Psychiatric Symptoms and Community Violence Among High-Risk
Patients: A Test of the Relationship at the Weekly Level
Jennifer L. Skeem
University of California, Irvine
University of Pittsburgh School of Medicine
Institute of Psychiatry
Edward P. Mulvey
University of Pittsburgh School of Medicine
Ohio State University
University of Massachusetts Medical Center
Given the availability of violence risk assessment tools, clinicians are now better able to identify
high-risk patients. Once these patients have been identified, clinicians must monitor risk state and
intervene when necessary to prevent harm. Clinical practice is dominated by the assumption that
increases in psychiatric symptoms elevate risk of imminent violence. This intensive study of patients
(N ? 132) at high risk for community violence is the first to evaluate prospectively the temporal relation
between symptoms and violence. Symptoms were assessed with the Brief Symptom Inventory and
threat/control override (TCO) scales. Results indicate that a high-risk patient with increased anger in 1
week is significantly more likely to be involved in serious violence in the following week. This was not
true of other symptom constellations (anxiety, depression, TCO) or general psychological distress. The
authors found no evidence that increases in the latter symptoms during 1 week provide an independent
foundation for expecting violence during the following week.
Keywords: mental illness, symptoms, violence, risk, anger
Over the past few decades, researchers have grappled with the
complex relationship between mental disorder and violence. Many
studies that differ methodologically converge on the message that
the two constructs are positively related, although “the relative
contribution of mental illness to the overall rate of violence in
society is quite small” (Norko & Baranoski, 2005, p. 21). Two
large community-based studies have indicated that mental disor-
ders are particularly likely to increase violence risk when they
co-occur with substance abuse disorders (Monahan, Steadman, et
al., 2001; F. Swanson, Holzer, Ganju, & Jono, 1990). Neverthe-
less, the nuances of the relationship between mental disorder and
violence remain elusive.
First, it is unclear whether mental disorders, specific symptom
constellations, or both contribute to violence. Most researchers
have addressed only whether psychiatric diagnoses predict vio-
lence (Hodgins, 1992; F. Swanson et al., 1990), despite sugges-
tions that psychiatric symptoms more strongly predict violence
(Link & Stueve, 1994; cf. Appelbaum, Robbins, & Monahan,
2000). This distinction matters because many diagnoses have
similar symptoms and because symptoms may be easier to accu-
rately assess and monitor than diagnoses. Second, and more im-
portant, the temporal relationship between mental disorder and
violence has yet to be firmly established. Most researchers have
examined how well an assessment of mental disorder at a partic-
ular point in time (e.g., during hospitalization) predicts whether
violence occurs during a relatively long follow-up period (e.g., a
year after release). A few researchers have demonstrated that
snapshots of active psychiatric symptoms (e.g., delusions) predict
violence more strongly than static diagnostic variables (e.g.,
schizophrenia; Link, Andrews, & Cullen, 1992). However, longi-
tudinal research has yet to show that increases in psychiatric
symptoms precede and increase the likelihood of proximate vio-
lence (see Kraemer, Kazdin, Offord, & Kessler, 1997).
Risk Status, Risk State, and Symptom Acuity
Greater clarity regarding the timing and specificity of the rela-
tionship between mental illness and violence is needed to advance
theoretical and practical understanding of this issue. Such clarity
Jennifer L. Skeem, Department of Psychology and Social Behavior,
University of California–Irvine; Carol Schubert and Edward P. Mulvey,
Department of Psychiatry, University of Pittsburgh School of Medicine;
Candice Odgers, Institute of Psychiatry, London, England; William Gard-
ner, Department of Pediatrics, Ohio State University; Charles Lidz, De-
partment of Psychiatry, University of Massachusetts Medical Center.
Data collection was supported by National Institutes of Mental Health
Grant R01 MH40030-13. We thank the following interviewers and staff on
this project: Brenda Cappy, Shawn Ellies, Kristen Eshman, Gordon Hod-
nett, Jennifer King, Debra Murray, Dennis Webster, and Jane Zoltun.
Correspondence concerning the article should be addressed to Jennifer
L. Skeem, Department of Psychology and Social Behavior, 3311 Social
Ecology II, University of California–Irvine, Irvine, CA 92697-7085.
