modern battered women’s movement in the United States, communities were
reluctant to apply criminal law to intimate partner violence (Fagen 1996; Schechter 1982;
Mullender 1996). Most took what Pence and McDonnell (1999) describe as a lasses-faire
approach in police and court responses. In an effort to make effective assistance available to
women in danger, advocates for battered women argued for increased involvement by the
criminal justice system, which gained empirical support from Sherman and Berk’s (1984) widely
publicized Minneapolis Experiment of arrest policies and specific deterrence effects.
While the replication studies of arrest policies were mixed in their outcomes (for a
summary, see Zorza and Woods 1994), the results from the Minneapolis Experiment study were
used along with several large lawsuits and pressure from advocacy groups to pass laws requiring
police departments to develop written policies on domestic violence arrests (Davis, Smith, and
Nickles 1998). Today, more than 91% of police departments have a written policy for domestic
3 In order to make the distinction between the single event and pattern within this paper, abuse will refer to a specific
behavior while battering will refer to a pattem of abusive behaviors over time. A busive behaviors include physical
abuse, sexual abuse, emotional abuse, economic abuse, destruction of property, coercion, and threats. Battering
refers to an ongoing pattern of abusive behaviors. Domestic violence includes both abuse and battering. Men who
batter will refer to men who use tactics of abuse and battering against their intimate partners (Pennsylvania Coalition
Against Domestic Assault 1992, Section K. Definitions). Hence, the terms ‘male batterers’, ‘batterers’, ‘abusers’ ,
and ‘assailant’ will be used interchangeably to refer to adult men (ages 18 and over) who batter women.
Domestic Violence 2
violence arrest, with 17% having a pro-arrest policy and 72% a mandatory arrest policy
(Kickman and Reaves 2003). Pro-arrest policies encourage police officers to make a warrantless
misdemeanor arrest when responding to calls if there is evidence of domestic violence (e.g.,
broken glass, over turned furniture, bruises or scrapes on a person). Mandatory arrest policies
require police officers to make an arrest.
Pro and mandatory arrest policies were expected to increase the safety of battered women
and accountability for male batterers (Miller 2001). While the increase in men arrested for
domestic violence was expected, the disproportionately greater increase in women arrested for
domestic violence was not (Hirschel and Buzawa 2002). Women now represent nearly 20% of
the intimate partner related arrests (Durose et al. 2005). This has serious negative consequences
for victims including disempowerment, losing custody of children, and revictimization (Das
Dasgupta 2002; McMahon and Pence 2003; Mills 1998; Osthoff 2002; Hirschel and Buzawa
2002; Rajah, Frye, and Haviland 2006; Miller and Meloy 2006).
A variety of potential explanations have been offered in the literature including increases
in women’s use of force against intimate partners, men becoming less violent, men manipulating
the criminal justice system, a more egalitarian criminal justice system, constraints on the
criminal justice system to effectively respond to social problems, and “net widening” (DeLeon-
Granados, Wells, and Binsbacher 2006). As applied to women arrested for domestic violence,
net widening refers to the effect where more women are arrested because pro and mandatory
arrest laws effectively lower the threshold for making an arrest. Considerable empirical evidence
suggests that juvenile and criminal justice reforms such as diversion and treatment programs
have resulted in net widening, gradually pulling in cases that would otherwise not been in the
system to begin with (Blomberg 1995).
The consequences of women arrests and variety of potential explanations has placed an
urgency to develop a more contextual and ecologically oriented understanding of women’s use
of force in domestic violence and women being arrested for domestic violence (Renzetti 1999;
DeLeon-Granados, Wells, and Binsbacher 2006; Swan and Snow 2006, 2002). These accounts
include attending to situational factors that help explain women’s use of force and to the
organizational dynamics driving the rise in women arrests.
In this paper, we describe a system dynamics model of women arrested for domestic
violence. The paper is organized as follows. Section 2 summarizes the literature on women’s
use of force and consequences of women arrested for domestic violence. Section 3 describes the
development of the model including the structure and data sources. Section 4 presents the
system behavior used to test the model. Section 5 discusses the validation of the model using
both numerical data and qualitative data from key informant interviews. Section 6 presents the
results from behavioral analysis of the model. Section 7 concludes with a discussion of the
results and their implications for policy, research, and practice.
2. Background
The effects of mandatory arrest policies on battered women are mixed. These concerns
range from needing to develop a better understanding of women’s use of force to lack of court
mandated programs for women and the impact of women being arrested on child custody and
child abuse and neglect investigations. This section begins with a discussion of women’s use of
force and then considers some of the effects on women of being arrested for domestic violence.
Domestic Violence 3
2.1. Women’s Use of Force
Whereas battering is generally recognized by researchers and scholars of domestic
violence as an ongoing pattern of coercion and control, the criminal justice response is largely
organized around single incidents (Hirschel and Buzawa 2002). From the perspective of the
police response, women are arrested because there is sufficient evidence of her use of force to
warrant the arrest. This incident-based focus on meeting criteria set forth in the criminal justice
codes and mandatory arrest polices prevents police from factoring in the context surrounding a
person's use of force. Several researchers and scholars have drawn attention to this problem by
seeking to understand women’s use of force more broadly than the narrow criminal definition in
domestic violence (e.g., Miller and Meloy 2006; Renzetti 1999; Osthoff 2002).
While it has often been acknowledged that women do use force in self-defense, there has
been a general reluctance to understand women’s agency in using force beyond an imminent
threat (Renzetti 1999). Conceptually, use of force in intimate partner relationships includes
everything from a single hit to an ongoing pattern of behaviors including threats, intimidation
and emotional abuse. Only the last would constitute battering. Failure to recognize this
distinction leads to conflating women’s use of force in self-defense or in response to an ongoing
pattern of battering with battering by women. This contributes to the misleading conclusion that
women and men are engaged in “mutual combat” (Osthoff 2002) and that the arrest of women
reflects a previously hidden population of women batterers.
Where women tend to use force defensively or to regain control, men who are arrested
for domestic violence use force to establish and maintain control over their partners (Das
Dasgupta 2002). Numerous studies have found women to differ from men in their motivations
for using force. Women’s motivations include self-defense, fear, defense of children, control,
retaliation, showing feelings that cannot be expressed in words, and provocation or being
“pushed over the edge” by her partner (Swan and Snow 2006; Stuart, Moore, Hellmuth et al.
2006).
Numerous studies also support the interpretation that women’s use of force in the context
of domestic violence operates defensively and in response to battering behavior. Studies have
found a high co-occurrence of domestic violence victimization with women’s use of force (for a
review, see Das Dasgupta, 2002). Men who were arrested for domestic violence are more likely
to have violated personal protection orders and had previous contact with police for domestic
violence than women (Henning, Jones, and Holdford 2003). Women are also much more
forthcoming about their use of force in contrast to men, increasing the likelihood that they as
opposed to the male abuser would be arrested by police (Larance 2006; Rajah, Frye, and
Haviland 2006). A common experience reported in the literature is that women cooperating with
the police investigation and being honest about their own actions expect that police officers are
going to help them, and are therefore shocked by their own arrest (Rajah, Frye, and Haviland
2006). Thus, while small fraction of domestic violence cases involves women battering men or
other women, research suggests overall that the increasing numbers of women arrested for
domestic violence are largely women victims of battering.
2.2. Mental Health
A number of studies have noted the higher co-occurrence of mental health needs related
to trauma in women arrested for domestic violence, which would be consistent with ongoing
patterns of victimization in a battering relationship. For example, Henning et al (2003) found that
Domestic Violence 4
women arrested for domestic violence have higher needs for mental health services than men
arrested for domestic violence. Stuart et al (2006) found that women arrested for domestic
violence experienced higher rates of mental illness, as well as high rates of sexual and
psychological abuse by partners. These studies have generally not distinguished between the
mental health needs arising from victimization, the arrest experience, and pre-existing mental
health conditions. This has contributed to a body of research arguing for specialized batterer
treatment programs for women that address these mental health needs, instead of more critically
considering how these needs may have resulted from the victimization experience or been
exacerbated by the arrest experience.