Journal of Consulting and Clinical Psychology
2006, Vol. 74, No. 5, 967–979
Copyright 2006 by the American Psychological Association
0022-006X/06/$12.00 DOI: 10.1037/0022-006X.74.5.967
would have implications for assessing risk for violence. Currently,
psychiatric variables contribute very little in assessing long-term
risk of violence compared with the contribution of such variables
as past violence (Bonta, Law, & Hanson, 1998). However, psy-
chiatric measures show promise for assessing imminent risk of
violence, given the relationship between active symptoms and
violence (Norko & Baranoski, 2005). A patient’s psychiatric
symptoms vary in their level of acuteness over time or movement
away from his or her baseline. Similarly, violence risk ebbs and
flows over time, even in high-risk patients (see Douglas & Skeem,
2005; Skeem & Mulvey, 2002). Nevertheless, no researchers have
examined directly the dynamic interplay between psychiatric state
and violence risk state.
Most research to date has advanced our understanding of risk
status. Clinicians are in a relatively good position to identify
high-risk patients through the use of increasingly sophisticated
violence risk assessment tools that incorporate historical variables
(Gardner, Lidz, Mulvey, & Shaw, 1996; Harris, Rice, & Quinsey,
1993; Monahan, Steadman, Appelbaum, et al., 2005; Webster,
Douglas, Eaves, & Hart, 1997). Once these high-risk patients are
identified, however, clinicians must determine how to monitor risk
state and intervene when necessary to prevent harm. In doing so,
clinicians are most likely to focus on psychiatric variables, given
their mental health training, setting, and resources. This might be
an appropriate focus. Although psychiatric variables do not con-
tribute greatly to assessing patients’ violence risk status, they may
offer important insights into violence risk state. Thus, there should
be substantial practical interest in determining whether and how
changes in mental health (general distress or particular symptom
constellations) increase risk of proximate violence among individ-
uals who are deemed high risk.
To date, the assumption that deterioration in psychiatric symp-
toms increases patients’ risk of imminent violence has dominated
legal standards and clinical practice. A chief criterion for invol-
untary hospitalization is a judgment that mental disorder creates a
“danger to self or others” (Monahan, 1996). Several standards for
outpatient commitment require patients to accept treatment if
decision makers believe there is a potential for psychiatric deteri-
oration that will lead to violence in the future (Appelbaum, 2001;
Monahan, Bonnie, et al., 2001). In practice, patients at high risk for
violence are provided with standard psychotropic and psychosocial
treatment, perhaps at a more intensive level than usual; treatment
rarely targets violence risk per se (Skeem, 2003). The assumption
is that provision of standard treatment will reduce symptoms,
which will reduce violence risk. Although involvement in standard
treatment relates to reduced violence potential (Monahan, Stead-
man, et al., 2001), it is wholly unclear whether treatment involve-
ment reduces symptoms, which reduces violence potential.
Clearly, there is a need to test the assumption on which so much
policy and practice rests.
In determining the relationship between psychiatric state and
violence risk state, one must assess not only the acuity of an
individual’s symptoms (the level of the reported symptom), but
also the symptom type. Particular types of symptoms (e.g., de-
pressed mood) are included in multiple diagnostic categories (e.g.,
major depressive and schizoaffective disorder). Given polythetic
criteria for diagnostic categories, patients who manifest different
symptom constellations can qualify for the same diagnostic cate-
gory (e.g., schizophrenia). Moreover, symptom clusters within an
individual in a particular diagnostic category can change dramat-
ically over time (e.g., during an acute vs. residual phase). Thus,
two individuals with bipolar disorder may appear quite different
from one another on the basis of their individual symptom con-
stellations and their clusters of “currently active” symptoms. To
account for these factors, researchers need to focus on the smaller
units on which a diagnostic category is built and recognize that
these units may change over time within an individual.
A small body of research is responsive to this need. Some
investigators have included symptom type and acuity in their
studies of violence risk. That work suggests that both of these
symptom dimensions are more predictive of violence than diag-
nostic status: Patients who were high in thinking disturbance,
hostility–suspiciousness, and agitation–excitement at hospital ad-
mission were more likely to be assaultive during their hospital stay
(Lowenstein, Binder, & McNiel, 1990; McNeil & Binder, 1994).
Other investigators have focused more narrowly on specific
types of symptoms (Link et al., 1992; Taylor, 1998), the most
common of which are threat/control-override (TCO) symptoms
(Link & Stueve, 1994). TCO symptoms ostensibly are delusional
beliefs that someone is seeking to do one harm (threat) or that
outside forces are controlling one’s mind (control override). Tra-
ditionally, these symptoms are measured on the basis of a small set
of self-report questions. Using such measures, Link, Stueve, and
Phelan (1998) found that both threat and control override symp-
toms were independently related to an increased risk for violence.
Similarly, J. Swanson and colleagues (1997) found that patients
who endorsed TCO symptoms were twice as likely to become
violent during a 1-year posthospitalization period than those who
Nevertheless, the link between TCO symptoms and violence is
far from settled. Appelbaum et al. (2000) attempted to replicate the
association previously found between TCO delusions and violence
on the basis of a large prospective study of psychiatric patients.
Despite multiple modes of analysis, the authors found that the
presence of delusions in general and TCO delusions in particular
was not associated with future violence. They were able to repli-
cate previous findings only when the methodological limitations of
earlier studies were mimicked (i.e., when self-report responses to
TCO questions were considered without an independent assess-
ment of whether such beliefs were actually delusional; when past
rather than future violence was considered). Further analyses sug-
gested that increased violence risk was attributable not to TCO
delusions but perhaps to a hostile, suspicious personality style
tapped by the TCO self-report questions. Similarly, Estroff (cited
in Swanson et al., 1997) found an inverse association, and J.
Swanson, Borum, Swartz, and Hiday (1999) found no association
between TCO symptoms and violence.
Together, these studies indicate that symptom acuity is more
important in predicting violence than diagnostic status. They pro-
vide only limited support for the notion that particular types of
symptom constellations predict community violence. Although
these studies looked beyond broad diagnostic categories to focus
on symptom type and acuity, they were unable to assess the crucial
SKEEM ET AL.
relationship between psychiatric state and violence risk state. Such
an assessment of within-person effects requires a longitudinal
design with repeated observations.
The Present Study
In this article, we report results from an intensive longitudinal
study of individuals selected because of their high potential for
repeated involvement in violence after visiting a psychiatric emer-
gency room. We refer to these participants as high-risk patients
because this label best conveys the group to whom the sample may
generalize. The vast majority (93%) of participants received psy-
chiatric treatment after their visit to the emergency room, and the
majority were still involved in treatment 6 months later (Skeem,
Mulvey, Lidz, Gardner, & Schubert, 2002). We focus on this small
group of patients at high risk for violence because the group
accounts for the majority of violence among individuals with
mental illness (see Skeem et al., 2004). Even for these high-risk
patients, however, proximate violence is not a foregone conclu-
sion. For example, sophisticated analyses of data from the
MacArthur Violence Risk Assessment Study yielded an actuarial
tool that predicts proximate violence well (Monahan, Steadman,
Appelbaum, et al., 2005). Nevertheless, of patients identified as
high risk by this tool, at least 24% (derivation sample) and up to
65% (cross-validation sample) were not violent during the 20
weeks after release from the hospital (Banks et al., 2004; Mona-
han, Steadman, Robbins, et al., 2005). Our goal in the present
study was to assess whether and how symptom fluctuations may
help explain why only a fraction of high-risk patients engage in
In this study, a group of individuals identified by an actuarial
tool as high risk during a psychiatric emergency room visit was
followed into the community. These patients and collateral infor-
mants completed interviews every week for 6 months to assess
their psychiatric state and involvement in violence. This design
permits a specific examination of the relationship between symp-
tom acuity and type on the one hand and violence on the other at
the weekly level. To date, this type of intensive temporal measure-
ment has been absent from the debate regarding the nature of the
relationship between symptoms and violence. The debate has
instead been based on static snapshots of symptoms in relation to
violence over relatively long periods.
We examined the relationship between symptoms and violence
in three ways. First, we examined the data in a “static” manner like
that used in past research; that is, we assessed the association
between general symptom acuity and violence across the full
follow-up period. Second, we assessed the concurrent relationship
between symptoms and violence during the same week. Then, we
assessed the time-ordered relationship between symptoms during 1
week and violence during the next week. This time-ordered rela-
tionship is the primary relation of interest for policy and practice.
A secondary focus of the current investigation was to inform the
ongoing debate about the utility of the TCO items by assessing (a)
whether TCO symptoms fluctuate over time (like symptoms of
psychosis) or remain static (like personality traits), and (b) how
strongly TCO state relates to violence risk state.
In this study, we focused on the relationship between symptoms and
violence within individuals over time. This approach requires a group of
individuals whose symptoms and involvement in violence vary over time.
This means that we needed to identify individuals who were likely to
engage in multiple violent incidents over a relatively short period of time.