2.3. Disempowerment
Pro and mandatory arrest policies can function to disempower victims even if the batterer
is arrested. One of the main criticisms of mandatory arrest policies for domestic violence has
been the loss of decision making power women experience in the police and prosecutorial
response (Rajah, Frye, and Haviland 2006). A primary concern is that with mandatory arrest, a
victim’s ability to choose whether an arrest is the best course of action is taken away (Danis
2003; Humphries 2002). The decision to take action against the abuse and leave an abusive
partner is complex and does not necessarily result in an end to the suffering. The certainty of
arrest may deter women from calling the police in a time of danger, especially if the arrest of a
husband or partner will result in serious consequences to the family, such as bringing shame
upon the family, loss of income, involvement with the child welfare system, being ostracized
from communities, deportation of husband, or retaliation by the abusive partner (Menjivar and
Salcido 2002; Shim and Hwant 2005; Mills 1999; Danis 2003; Zanipatin et al. 2005). Abusive
behavior and violence continue and may even be heightened when a battered woman takes action
to end an abusive relationship (Fleury, Sullivan, and Bybee 2000; Barnett 2000).
Women who are arrested are also isolated from needed supports. Women become less
likely to seek assistance from police or other social services, and are therefore less likely to
report continued victimization. Those who do seek assistance subsequent to arrest may find that
their arrest record disqualifies them for services. Women who have been arrested are also less
likely to seek assistance from police or seek assistance from social services as a consequence of
the arrest decision, and therefore less likely to report continued victimization. Victims of
domestic violence that have been arrested lose access to transportation, shelters and temporary
housing, personal protection orders, victim assistance, and counseling and empowerment
programs along with missing work, risk of deportation, increased economic hardship, and losing
custody of children (Hirschel and Buzawa 2002; Humphries 2002). The result is that by
increasing the state’s ability to hold batterers accountable and provide for women’s safety
through changes in the criminal justice system, many victims of battering now have less legal
protection, fewer services, and are sometimes inaccurately labeled as a suspect, defendant, or
perpetrator (Miller and Meloy 2006).
In addition, low income communities, immigrants and communities of color are
disproportionately affected by the unintended consequences of mandatory arrest policies (Mills
1999; Chesney-Lind 2002; Houry and Kellermann 2005). In some communities, women are
more likely to use force to fight back, placing them at greater risk as being seen as an
unsympathetic victim (Miller and Meloy 2006; Swan and Snow 2006, 2003). There is also the
greater isolation associated with recent immigration as women are separated from extended
families and support in their countries of origin (Swan and Snow 2006). These factors
Domestic Violence 5
underscore the need to develop accounts that understand women’s use of force and domestic
violence arrests contextually (Osthoff 2002; Renzetti 1999).
Understanding women’s use of force and arrests contextually requires us to understand
the underlying causal story. In this paper, we seek to understand how the pattern of women
arrested for domestic violence develops over time in terms of specific feedback mechanisms.
Our main research question is: Which mechanisms account for the dynamics of women arrested
for domestic violence?
3. A System Dynamics Model of Domestic Violence Arrest
The system modeled is a community implementing a mandatory domestic violence arrest
policy. The simulation model and documentation are available from the first author and can be
run on the personal learning edition (PLE) of Vensim (2006), available at no cost at
www.vensim.com. The model itself was developed using V ensim DSS Version 5.6a.
3.1. System Structure and Behavior Data
Basic model structures such as the aging chain of a domestic violence case were based on
authors’ understanding of the criminal justice processes related to domestic violence. Forrester
points out that direct observational experiences of a system are some of the richest sources of
information (Forrester 1980). The authors have more than 24 years of combined experience in
the domestic violence movement researching, volunteering, serving on non-profit boards,
working as staff or program director, and developing programs across five different communities
in the United States. While direct experience in a system provides an invaluable source of
knowledge of key structures, assumptions must be checked and hypotheses tested against other
participants’ perceptions of the system and as well as numerical data whenever possible.
A better understanding of system behavior and structure came through comparison of
model structure and behavior against qualitative and quantitative data from an earlier study of the
impact of mandatory arrest. The study followed a single case study research design (Yin 1994)
of a rural Midwestern community over a four year period from 1998 through 2002. The
community had a population of approximately 60,000 people that was 95% percent non-Hispanic
white with median household income of approximately $37,000 and 10% of the population
living below the poverty line. The initial model from the study pointed to coordination of
services and the risk to clients as important structures to consider.
Key informant interviews from the study were used to confirm or disconfirm community
trends and assess the initial model structure. Key informants also suggested structures that
needed to be included in the model such as cooperation. For example, the initial model did not
include coordination of services as a mechanism. However, key informants accounts of changes
in the community identified coordination as an important mechanism that needed to be included
in the model.
Informants were selected using snowball sampling, and included on the basis of their
expertise with domestic violence. Snowball sampling is a method that uses data from one
participant to identify additional participants. Two initial contacts were made to a domestic
violence coalition coordinator and assistant prosecuting attorney. A total of seven interviews
were conducted with 5 key informants. Three of the key informants had victim advocacy
positions, while the other two key informants worked for the prosecutor's office. All 5 key
Domestic Violence 6
informants had attended a variety of domestic violence trainings, and had between 4 and 10 years
of experience working with victims of domestic violence.
The numeric behavior data came from the prosecutor's office violence against women
database. In 1998, the community received funding for a domestic violence prosecution unit.
This allowed the prosecutor's office to maintain a database on violence against women that
included arrest warrants for domestic violence. Each record included information on the name of
the suspect, victim, victim-offender relationship and demographics along with other variables
describing the circumstances of the arrest, use of force, and dates of critical events (e.g., arrest,
authorization of warrant, court, disposition). The resulting sample covered 1,455 domestic
violence related arrests from January 1, 1998 through December 31, 2001 involving 2,050
individuals. For the purpose of modeling, we selected a subset of cases that were clearly
identified by their victim-offender relationship as domestic violence. For example, cases that
included victim-offender relationship of wife-ex-husband or girlfriend-boyfriend were included.
Cases that involved daughter-father, sister-brother, or wife-brother-in-law were excluded. This
yielded 522 incidents involving 417 individual suspects. These data were used to generate time
series of first and repeat arrests by victim-offender relationship and gender.
These time series were triangulated against other data in the community and found to be
consistent in their general trends. For example, we asked all key informants what if any trends
they had observed in their community. If they mentioned changing trends such as increasing
arrest of women that had since been improving, then we inquired about when they noticed those
changes. We also compared key events against public documents and other longitudinal data
such as personal protection order petitions.
3.2. Units of Analysis
There are a number of ways that one might consider the units of analysis in domestic
violence, from incidents and arrests to individuals, intimate partners or couples, families, and
households. In this model, the units of analysis are intimate partner relationships. Theoretically,
either or both parties could be arrested for domestic violence. However, policies and laws that
require police officers to screen for the primary aggressor eliminated the possibility of dual
arrests. So at any given time, cases of domestic violence have only men or women at risk of
arrest in a heterosexual relationship, but not both.
3.3. Model Structure
This section describes the resulting system dynamics model of women arrested for
domestic violence. The model represents domestic violence cases moving through three risk
categories of arrest in the criminal justice system. Each risk category is represented by a stock.
In Figure 1, domestic violence cases enter the system from the bottom when the abuse has
escalated to the point of criminal behavior where police can make an arrest. In the real system
this flow is driven by all the factors that contribute to the onset and escalation of domestic
violence by men (e.g., exposure to domestic violence as a child, substance abuse, attitudes
toward women). Cases can leave the system from Cases exiting with men at risk and no priors,
Cases exiting with men at risk and priors, and Cases exiting with women at risk and no priors
through cessation of domestic violence (possibly from counseling) or leaving the jurisdiction.