Using a two-stage screening process that involved a medical record review
and a subsequent screening interview, potential participants for this study
were identified from among psychiatric emergency room patients in a
large, university-based psychiatric hospital in an urban area. Enrollment
criteria were adapted from an actuarial prediction model developed by
Gardner et al. (1996) and included (a) young age (14–30 years), (b) a
history of violence and recent violence in the 2 months prior to the
emergency room visit, (c) recent and heavy substance use, (d) a score of 7
or higher on the Brief Symptom Inventory (BSI; Derogatis & Melisaratos,
1983) Hostility subscale, and (e) absence of current thought disorder (i.e.,
current diagnosis of schizophrenia or current report of delusions). The
latter criterion, which was empirically derived (Gardner et al., 1996), is
consistent with other research indicating that symptoms of psychosis,
although often clinically relevant for understanding violence in a small
proportion of individuals, are generally weak predictors of violence when
applied to broad samples of mentally ill individuals (e.g., Monahan, Stead-
man, et al., 2001; Wessely, Castle, Douglas, & Taylor, 1994; for a review,
see Douglas & Skeem, 2005). Those patients meeting these criteria were
invited to participate in weekly interviews in the community for 26 weeks
after their baseline interview.
Hospital records for a pool of 3,356 patients were reviewed to complete
the first of two stages of screening in the emergency room. Of those
individuals, 1,004 were deemed eligible in Stage 1, and attempts were
made to invite them to the second stage of screening. Among the individ-
uals eligible to participate in Stage 2, 20% refused to participate and 31%
could not be located. The remaining 517 individuals completed the second
stage of screening, and 171 of those were deemed eligible for study
participation. Of those individuals, 89% enrolled in the study.
Our final sample comprised 132 individuals who completed 92% of their
weekly follow-up interviews. Details concerning the enrollment process
and sample characteristics as well as the methods involved in conducting
weekly interviews are presented elsewhere (Schubert, Mulvey, Lidz, Gard-
ner, & Skeem, 2005; Skeem et al., 2002). Briefly, the sample comprised
young (M ? 21 years, SD ? 6) men (48%) and women (52%) who were
equally likely to be White or African American (49%; “other” ? 2%). Of
the 83 research participants aged 18 and older, 65% had attained at least a
high school diploma, and one third lived with their parents. Research
participants had (nonprimary) hospital chart diagnoses of affective disor-
ders (76%; chiefly major depression and bipolar disorders), psychotic
disorders (11%; chiefly schizoaffective and psychosis not otherwise spec-
ified [NOS]), and other Axis I disorders (57%; chiefly anxiety disorders).
Of participants, 45% had comorbid Axis I and substance abuse disorders.
They had an average of 1.7 prior psychiatric hospitalizations (SD ? 2.5),
and 60% had a recorded history of attempted suicide.
BSI (Derogatis & Melisaratos, 1983).
toms was obtained each week using the BSI, a 53-item self-report inven-
tory in which participants rate on a 5-point scale ranging from 0 (not at all)
to 4 (extremely) the extent to which they have been bothered during the
past week by various symptoms. The BSI includes three scales that capture
global psychological distress, including the General Severity Index (GSI).
The BSI is designed to include nine subscales that assess individual
Information regarding symp-
PSYCHIATRIC SYMPTOMS AND COMMUNITY VIOLENCE
symptom constellations: Somatization, Obsessive–Compulsive, Interper-
sonal Sensitivity, Depression, Anxiety, Hostility, Phobic Anxiety, Paranoid
Ideation, and Psychoticism. An individual’s score on each subscale is his
or her average for items included in that subscale.
The scales and subscales of the BSI demonstrated good internal consis-
tency (? ? .71–.85), and test–retest reliability (r ? .68–.91) in the
normative study (Derogatis & Melisaratos, 1983) and good internal con-
sistency in the present research (subscale ?s ? .76–.91; GSI ? ? .97).
With respect to validity, the BSI generally manifests a theoretically coher-
ent pattern of association with the scales of the Minnesota Multiphasic
Personality Inventory (Derogatis & Melisaratos, 1983). Several studies
have indicated that the BSI is sensitive to change (Benishek, Hayes,
Bieschke, & Stoffelmayr, 1998; Carscaddon, George, & Wells, 1990;
Holden, Starzyk, McLeod, & Edwards, 2000; Pekarik, 1983; Piersma,
Boes, & Reaume, 1994; Piersma, Reaume, & Boes, 1994).
Given such positive psychometric properties, most researchers view the
BSI as an appropriate measure of general psychopathology and psycho-
logical distress (Benishek et al., 1998; Bonynge, 1993; Boulet & Boss,
1991; Brophy, Norvell, & Kiluk, 1988; Hafkenscheid, 1993; Hayes, 1997;
Heinrich & Tate, 1996; Piersma, Boes, & Reaume, 1994; Ruiperez, Ibanez,
Lorente, Moro, & Ortet, 2001). Nevertheless, the utility of the BSI at the
discrete symptom or item level is questionable (cf. Benishek et al., 1998).
Multiple factor analytic and other studies have suggested that the discrimi-
nant validity of several BSI subscales is poor. Although researchers have
identified anywhere from one to seven factors for the BSI, four factors
repeatedly emerge across most factor analytic studies: depression, anxiety,
somatization, and hostility (Brophy et al., 1988; Gavazzi, Julian, & Mc-
Kenry, 1996; Hafkenscheid, 1993; Hayes, 1997; Ruiperez et al., 2001;
summary table is available from the authors).
Exploratory factor analysis of data collected from all patients screened in
the present study indicated a highly similar four-factor solution. Confir-
matory factor analyses indicated an adequate fit for this four-factor solu-
tion, ?2(248, N ? 532) ? 786, p ? .001, comparative fit index (CFI) ? .92,
root mean square error of approximation (RMSEA) ? .06, on this sample.
This four-factor solution fit better than a single-factor solution using the
same items, ?2(263, N ? 532) ? 1,732, p ? .001, CFI? .78, RMSEA ?
In the present study, we operationalized symptoms at two levels: symp-
tom constellations and general distress. First, we used scores on three of the
four scales identified to reflect patients’ depression, anxiety, and hostility
symptom constellations each week. Negative affectivity and hostility have
been clearly linked with violence in past research (for a review, see
Douglas & Skeem, 2005; see also Gardner et al., 1996). This is not the case
with somatization, and we excluded this symptom constellation from the
study as a result. Second, we used the GSI to reflect patients’ psychological
distress each week. The GSI is the average score on all 53 items of the BSI.
TCO symptoms (Link & Stueve, 1994; Link et al., 1998).
toms were assessed by asking patients to rate the following three questions:
“How often you have felt that (a) your mind was dominated by forces
beyond your control; (b) thoughts were put into your head that were not
your own; and (c) there were people who wished to do you harm?” Each
item was rated for the time frame of the past week on a 5-point scale
ranging from 0 (never) to 4 (very often). TCO scores were computed as the
average of these three items. Unlike Appelbaum et al. (2000), our measure
of TCO does not include a clinical assessment of the extent of the
delusional quality of these self-reported symptoms.
The nature, frequency, and severity of the participant’s involvement in
violent incidents was explored at each weekly interview. Each participant
was asked whether he or she had engaged in any of nine categories of
aggressive acts (e.g., pushing, hitting, using a weapon) on the basis of the
adaptation (Lidz, Mulvey, & Gardner, 1993) of the Conflict Tactic Scale
(CTS; Straus & Gelles, 1990). For each category endorsed, respondents
were asked to list the number of times the act had occurred. Contextual
information (date, location, coparticipants, injury level, outcomes) for each
incident was also recorded. On the basis of this information, we coded
violence into two levels of severity. Serious violence was defined as an
incident that resulted in a physical injury, a sexual assault, a threat made
with a weapon in hand, or an aggressive act that involved the use of a
weapon (see Steadman et al., 1998). Minor violence included aggressive
acts that did not result in injury. For the purposes of this article, we focus
on serious violence and comment when results differ for any violence.
Each week was coded to indicate whether or not a “serious” incident or
“any” incident had occurred.
Violence was converted to the weekly level (i.e., whether at least one
event occurred in the past week) from data originally assessed at the daily
level (i.e., events on particular days in the past week). This conversion was
necessary to produce a time frame consistent with that of the symptom
data. To capture the context of the violence, participants’ place of residence
during the majority of each week was coded. If more than half of the days
in the recall week were spent in the community (as opposed to jail,
hospital, or other institution), the week was considered a “community
week.” Using this classification, the majority of participants (59%) were in
the community for more than 24 weeks of the 26-week follow-up period
(M ? 22.7 weeks, SD ? 5.3). In this study, only community weeks were
included. Although the relation between symptoms and violence within
institutional contexts is also of interest to researchers and policy makers,
these settings are arguably qualitatively different from community life.
Institutional contexts vary from time in the community in the type of
individuals with whom one interacts; external constraints placed on one’s
behavior; medications and interventions to which one is subject; and
availability of substances, weapons, and other violence-relevant materials.