Calibration of the model resulted in no Cases entering with women at risk of arrest. The
Domestic Violence 7
remainder of the discussion will therefore assume that cases of domestic violence only enter the
system with men at risk of arrest.
Figure 1
Stocks and Flows in a Domestic Violence Model
i rn t
Cases exiting Cases exiting Cases exiting with
with men at tisk with men at women at risk
and no priors risk and priors and no priors
First arrest, First arrest of
Cases with of men Cases with Crossovers Cases with women
menat risk pe} menat risk >| women at risk nee
and no priors and priors and no priors
Cases entering Cases entering
‘with men at with women at
risk of arrest tisk of arrest
The three stocks are related through a system of feedback mechanisms represented by the
causal loop diagram in Figure 2. The main reinforcing feedback mechanism is cooperation (R1
in Figure 2). The arrest of women decreases cooperation, which leads to a reduction in the
amount of positive and effective communication between police and victim advocates. This
increases the risk or hazard rate ratio (HRR) of crossovers represented by HRR of cooperation on
crossovers in Figure 2, which increases the number of crossovers and leads to further increase in
women arrested for domestic violence. Similarly, a decrease in women arrests will lead to
improved communication and the lower risks of cases crossing over from men at risk to women
at risk. This will result in fewer women being arrested and reinforce the trend toward improved
cooperation and communication between victim advocates and police.
There a number of balancing feedback mechanisms shown in Figure 2. Feedback loop
B1 represents general deterrence for first arrests, and B2 represents general deterrence for repeat
arrests. Mandatory arrest policies are partly motivated by the expectation that these two feedback
mechanisms will dominate the behavior of the system. Feedback loops B3 and B10 represents
the effect of cooperation on cases of domestic violence being reported to police and leading to
first and repeat arrests respectively.
Domestic Violence 8
Figure 2
Major Feedback Loops Related to Risk Categories
cases ex
wihmen at nek
(Gof and no priors
Reporting of cases . (af
ry B9 +
Satie prtors Hazard rate of,
P arrest of men
Cases entering + Cases with men at
with men at risk tisk and no priors
of arrest
A) + vat of cases
with no priors
+
mS, from
first arrest First arrest
ofmen
Total
amrests nent / HRR of
cooperation on.
crossovers
+ fo
Cases & Gy \
D Jo from men at risk
repeat arrest and priors + VO Rl
" |. ‘
Repeat arrest Cooperation Cooperation
of men B6 | ;
+ = '
Cases exiting
with men at Cases wih
risk and priors - ‘women at risk B7
and no priors
48) Nt Bet arrest
Cases exiting £ of women
‘with women,
atrisk and no
priors
There are six balancing feedback mechanisms regulating the number of cases leaving a
risk category (B4, B5, B6, B7, B8, and B9 in Figure 2). These represent the underlying logic of
the system that cases cannot transition out of a risk category if there are no cases in the category
to begin with. For example, in B4 the more cases there are with men at risk of arrest and no
priors, the more first arrest of men there will be. With each arrest, there is one less case that is
still at risk. Thus the effect of first arrest of men is to lower the number of cases with men at risk
and no priors. An economically austere view of systems would try to reduce all social
phenomena to this logic. While this would be wrong because behavior of social systems and
services are often better explained by the perceptions and policies of decision makers, sometimes
it is this logic of constraints that drives system behavior. For example, the decline of first arrest
of men (see Figure 3) can be understood strictly in terms of the balancing feedback mechanism
B4 as a draining process. In contrast, the first arrest of women cannot be accounted for through
this simple mechanism and requires a more complicated structure.
Domestic Violence 9
4. System Behavior
Figure 3
Trends of First Domestic Violence Arrests by Gender of Suspect
120
-O-First arrests of women
i First arrests of men
eee ~O
20
Year
Figure 3 shows the system behavior of first and repeat arrests from the prosecutor's
database. Numeric values and percentages are listed in A ppendix A, Table 1. Domestic violence
arrests of women increased from 15% of all domestic violence arrests in year 1 to 27% in year 4.
Meanwhile, the number of men arrested dropped monotonically from 85% to 73% over the same
period. Figure 3 shows the trends for the first domestic violence arrests by gender of suspect
5. Model Validation
The complexity of the changes in system conditions being modeled (e.g. shifting locus of
risk) and newness of the application of system dynamics to the issue of domestic violence make
model validation particularly important. Using several tests described by Sterman (2000), the
model was tested for structural similarity to the system structure, consistency, reasonableness of
behavior, and similarity of model behavior to system behavior. Confidence in the similarity of
the model structure to the system structure was increased by the inclusion of key mechanisms.
Two of the model structures considered to be of particular importance and therefore in need of
validation were cooperation mechanisms and the hypothesized crossover of cases from men at
risk to women at risk. For example, cooperation emerged as a central theme in key informant
interviews:
e Three key informants explained how the relationship between the domestic violence
shelter and police department deteriorated from year 1 to year 3 over the police
departments handling of domestic violence cases. One elaborated that this had to do
with the fact that the police department would arrest women who the shelter knew to
Domestic Violence 10
be victims of domestic violence. The conflict peaked in year 3 with a standoff
between the local police and shelter. The fallout forced the resignation of both the
police chief and executive director of the shelter. Key informants’ accounts of this
conflict were corroborated with public documents.
Several key informants shared how changes in the leadership of both the police
department and domestic violence shelter in year 3 started a new sense of cooperation
and access to police officers. With an increased cooperation and access, advocates
started working with police officers on a case-by-case basis after a victim was
arrested. These conversations focused on helping officers understand the dynamics of
domestic violence and identify victims in the future.
Key informants also explained how the prosecutor's office started in year 3 to have
one assistant prosecuting attorney handle all of the domestic violence cases with the
exception of sexual assaults. This improved communication and cooperation between
the prosecutor's office and victim advocates in the domestic violence shelter, which
made it much easier for victim advocates to respond quickly in cases where the victim
had been arrested and share information with the assistant prosecuting attorney.
Similarly, we gained confidence in the crossover mechanism through the key informant
interviews. For example, several key informants explained how the batterers’ initial contacts led
to an increased risk of victims becoming suspects in subsequent investigations:
One informant shared how in talking with women who had been arrested, victims
took matters into their own hands because nothing came from the previous arrest or
the batterer provoked the victim. For example, the “Perps [perpetrators] will say, ‘Hit
me, this time you're going down [to jail].’” When the police arrive and ask what she
did, she would say “Y eah, I pushed him” and then get arrested.
Another informant pointed out that if we were concemed about batterers
manipulation of criminal justice professionals, then we should also be concerned with
they way they manipulate social workers in child protective services. For example,
the batterer would initially be the one investigated for child abuse, but after the initial
investigation he would brag to the victim about how he now knew how to get the
children taken away from the victim.
Because we had individual level data with identifiers in the prosecutor’ s violence against
women database, we were also able to test the existence of the crossover mechanism using
statistical relationships. The stock-flow structure in the model supported by key informant
accounts would entail two service pathway hypotheses that we could test: (1) women suspects
were more likely than men to appear as victims in an earlier incident, and (2) this effect would be
stronger in cases clearly identified as domestic violence. Chi-square tests were used to compare
the expected frequencies of women’s pathways through criminal justice system with men’s
pathways. Women suspects were more likely than men to enter the system initially as victims
with nearly 37% of women suspects appearing initially as victims in an earlier incident versus
7% of men (Phi = 0.26, Chi-Square (1,1128) = 77.4, p<0.001).