Therefore, the two settings are likely to yield a different set of relationships
between the two behaviors.1
Participants were screened for eligibility in the psychiatric emergency
room, using a two-stage record review and screening interview process.
Eligible patients identified as high risk were recruited for the study and
followed into the community. We attempted to interview recruited partic-
ipants every week for 6 months after their enrollment in the study, totally
to a baseline interview and 26 weekly interviews. When 2 or fewer
interviews were missed, research associates extended the recall period for
violence in the next interview to include the time period that was missed.
Thus, in a small number of instances, the violence recall period was 14
days instead of 7 days. When 5 or more consecutive interviews were
missed, the participant was dropped from the study and replaced by a new
participant. Approximately 20 participants were dropped for this reason.
Interviews covered involvement in violence and changeable risk factors for
violence, including the participant’s living situation, employment or school
activities, positive and negative social support, relationship quality, treat-
ment involvement, contacts with the legal system, drug and alcohol use and
symptoms, and mental health symptomatology. Interviews required ap-
proximately 1 hr to complete, and participants were paid $10.
1Weekly violence data were coded as missing when the week was
classified as a community week but the violence that occurred during that
week occurred within an institution. This affected less than 1% of the total
weeks represented in this study. The fact that violence in institutions was
so rare counters concern that excluding institutional violence would
weaken the apparent relation between psychiatric deterioration and vio-
lence. Indeed, we found that patients were rarely placed in an institution
even after a violent incident had occurred; this happened after only 2% of
incidents (Lidz et al., 2006).
SKEEM ET AL.
A collateral informant for each participant was interviewed on the same
schedule. These informants were chosen on the basis of the participant’s
nomination of individuals who knew him or her well, and information was
obtained weekly on each individual’s frequency and duration of contact
with, and judged closeness to, the participant. If a collateral informant had
no contact with the participant or had no new knowledge about him or her
during a given week, he or she did not complete an interview that week.
When a collateral had no contact with, or new information about, a
participant for 3 consecutive weeks, a new collateral who was more
familiar with the individual’s current activities was chosen to replace the
old one. On the basis of application of these rules and the ability of the
interviewers to engage collateral involvement, 73% of the follow-up in-
terviews completed with participants had an accompanying collateral in-
Although providing a complete picture, the use of multiple sources of
information can also produce conflicting reports. When the conflict con-
cerns whether a violent incident occurred, the most likely sources of error
are arguably that the event is unknown to a source (collaterals) or a source
does not wish to acknowledge the event (participants or collaterals).
Therefore, any report of the occurrence of a violent incident was assumed
to be a correct report. When the conflicts were about the details of a violent
incident (e.g., the identity of a cocombatant), a system relying on group
consensus was used to devise a “most plausible account” of the incident.
We first provide general descriptive information for psychiatric
symptoms (i.e., general psychological distress, specific symptom
clusters, and TCO symptoms) and violence. Then, we examine the
relations between these two domains in three ways. First, we
assess the relation between symptom status and violence status,
averaging (symptoms) and summing (violence) the data from each
participant’s time series. Second, we assess the concurrent relation
between symptoms and violence during the same week. Third, we
assess the time-ordered relation between symptoms during 1 week
and violence during the next week. This third examination ad-
dresses the primary aim of the study, that is, to assess the temporal
relation between symptom type and acuity on the one hand and
violence on the other at the weekly level.
To characterize the general level of distress in the sample,
summary scores were computed for each participant by averaging
his or her GSI scores across the 26-week series. We call these
person mean GSI scores. We then averaged these person mean GSI
scores across participants. The average person mean GSI score was
1.17 (SD ? 0.8). This mean score is significantly lower, t(1,
132) ? 2.2, p ? .05, than normative data provided for adult
psychiatric outpatients (M ? 1.32, SD ? 0.72; Derogatis &
Melisaratos, 1983). However, outpatients may not be an ideal
comparison group, given that 60% of the participants were not
regularly active in outpatient treatment by the time of their final
study interview. At the time of their screening interview, these
individuals obtained an average GSI score (M ? 2.15, SD ? 0.8)
that fell above the range expected for psychiatric inpatients (see
Skeem et al., 2002). Also at the beginning of the study, partici-
pants’ subscale scores generally were within the range expected of
psychiatric inpatients, although their Hostility subscale scores
were higher given that hostility was a criterion for study inclusion.
The latter finding suggests that specific symptom constellations
may be particularly characteristic of this sample. To examine the
sample’s profile across specific symptoms, the same procedures
described for GSI scores were applied to BSI symptom constella-
tion and TCO scores. Table 1 presents the descriptive information
for these variables. To interpret the data in the table, it is important
to recall that the BSI assesses the degree to which a participant is
bothered by a particular symptom, whereas the TCO score reflects
the frequency of the presence of certain beliefs. As shown in Table
1, hostility (and perhaps TCO) symptoms appear particularly char-
acteristic of the sample.
The majority of participants (61%) had at least 1 week through-
out the observation period when they engaged in serious violence,
and 87% of participants had at least 1 week when they engaged in
any violence. The sample had an average of 2 (SD ? 2.5) com-
munity weeks with serious violence and 4.9 (SD ? 4.4) commu-
nity weeks with any violence. To simplify the presentation of
results, we focus on results for serious violence, which arguably is
of greater interest to researchers and policy makers. In the rare
event that the pattern of results for any violence was different from
that for serious violence, we note the discrepancy in the following
Symptom Status and Violence Status (Across the Series)
The first set of analyses focused on risk status. These analyses
tested the static relationship between symptoms and violence on
the basis of summary data from the entire 26-week series. Specif-
ically, we assessed the relation between a participant’s symptom
level (mean across the series) and the number of serious violent
weeks in the community. As before, we examined symptoms at
both general (psychological distress) and specific (symptom con-
stellation) levels of resolution.
A Poisson regression model with block entry was applied to
assess the degree of variance in the number of serious violent
weeks that could be explained by general distress (average GSI
score), after controlling for time at risk (i.e., the number of weeks
in the community). A Poisson model was used to correct for the
skewed nature of the dependent variable and to obtain more
accurate estimates of the state-level relationships between symp-
toms and involvement in violent events (Gardner, Mulvey, &
The results indicate that participants’ average GSI score was
unrelated to the number of weeks with incidents of serious vio-
lence (? ? .04, ns). Parallel analyses were conducted to test each
Descriptive Data for Symptom Clusters
TCO ? threat and control override.
PSYCHIATRIC SYMPTOMS AND COMMUNITY VIOLENCE
of the symptom constellations. As was the case for general dis-
tress, after controlling for time at risk, average anxiety (? ? .00,
ns), depression (? ? .05, ns), and TCO (? ? .01, ns) scores were
not related to the number of serious violent incidents. However,
after controlling for time at risk in the community (? ? .81, p ?
.001), participants’ average hostility score was strongly related to
the number of serious violent incidents (? ? .56, p ? .001).
Thus, during the 6 months after visiting the emergency room,
participants’ average levels of hostility, but not their general dis-
tress or other symptoms, related to their number of serious violent
incidents. Because these analyses do not indicate how symptoms
and violence relate to one another across time, we next examined
the concurrent and serial relationships between symptoms and
violence at the weekly level.
Symptoms and Violence Within Weeks (Concurrent
We examined the concurrent relation between symptom levels
and violence by testing whether weeks that individuals reported
higher symptom levels were also more likely to be characterized
by serious violence. Results from fixed-effects logistic regression
analyses are presented in Table 2.
Fixed-effects regression methods focus on the within-person
variation and control for unmeasured stable characteristics of par-
ticipants. Allison (2005) argued that fixed-effects models provide
some of the advantages of randomized experiments because they
control for unmeasured individual differences by using each indi-
vidual as his or her own control. The odds ratios (ORs) reported in
Table 2 were derived from fixed-effects models and can be inter-
preted as the expected increase in the odds that a participant will
engage in violence during a given week if his or her symptom
score is 1 unit (i.e., 1 SD) above the series mean during that same
week. For example, the odds of engaging in serious violence
increased by 2 times during a week when a patient obtained a score
of 3 versus 2 on the GSI scale.
Significant associations were found for general distress and all
symptom clusters. These analyses indicate that both general psy-
chological distress and specific symptom constellations relate
moderately to violence during a given 1-week period. Because the
fixed-effects model uses participants as their own controls, these
results show that there was an increased likelihood of violence
during weeks when an individual’s symptoms were above his or
her series average.
Symptoms and Violence Across Weeks (Time-Ordered
Although these findings provide support for the co-occurrence
between elevated symptoms and engagement in violence within
the same week, they do not tell us what comes first. Do increasing
symptoms lead to violence? Are symptoms elevated following a
violent incident? Or, is there a reciprocal relationship between
symptoms and violence? We examined the temporal ordering of
symptom levels and violence by testing whether symptom levels in
1 week were predictive of violence in the following week, and vice
The time-ordered relationship between symptoms and violence
was examined using a structural cross-lagged longitudinal model.