Domestic Violence 11
Next, the strength of association between gender and pathway was considered for cases
with clear domestic violence (victim-suspect relationship identified as current or former spouse
or boyfriend/girlfriend). Within this subset of adult domestic violence cases, 32% of women
appeared as victims in previous cases versus 6% of men (Phi = 0.35, Chi-Square (1,445) = 53.9,
p<0.001). Phi increased from 0.26 to 0.35 indicating cases clearly identified as domestic
violence had a stronger association between victim status from the first incident and gender.
These results were corroborated by inspecting the tactics arresting officers reported being used
and comparisons of dispositions from prosecution. This provided support for the existence of a
crossover mechanism where cases with men at risk of arrest crossed over to being cases with
women at risk of arrest.
Figure 4
Comparison of Model Simulations A gainst System Behavior
120
-+ First arrest of men (model)
=i First arrest of men (data)
-O- First arrest of women (model)
—O- First arrest of women (data)
Year
It is important to emphasize that the use of statistical tests here was to empirically test
hypotheses identified through the modeling based on key informant interviews as opposed to
simply testing the fit of a model against data. That is, we had an initial model that was wrong
according to key informant interviews. The interviews suggested a crossover mechanism. The
existence of a crossover mechanism would imply two hypotheses about pathways. We then
tested these hypotheses and found independent support for the existence of a crossover
mechanism in this community.
Other standard tests of the validity of system dynamics models were also used.
Equations were reviewed for consistency with key informant interviews and the model uses a
consistent set of units. The model also generates reasonable behavior over a wide range of
exogenous parameter values. For example, the model reproduces realistic qualitative behavior
patterns of arrests over a wide variation of values for Effect size of arrest intervention, Effect size
of coordination intervention, Initial cases entering with men at risk, Baseline hazard of arrest
rate, Time at risk, and Baseline hazard rate of crossover. The model also passed a number of
extreme conditions tests, and reproduces the known behavior. Figure 4 shows the comparison of
the model simulations against observed frequencies from the community for both first arrest of
Domestic Violence 12
men and first arrest of women. Based on these tests the model is considered valuable for
investigating the impacts of pro and mandatory arrest policies in domestic violence cases.
6. Results
Having a simulation model that provides a logically consistent explanation with multiple
data sources provides an important tool for understanding the relationship between the structure
of feedback loops, policies, and system behavior. Experimental manipulation is an important
scientific tool for understanding the causes of behavior. Community studies, however, are
complex and fraught with uncertainties, politics, and ethical concems that often constrain the
types of experiments a community might be willing and able to undertake. Moreover, when we
are able to design and conduct such studies, we generally do not have a means or the luxury to
replicate the experiment precisely as we might in a lab to test our analysis and explanations.
Such limitations severely constrain the ability of policy makers, advocates, and researchers to
anticipate and understand the consequences of policy changes such as mandatory arrest.
However with empirically based simulation models, one is able to conduct experiments
that are limited only by time and computing power. One is able to study the behavior of the
system as parameters are changed; whole structures removed or added; policies and interventions
modified; and all of this over a wide population of communities and over longer periods of time.
This facilitates a much more rigorously tested set of hypotheses and explanations about the
relationship between the structure of the system and its behavior.
The analysis reported here focuses on understanding the relationship between the model
structure and first arrests of women. The goal of this analysis is to provide an explanation of
how the simulated behavior is generated in terms of the feedback mechanisms shown in Figure 2.
The influence of these feedback mechanisms was identified and verified through simulation
experiments. This included a behavioral approach to identifying dominant feedback loops (Ford
1999) implemented in the S-programming language using R Version 2.3.1 (available at www.r-
project.org). The S functions are available from the first author.
The baseline simulation in Figure 4 represents the behavior of the community under the
observed conditions. One common source of policy design failure is not anticipating the
consequences of policies over longer time horizons (Sterman 2000). So to more fully assess this
relationship, the simulation of the model calibrated to the community data is run over the longer
time horizon of 10 years. While arrest data are typically reported as the number of arrests over a
period of time (month or year), the dynamics driving the system are continuous over time. To
get a clearer picture of these dynamics, simulation results are plotted as they are represented in
the computer simulation as opposed to how they appear in aggregated annual reports.
Figure 5 shows the time varying behavior of the variable First arrest of women over a 10
year time horizon. The simulation can be understood as single system research design with three
phases: baseline phase (Phase A), mandatory arrest intervention (Phase B), and mandatory arrest
with the addition of a coordination intervention and advocacy intervention (Phase C). The initial
baseline phase represents the model in a dynamic equilibrium. This baseline phase provides a
way to establish that the dynamics are the result of a system’s response to the interventions. It is
a synthetic state of a community since communities are rarely if ever in equilibrium.
Importantly, the system being in a dynamic equilibrium does not mean that there are no
dynamics, but just that the changes that do occur have no net effect on the number of cases in the
various stocks throughout the model. That is, the number of cases entering each of the risk
Domestic Violence 13
categories in Figure 1 equals the number of cases exiting each of the risk categories. The point
to be aware of is that in this initial baseline phase, these risk categories are not empty. Prior to
implementation of mandatory arrest, there are cases with men who are at risk of arrest and no
priors, cases with men at risk of arrest and priors, and cases with women at risk and no priors.
Any policy intervention will therefore initially act on the cases in the risk categories, with
subsequent effects unfolding over time as the changes move through the system.
Figure 5
Simulated First Arrests of Women over Ten-Y ear Time Horizon
A B,B) By C, CG C3 Cy Ce
|
30
VIZ
—__
Year
The first intervention phase starts in year 1 and represents the introduction of a
mandatory arrest policy. The immediate effect is to see an increase in the first arrest of women.
Over the next several months (Phase B,), the number of first arrests of women actually decreases
after the initial increase as the number of cases leaving Cases with women at risk of arrest and
no priors is greater than the number of cases coming into the category from Crossovers. The
primary loop causing this behavior is B7 in Figure 2. However, the number of arrests soon
begins to increase in phase Bz as an increase in arrests of men has led to more Cases with men
with at risk of arrest and priors. This causes an increase in Crossovers, and hence more Cases
with women at risk of arrest and no priors. The result is an increase in the First arrest of women
with B7 driving the behavior again at the end of phase B».
The increase in First arrest of women causes a decline in Cooperation. This decline in
cooperation between advocates and police leads to less communication and less effective
advocacy on specific cases, which increases the risk of crossovers. This increases the number of
women being arrested for domestic violence until the erosion of advocacy within cooperative
Domestic Violence 14
relationship begins to bottom out. Whatever benefit that might come from cooperation between
victim advocates and police has now disappeared. This erosion of the effectiveness of direct
advocacy with clients ultimately limits the rate of growth in the first arrest of women in phase
B3.
While direct advocacy with police has eroded, advocacy at the community level is still
important. It is arguably this deterioration of the relationship between police and victim
advocates that ultimately lead to major changes in the coordinating council including changes in
the police chief, prosecutor handling domestic violence cases, and executive director of the
domestic violence shelter. These changes reflect important feedback effects within this
particular community, but it would be difficult to say whether or not such feedback effects could
realistically be expected in another community. Differences in personality of the coordinating
council members, local politics, and community history would all have to be considered. While
these dynamics of local politics and functioning of coordinating councils are outside the
boundary of this model, one might ask, what kinds of political processes would change the
direction of cooperation? And, what types of interventions would result from such changes? In
this model, the resulting changes are represented as exogenous variables affecting victim
advocacy and cooperation.
The second intervention phase starts in year 3 with the addition of a victim advocacy
intervention in the prosecutor's office and renewed effort to improve the cooperation on the
coordinating council. These interventions lower the number of first arrests of women. In phase
C, the drop in crossovers resulting from the improved advocacy causes a decline in the number
of Cases with women at risk and no priors. This decline is immediately reflected in a reduction
of the number of women arrested relative to the peak in year 3. The decline in the number of
women arrested leads to an improvement in cooperation that increases the effectiveness of
advocacy on reducing crossovers. Thus in phase C; the combination of fewer cases (B7) and
improved cooperation (R1) drives the reduction in first arrest of women.
The decrease continues in phase C2, but begins to slow. There are several causes for this
decrease to slow down. First, there are diminishing returns from improving cooperation on the
effectiveness on advocacy. Second, slowing the crossover rate of cases has the effect of
increasing the number of men at risk of arrest. This increase in the number of men at risk of
arrest begins to counteract the benefits of a reduced hazard rate of crossover. Simply put, while
the risk is lower, the number of cases exposed to the risk increases, and crossovers start to
increase again.
Beginning in phase C3, the number of first arrest of women begins to decline again. This
is caused by the number of cases crossing over peaking and then declining. This is the result of
the effects from the initial mandatory arrest on cases with men at risk and no priors catching up
to the rest of the system. Specifically, while mandatory arrest policies lead to an initial increase
of men arrested for domestic violence in year 1, the men arrested for domestic violence shows a
steady decline as the number of cases with men at risk of arrest and no priors was effectively
drained. With fewer first arrests of men, there are more cases leaving the category of Cases with
men at risk and priors. This causes a net outflow of cases that lowers the number of Cases with
men at risk and priors, lowers the number of Crossovers, and lowers the inflow of Cases with
women at risk and priors. The upshot of this is that the benefits of mandatory arrest are
beginning to take hold throughout the system in phase C3, although not without some significant
unintended consequences along the way including the increased level of women being arrested
for domestic violence.
Domestic Violence 15
By phase C4, the number of Cases with men at risk and priors is now in a steady linear
decline. This is because the number of Cases with men at risk and no priors is now nearly
constant, and hence the number of first arrests entering the category is nearly constant. The
decrease in the number of men arrested, however, has a weak but unintended consequence of
reducing the effects of deterrence on new cases entering Cases with men at risk and no priors.
This change is reflected in phases C; and Cy. While the rate that new cases are entering the
system is increasing, it is less than the number of cases leaving the system. The net result is that
that the first arrest of women continues to decline in phase C7. One important implication of this
is that the effects of deterrence on the system do not appear until well after the initial
intervention for which the policy was designed, and when they do, it is in the weakening of
deterrence resulting from fewer arrests that matters.
Two points are worth emphasizing in the results from the behavioral analysis. First, the
dynamic behavior shown in Figure 5 arises from the feedback mechanisms. That is, the results
were generated by the feedback loops and stocks interacting over time in response to the
interventions in year 1 and year 3. Second, the additional feedback mechanism of cooperation
can explain the pattern of women arrested for domestic violence in a way that is both logically
and empirically consistent with data from this particular community. It is obviously not the only
possible explanation or model of the phenomenon of women arrested for domestic violence in
this community. Nor is the model one that can be immediately generalized to other communities
without modification and additional data. It is, however, one that is explicitly formulated,
transparent, replicable, and therefore refutable.
up Discussion
These results also illustrate some of the inherent complexities in the relationship between
mandatory arrest policies and women arrested for domestic violence. The promised benefits of
the initial intervention do not appear to affect the system until much later. The appreciable
declines in the number of men arrested have more to do with increasing the number of people
moving into a system than an effective deterrence policy. The causes of the unintended
consequences on women shift over time with the dynamics of the system from an initial increase
in the arrest rate, to a declining stock of cases at risk, and erosion of cooperative relationship that
leads to a second intervention. In some phases it is the number of cases within a particular
category that is driving a change. In other phases it is one or more feedback loops.
These results emerge not from programming in the values, but from deductions based on
the underlying logic of how cases move through service systems, principles of system dynamics,
some numeric data, and key informant interviews. The model is a simplification of reality that
practitioners, administrators, and policy makers must contend with in trying to make decisions
about mandatory arrest and other domestic violence policies. Yet even at this stage of
development, the model provides important conceptual insights into the arrest of women for
domestic violence.
First, there is empirical support for the existence of a crossover mechanism for domestic
violence cases. This mechanism represents a shift in the risks from the batterer to the victim
through efforts to leam and game the system response. Key informants pointed to examples of
where this happened with the police response, but also pointed to a similar effect in child
protective service investigations. These reports were supported through subsequent statistical
analyses using the violence against women data set. Simulation analyses of these effects showed
Domestic Violence 16
how this crossover mechanism could play a central role to the arrest of women, and be counter-
acted through interventions to improve victim advocacy in the prosecutor's office and
cooperation between service providers.
Such a mechanism would have important implications for both child welfare and
domestic violence policies. Part of the response to dual arrests has been the development of
protocols and primary aggressor screenings. For example, some have argued for more protocols
and statues encouraging the identification of primary aggressors to address rigid criminal justice
procedure and better training of police officers (Miller and Meloy 2006). Such approaches
presume that providing standardized instruments to police and other professionals intervening in
domestic violence will increase the likelihood that the primary aggressor can be identified.
However, the existence of a crossover mechanism would imply that batterers how to manipulate
the system learn through their initial contact. To the extent that such protocols are accompanied
by increased training that reduces the variation in responder’s behavior, the opportunity to lean
and manipulate the system would increase, exacerbating the risk of cases crossing over rather
then mitigating them. What is forgotten in these protocols is that people can and often do have
previous contacts with service systems and learn from the experience. Similar efforts to screen
for domestic violence cases in child welfare, family courts, and mediation are underway. If the
possibility of crossover mechanisms is ignored or not monitored, these effects will not only be
ineffective, but put women and other victims at even greater risk.
However, crossover mechanisms are readily researchable. They appear as cases
involving multiple parties tracked over time where the suspects appear initially as victims. Many
administrative databases from child welfare and courts keep track of both victim and suspect
identifiers. Conducting secondary analyses of administrative data along with qualitative analysis
with victims, advocates, and batterers to investigate crossover mechanisms in cases of domestic
violence should be a top priority among domestic violence, child welfare, and criminal justice
researchers. The aim should be to develop strategies for improving victim safety.
Second, the model shows how both feedback mechanisms and the stock-flow distinction
play substantively important roles over the course of service system response to a change in
policy. For example, the number of cases crossing over and impacting the number of women
arrested for domestic violence is sometimes driven by the number of cases at risk and at other
times by one or more feedback mechanisms. Both are required to understand how the dynamics
evolve in this system.
In this model, some risk categories or stocks increased with successful interventions on
crossovers, which in tum had an unintended consequence later on of undermining the initial
benefits. For policies to be more effective and sustainable, it is important to find ways to both
reduce the risk of adverse events like crossovers, but also find ways of draining the problematic
stocks. Efforts to do more aggressive outreach with women who have already come in contact
with the criminal justice system (e.g., risk assessment for high-risk cases or domestic assault
response teams for cases of re-victimization) would help drain the stock of cases when women
are at risk and have no priors. Likewise, efforts to increase monitoring or develop more effective
batterer intervention programs would help drain the stock of cases with men at risk of arrest and
priors.
The benefits of such interventions will depend on the specifics of the community along
with the specific sequence of interventions. Additional empirical and simulation work is
required to fully evaluate these policy options. In particular, both quantitative analyses using
administrative data with identifiers and qualitative analysis with community stakeholders
Domestic Violence 17
working in the area of domestic violence across multiple communities are needed in order to
better understand the structure of such systems. The primary goal here should be to test the
model across the widest variation of communities.
This seems imminently feasible using system dynamics and drawing on existing or
ongoing studies of domestic violence. In particular, system dynamics provides a way to model
the underlying structure of systems in terms of feedback loops that researchers can use as a basis
for comparative analyses. Data can come from a wide selection of sources, including numeric
data from administrative databases and qualitative data from focus groups and key informant
interviews. More to the point, system dynamics provides a tool to rigorously synthesize the
results from diverse community studies. Such studies would not only develop and illustrate how
one could conduct more flexible evaluations of complex social problems, as Goodman and
Epstein (2005) have argued for, but also deepen our understanding of the community ecology of
domestic violence and criminal justice response.
Lastly, the paper highlights the importance of cooperation in community responses to
domestic violence. Cooperation provided the central reinforcing mechanism driving some of the
benefits, but also some of the unintended and negative consequences of mandatory arrest
policies. Allen (2005) has studied cooperation within the context of coordinated community
responses (CCR), and pointed out that these types of interventions might not always be
beneficial. For example, CCR efforts often increase expectations that agencies will share
information about victims and offenders. This information sharing is not always symmetrical,
nor beneficial in terms of improving victim safety or increasing assailant accountability, and can
lead to conflicts that undermine the effectiveness of coordinated community response efforts.
The model illustrates how the benefits and dangers of CCR efforts might shift over time in
response to various policy interventions. Moreover, the short and long-term benefits of efforts to
improve cooperation between stakeholders may be dependent on the state of the service system.
In particular, the model shows how the same structure can produce both types of effects. Further
empirical and simulation work should be undertaken to understand these effects in hopes of
developing more robust strategies for successfully implementing interventions.
With the success of the battered women’s movement in raising awareness of domestic
violence, more communities have implemented mandatory arrest policies than ever before and
are motivated to address issues of victim safety and batterer accountability. In the next period of
the policy reforms, the issues facing coordination efforts will require more sophisticated
approaches for involving multiple stakeholders, addressing political issues, and informing the
debates through sound empirical analysis. Moreover, domestic violence issues are, like many
other social issues, increasingly being understood through the lens of public health and related to
other issues such as health and mental health, immigration, child welfare, poverty, schooling, and
economic development. Decision makers recognize that that many of the most challenging
problems we face as a society, both in the more developed countries and developing countries,
are related. System dynamics and group model building can play a vital role in helping
stakeholders understand and solve dynamically complex problems. The readiness of
stakeholders to apply system dynamics toward social problems is in no small part a measure of
the success of system dynamics over the last 50 years. While this bodes well for the future
prospects of system dynamics, the important point is that with policies that are better and more
sustainable, people’s lives will improve.
Domestic Violence 18
References
Allen, Nicole E. 2005. A multi-level analysis of community coordinating councils. American
Journal of Community Psychology 35 (1/2):49-63.
Bamett, Ola W. 2000. Why battered women do not leave, part 1. Trauma, violence & abuse 1
(4):343-372.
Blomberg, Thomas G. 1995. Beyond metaphors: Penal reform as net-widening. In Punishment
and social control, edited by T. G. Blomberg and S. Cohen. New Y ork, NY: Aldine de
Gruyter, Inc.
Chesney-Lind, Meda. 2002. Criminalizing victimization: the unintended consequences of pro-
arrest policies for girls and women. Criminology and public policy 2 (1):81-90.
Danis, F. S. 2003. The criminalization of domestic violence: What social workers need to know.
Social Work 48 (2):237-246.
Das Dasgupta, S. 2002. A framework for understanding women's use of nonlethal violence in
intimate heterosexual relationships. Violence against Women 8 (11):1364-1389.
Davis, R. C., B. E. Smith, and L. B. Nickles. 1998. The deterrent effect of prosecuting domestic
violence misdemeanors. Crime and Delinquency 44 (3):434-442.
DeLeon-Granados, William, William Wells, and Ruddyard Binsbacher. 2006. Arresting
Developments: Trends in Female Arrests for Domestic Violence and Proposed
Explanations. Violence Against Women 12 (4):355-371.
Durose, Matthew R., Caroline Wolf Harlow, Patrick A. Langan, Mark Motivans, Romona R.
Rantala, and Erica L. Smith. 2005. Family violence statistics: Including statics on
strangers and acquaintances: Bureau of Justice Statistics, U. S. Department of Justice
Office of Justice Programs.
Fagen, J. 1996. The Criminalization of Domestic Violence: Promises and Limits. Washington,
DC: National Institute of Justice, U. S. Department of Justice.
Fleury, Ruth E., Cris M. Sullivan, and Deborah I. Bybee. 2000. When ending the relationship
does not end the violence. Violence against women 6 (12):1363-1383.
Ford, David N. 1999. A behavioral approach to feedback loop dominance analysis. System
Dynamics Review 15 (1):3-36.
Forrester, J. W. 1980. Information sources for modeling the national economy. J ournal of the
American Statistical Association 75 (371):555-566.
Goodman, Lisa, and Deborah Epstein. 2005. Refocusing on Women: A New Direction for Policy
and Research on Intimate Partner Violence. J ournal of Interpersonal Violence 20
(4):479-487.
Henning, Kris, Angela Jones, and Robert Holdford. 2003. Treatment Needs of Women Arrested
for Domestic Violence: A Comparison with Male Offenders. J ournal of Interpersonal
Violence 18 (8):839-856.
Hirschel, D., and E. Buzawa. 2002. Understanding the context of dual arrest with directions for
future research. Violence against Women 8 (12):1449-1473.
Houry, Debra, and Arthur Kellermann. 2005. Arrests for intimate partner violence in female
detention patients. American journal of emergency medicine 23:96-97.
Humphries, Drew. 2002. No easy answers: public policy, criminal justice, and domestic
violence. Criminology and public policy 2 (1):91-96.
Domestic Violence 19
Kickman, Mattew J., and Brian A. Reaves. 2003. Local police departments 2000: Bureau of
Justice Statistics, U. S. Department of Justice Office Justice Statistics.
Larance, Lisa Y oung. 2006. Serving Women Who Use Force in Their Intimate Heterosexual
Relationships: An Extended View. Violence Against Women 12 (7):622-640.
Largo, T. W., P. K. Smith, J. C. Thrush, N. McCrohan, J. Rafferty, and A. Christensen. 1999.
Violence in the lives of Michigan Women: Results of a statewide survey: Michigan
Department of Community Health.
McMahon, M., and E. Pence. 2003. Making social change reflections on individual and
institutional advocacy with women arrested for domestic violence. Violence against
Women 9 (1):47-74.
Menjivar, Cecilia, and Olivia Salcido. 2002. Immigrant women and domestic violence: common
experiences in different countries. Gender & society 16 (6):898-920.
Miller, S. L. 2001. The paradox of women arrested for domestic violence. Violence against
Women 7 (12):1339-1376.
Miller, Susan L., and Michelle L. Meloy. 2006. Women's Use of Force: Voices of Women
Arrested for Domestic Violence. Violence Against Women 12 (1):89-115.
Mills, L. G. 1998. Mandatory arrest and prosecution policies for domestic violence - A critical
literature review and the case for more research to test victim empowerment approaches.
Criminal Justice and Behavior 25 (3):306-318.
———. 1999. Killing her softly: Intimate abuse and the violence of state intervention. Harvard
Law Review 113 (2):550-613.
Mullender, A. 1996. Rethinking domestic violence: The social work and probation response.
New Y ork: Routledge.
Osthoff, Sue. 2002. But, Gertrude, I Beg to Differ, a Hit Is Not a Hit Is Not a Hit: When Battered
Women Are Arrested For A ssaulting Their Partners. Violence Against Women 8
(12):1521-1544.
Pence, E. L., and C. McDonnell. 1999. Developing policies and protocols. In Coordinating
community responses to domestic violence: Lessons from Duluth and beyond, edited by
M.F. S.a. E. L. Pence.
Pennsylvania Coalition Against Domestic Assault. 2002. Accountability: Program standards for
batterer intervention services 1992 [cited November 23 2002]. Available from
http://coment.org/hisc/standards/pcadv.html.
Rajah, Valli, Victoria Frye, and Mary Haviland. 2006. "Aren't I a Victim?": Notes on Identity
Challenges Relating to Police Action in a Mandatory Arrest Jurisdiction. Violence
Against Women 12 (10):897-916.
Renzetti, C. M. 1999. The challenge to feminism posed by women's use of violence in intimate
relationships. In New versions of victims: Feminists struggle with the concept, edited by
S. Lamb. New Y ork, NY: New Y ork University Press.
Schechter, S. 1982. Women and male violence: The visions and struggles of the battered
women's movement. Boston: South End Press.
Sherman, L. W., and R. A. Berk. 1984. The specific deterrent effects of arrest for domestic
assault. American Sociological Review 49:261-272.
Shim, Woochan S., and Myung Jin Hwant. 2005. Implications of an arrest in domestic violence
cases: learning from Korean social workers' experiences in the U.S. J ournal of family
violence 20 (5):313-328.
Domestic Violence 20
Sterman, J. D. 2000. Business dynamics: Systems thinking and modeling for a complex world:
Irwin McGraw-Hill.
Stuart, Gregory L., Todd M. Moore, Kristina Coop Gordon, Susan E. Ramsey, and Christopher
W. Kahler. 2006. Psychopathology in Women Arrested for Domestic Violence. J ournal
of Interpersonal Violence 21 (3):376-389.
Stuart, Gregory L., Todd M. Moore, Julianne C. Hellmuth, Susan E. Ramsey, and Christopher
W. Kahler. 2006. Reasons for Intimate Partner Violence Perpetration Among Arrested
Women. Violence Against Women 12 (7):609-621.
Swan, Suzanne C., and David L. Snow. 2002. A Typology of Women's Use of Violence in
Intimate Relationships. Violence Against Women 8 (3):286-319.
———. 2003. Behavioral and psychological differences among abused women who use violence
in intimate relationships. Violence against Women 9 (1):75-109.
———. 2006. The Development of a Theory of Women's Use of Violence in Intimate
Relationships. Violence Against Women 12 (11):1026-1045.
Tjaden, P., and N. Thoennes. 1998. Prevalence, incidence, and consequences of violence against
women: Findings from the National Violence Against Women Survey: National Institute
of Justice and Centers for Disease Control and Prevention.
World Health Organization. 2005. WHO multi-country study on domestic violence against
women: summary report of initial results on prevalence, health outcomes and women's
responses. Geneva, Switzerland.
Yin, R. K. 1994. Case study research: Design and methods. Thousand Oaks, CA: SAGE
Publications.
Zanipatin, Jeanette, Stacy Shaw Welch, Jean Yi, and Patty Bardina. 2005. Immigrant women and
domestic violence. In Race, culture, psychology & law, edited by K. H. Barrett and W. H.
George. London: Sage Publications.
Zorza, J., and L. Woods. 1994. Analysis and policy implications of the new domestic violence
studies: National Center on Women and Family Law.
Domestic Violence
Appendix A: Data
Table 1
Distribution of Arrests by Gender, First/Repeat Arrest and Y ear
Year
1 2 3 4
First arrests of women
Number 19 26 28 26
Percent of first arrests 16% 24% 28% 29%
Repeat arrests of women
Number 2 3 4 8
Percent of total repeat arrests 13% 13% 13% 23%
Total arrests of women
Number 21 29 32 34
Percent of total arrests 15% 22% 25% 27%
First arrests of men
Number 101 81 71 65
Percent of first arrests 84% 76% 72% 71%
Repeat arrests of men
Number 14 20 27 27
Percent of repeat arrests 88% 87% 87% 77%
Total arrests of men
Number 115 101 98 92
Percent of total arrests 85% 78% 75% 73%
Total first arrests 120 107 99 91
Total repeat arrests 16 23 31 35
Total arrests 136 130 130 126
21
Domestic Violence 22
Appendix B: Model Equations
Listing of equations for model
SR RR RRR RKC
.Control
SCC RR KKK CARR
(01) FINAL TIME =10
Units: Y ear
The final time for the simulation.
(02) INITIAL TIME =0
Units: Y ear
The initial time for the simulation.
(03) SAVEPER =
TIME STEP
Units: Y ear
The frequency with which output is stored.
(04) TIME STEP =0.0078125
Units: Y ear
The time step for the simulation.
SRR KR KK RRR KCK KCK
.Cooperation
GC KR RRR RRR KK
(05) AT decrease in cooperation=
0.5
Units: Y ear
Adjustment time to decrease cooperation.
(06) AT to increase cooperation=
Baseline AT to increase cooperation *
(1+STEP(Effect size of coordination intervention, Time of coordination efforts))
Units: Y ear
Adjustment time to increase cooperation between police and victim advocates.
Coordination interventions facilitate the development of relationships, and thereby
shorten the time that it takes to increase cooperation.
(07) Baseline AT to increase cooperation=
1
Units: Y ear
Adjustment time to increase cooperation in the baseline phase.
Domestic Violence 23
(08)
(09)
(10)
Cooperation= INTEG (
(Max cooperation - Cooperation)/AT to increase cooperation -
Cooperation/AT decrease in cooperation *
Effect of women arrests on cooperation (First arrest of women/Initial arrest rate of
women),
Initial cooperation)
Units: Cooperation
Cooperation refers to the strength of positive professional relationships between different
sectors, for example, between domestic violence advocates and police departments.
Cooperation is treated as an abstract value between 0 and 100, with 0 representing no
cooperation and 100 representing the highest level of possible cooperation. Cooperation
increases naturally through social interaction between agencies and relationship building.
Cooperation also decreases naturally through staff tumover in agencies. Cooperation also
decreases faster if First arrest of women increases relative to the initial rate of first
arrests.
Effect of women arrests on cooperation(
[(-0.02,0)-
(2,2)],(0,0.4),(0.270336,0.464912),(0.501529,0.561404),(0.770703,0.72807),(1,1)
(1.23401, 1.25439),(1.46177,1.41228), (1.66024, 1.54386),(1.8085,1.61404),
(1.99388, 1.66667))
Units: Dmnl
Effect that arrest of women has on cooperation between victim advocates and police
departments. Increasing number of arrests above the initial value leads to an increase in
the rate that cooperation decreases.
2.0
Effect of first arrest of women on
decrease in cooperation
b
°
0.0 +
0 1 2
First arrest of women/Initial arrest of women
Initial cooperation= INITIAL(
Domestic Violence 24
Max cooperation * AT decrease in cooperation/
(AT decrease in cooperation + Baseline AT to increase cooperation))
Units: Cooperation
Initial value of cooperation assuming that the system is in equilibrium.
Max cooperation=
100
Units: Cooperation
Maximum value of cooperation.
SC RRR RRR
.Crossovers.
SC RRR RK
(12)
(13)
(14)
Baseline hazard rate of crossover=
0.705
Units: Dmnl/Y ear
Initial risk of cases crossing over from men being at risk of arrest to women being at risk
of arrest.
Crossovers=
HRR of cooperation on crossovers *
Hazard rate of crossover *
Cases with men at risk and priors
Units: Cases/Y ear
Cases of DV where the risk of arrest crosses over from men being at risk of arrest to
women being at risk of arrest.
Effect of cooperation on crossovers(
[(0,0)-(2,2)],(0,1.5),(1,1),(2,0.25))
Units: Dmnl
Effect of cooperation on crossovers. The greater the level of cooperation the more
effective the communication and reduction on crossovers.
Domestic Violence 25
(16)
N
°
Effect of cooperation on crossovers
Pr
°
S
°
1 2
Cooperation/Initial cooperation
°
Hazard rate of crossover=
Baseline hazard rate of crossover *
(1 + STEP(Effect size of advocacy intervention, Time of victim advocacy
intervention))
Units: Dmnl/Y ear
Risk of cases crossing over from men being at risk of arrest to women being at risk of
arrest.
HRR of cooperation on crossovers=
Effect of cooperation on crossovers(C ooperation/Initial cooperation)
Units: Dmnl
The Hazard Rate Ratio (HRR) of cooperation on crossovers refers the effect that
cooperation has on crossovers resulting from the ability of advocates, police, prosecutors,
and other professional to effectively communicate and educate each other to lower the
risk of crossovers. This includes communicating between sectors about previous cases as
a means to educate other professionals as well as advocate on behalf of a specific client.
Time at risk=
10
Units: Y ear
Average time of cases at risk.
Domestic Violence 26
GE RKC RKC KK
Interventions
SC CR RRR RRR CK
(18)
(21)
Effect size of advocacy intervention = -0.094
Units: Dmnl
The effect of advocacy intervention on reducing crossovers. Negative values lower the
risk of cases crossing over, while positive values increase the risk of cases crossing over.
Effect size of arrest intervention = 1.932
Units: Dmnl
The fraction amount of increasing arrests relative to the baseline. Positive values increase
the rate of arrests, while negative values decrease the risk of arrest.
Effect size of coordination intervention = -0.59375
Units: Dmnl
Fractional amount of increasing adjustment time. Positive values increase the adjustment
time, while negative values decrease the adjustment time.
Time of arrest intervention = 1
Units: Y ear
Time of mandatory arrest intervention.
Time of coordination efforts =3
Units: Y ear
Time of the coordination intervention.
Time of victim advocacy intervention = 3
Units: Y ear
Time of the victim advocacy intervention.
SR RK KKK RKC KK
.Risk categories
SC CK RRR RK
(24)
(25)
Baseline hazard rate of arrest = 0.097
Units: Dmnl/Y ear
The initial risk of arrest at the start of the simulation in equilibrium.
Cases entering with men at risk of arrest =
Initial cases entering with men at risk*
Effect of arrests on new cases(T otal arrests men/Initial arrests of men)
Units: Cases/Y ear
New cases of DV where men at risk of domestic violence arrest with no priori contact
with the prosecutor's office
Domestic Violence 27
(26)
(28)
Cases exiting with men at risk and no priors =
Cases with men at risk and no priors/Time at risk
Units: Cases/Y ear
Cases of DV leaving the risk category of men at risk of arrest before an arrest, i.e., prior
to an arrest. This represents cases that are leave the risk category of arrest because one or
more individuals leave the jurisdiction, violence has shifted to non-criminal behaviors, or
ceased.
Cases exiting with men at risk and priors =
Cases with men at risk and priors *
HRR of time at risk for cases with priors/Time at risk
Units: Cases/Y ear
Cases of DV leaving the risk category of men at risk of arrest with prior arrests. This
represents cases that are leave the risk category of arrest because one or more individuals
leave the jurisdiction, violence has shifted to non-criminal behaviors, or ceased.
Cases exiting with women at risk and no priors =
Cases with women at risk and no priors *
HRR of women on time at risk/Time at risk
Units: Cases/Y ear
Cases of DV leaving the risk category of women at risk of arrest with women not having
any priors. This represents cases that leave the risk category of arrest because one or
more individuals leave the jurisdiction, violence has shifted to non-criminal behaviors or
ceased.
Cases with men at risk and no priors= INTEG (
Cases entering with men at risk of arrest -
Cases exiting with men at risk and no priors -
First arrest of men,
(Initial cases entering with men at risk * Time at risk)/
(1 + Baseline hazard rate of arrest * Time at risk))
Units: Cases
Cases of DV with men at risk of arrest for criminal domestic violence.
Cases with men at risk and priors= INTEG (
First arrest of men -
Cases exiting with men at risk and priors - Crossovers,
(First arrest of men * Time at risk) /
(HRR of time at risk for cases with priors +
Baseline hazard rate of crossover * Time at risk))
Units: Cases
Cases of DV with men at risk of arrest and having prior domestic violence arrests.
Cases with women at risk and no priors= INTEG (
Domestic Violence 28
Crossovers + New cases of women -
Cases exiting with women at risk and no priors - First arrest of women,
-((Crossovers +New cases of women ) * Time at risk)/
(HRR of women on time at risk + Baseline hazard rate of arrest *
HRR of women for arrest * Time at risk))
Units: Cases
Cases of DV where women are at risk of arrest and do not have any priors. These can
include cases that involve men with priors as might happen when cases crossover, or if
cases enter the stock with no priors.
Effect of arrests on new cases(
[(0,0.8)-(2,2)],(0,1.2),(1,1),(2,0.8))
Units: Dmnl
Represents the effect that total arrests of men has on new cases of DV entering the
category of being at risk of arrest. The more total arrests of men there are, the more men
are deterred from escalating to criminal domestic violence.
Effect of arrests of men
on new cases
Total arrests of men/ Initial arrests of men
Effect of cooperation on arrests(
[(0,0.8)-(2,2)],(0,1),(1,1), (1.18654, 1.10526),(1.43731,1.24561), (1.65138,
1.32456), (1.8104,1.36842),(1.99388,1.38596))
Units: Dmnl
The effect of cooperation on arrests represents the impact that cooperation has on
reporting of domestic violence by victims to police. Better cooperation leads to more
reporting of domestic violence cases. The model assumes, however, that the initial level
of reporting in the community is independent of cooperation.
Domestic Violence 29
(34)
(36)
(37)
N
°
Effect of cooperation on arrests
r
fo}
<4
°
°
H
N
Cooperation/Initial cooperation
First arrest of men =
Hazard rate of arrest of men * Cases with men at risk and no priors
Units: Cases/Y ear
First arrest of men
First arrest of women =
HRR of women for arrest *
Hazard rate of arrest *
Cases with women at risk and no priors
Units: Cases/Y ear
The first arrest women in cases of DV
Hazard rate of arrest =
Baseline hazard rate of arrest*
(1+STEP(Effect size of arrest intervention, Time of arrest intervention))
Units: Dmnl/Y ear
This represents the overall hazard rate of arrest for both women and men. The
introduction of a mandatory arrest policy is represented as a step function, where the
effect size of the arrest intervention is the fraction above the baseline hazard rate.
Hazard rate of arrest of men =
Hazard rate of arrest *
Effect of cooperation on arrests(Cooperation/Initial cooperation)
Units: Dmnl/Y ear
Refers to risk of men being arrested for domestic violence, which is a function of the
overall hazard rate of arrest multiplied by the impact of cooperation on the hazard rate.
Cooperation affects the hazard rate of arrest by increasing the likelihood of victims
making reports.
Domestic Violence 30
(38)
(39)
(40)
(41)
(46)
HRR of time at risk for cases with priors = 1.259
Units: Dmnl
The ratio of the hazard rate of cases exiting if there is a prior.
HRR of women for arrest = 1.816
Units: Dmnl
The ratio of the hazard of arrest of women to the general hazard of arrest.
HRR of women on time at risk = 5.41
Units: Dmnl
The ratio of the hazard rate of cases exiting if women are at risk to the general rate of
leaving.
Initial arrest rate of women = INITIA L(First arrest of women)
Units: Cases/Y ear
The initial rate of women being arrested at the start of the simulation.
Initial arrests of men = INITIAL(Total arrests men)
Units: Cases/Y ear
Initial arrests of men at the start of the simulation.
Initial cases entering with men at risk = 69.7
Units: Cases/Y ear
The initial rate that DV cases enter the men at risk category.
New cases of women = 0
Units: Cases/Y ear
New cases of women at risk of domestic violence arrest with no priori contact with the
prosecutor's office
Repeat arrest of men=
Hazard rate of arrest of men * Cases with men at risk and priors
Units: Cases/Y ear
Repeat arrests of men
Total arrests men=
First arrest of men + Repeat arrest of men
Units: Cases/Y ear
Total number of arrests of men