This structural equation model (SEM) allowed for an examination
of the direction of the relationship between symptoms and violence
over time (the cross-lagged regression) while controlling the rela-
tion of symptoms and violence on themselves from 1 week to the
next (autoregression). This model is based on the early work of
Joreskog and colleagues (Joreskog, 1970; Joreskog & Sorbom,
1979) and has been applied within the psychological literature to
examine time-ordered data (Ferrer & McArdle, 2003; McArdle &
Bell, 2000). Given our use of Full Information Maximum Likeli-
hood to deal with missing data across occasions, the N for each of
the models was 132.
The models were fit in Mplus (Version 3.12; Muthe ´n & Muthe ´n,
2003) using the categorical and multilevel data options. These
options allowed for the integration of binary (violence) and con-
tinuous measures (BSI scores) into the SEM and took into account
the multilevel structure of the data (the nesting of occasions within
persons).2Separate models were estimated for each symptom type
(e.g., GSI general distress and specific symptom constellations of
depression, anxiety, hostility, and TCO).3For each symptom clus-
ter, the full model (as depicted in Figure 1) was fit to evaluate
whether symptoms and violence were related to each other over
time. Next, each of the cross-lagged parameter estimates (?1, ?2)
was constrained to equal 0; these models tested whether a lagged
relationship between symptoms the preceding week and violence
this week (?1) or between violence the preceding week and symp-
toms this week (?2) was needed to explain the data. Comparisons
between models were based on standard SEM fit indices, with
particular attention paid to the weighted root mean residual
(WRMR ? .90), which is recommended for use with categorical
data (Yu & Muthe ´n, 2002).
Results indicated that there were no significant cross-lagged
relationships between violence and general distress (GSI scores)
across time. The full model, ?2(7, N ? 132) ? 5.7, p ? .58, CFI ?
.99, WRMR ? .47, fit as well as a model in which the cross-lagged
2The binary measurement of violence at each week (yes/no) was ac-
counted for by using the categorical data specification and numerical
3The cross-lagged regression parameters (parameter ‘a ˜1’ or ‘a ˜2’ in
Figure 1) can be interpreted as the effect of symptoms on violence, or vice
versa, from this week to next week, after controlling for the concurrent and
autoregressive relationships. The advantage of using this type of structural
model is that the relationship between symptoms and violence can be
modeled simultaneously, while controlling for important time-dependent
(read autoregressive and concurrent) relationships.
Concurrent (Within-Week) Relationship Between Symptoms and
GSI (general distress)
index; TCO ? threat and control override.
OR ? odds ratio; CI ? confidence interval; GSI ? general severity
SKEEM ET AL.
relationships (?1and ?2) were constrained to equal 0, ?2(9, N ?
132) ? 9.9, p ? .36, CFI ? .99, WRMR ? .62 (Figure 1). Similar
results were found for depression, anxiety, and TCO. That is, there
was no evidence of a relationship between these symptom clusters
and violence from week to week.
This pattern did not hold, however, when the relation between
hostility and violence was examined. As shown in Figure 2, the
model that demonstrated the best fit when depicting the time-
ordered relation between hostility and violence allowed hostility
last week (t – 1) to influence serious violence this week (t), ?2(7,
N ? 132) ? 6.7, p ? .15, CFI ? .99, WRMR ? .55; this model
demonstrated a better fit than a model that constrained the relation
between hostility last week and violence this week equal to 0, ?2(5,
N ? 132) ? 11.2, p ? .04, CFI ? .96, WRMR ? .87. It is
important to note that the relationship between violence last week
and hostility this week (?2) was not significant in any of the
Figure 2 conveys three main findings from the SEM analyses.
First, violence was moderately related to itself from week to week
(violence 3 violence); a violent incident increased the odds of
violence occurring in the following week by 1.4 times (p ? .01).
Second, hostility was strongly related to itself from week to week
(hostility 3 hostility; ? ? .79, p ? .001). Recall from Table 1 that
the concurrent relationship between hostility and violence was
strong (hostility 4 3 violence; OR ? 2.0, p ? .02). The third and
most important finding conveyed in Figure 2 is that, after conser-
vatively taking into account these three relations, a 1-unit change
in hostility was independently related to violence the following
week (OR ? 1.2). In summary, hostility has a moderate to strong
relationship with violence during the same week. After controlling
for this relation as well as controlling for the relation of violence
to itself from week to week, hostility uniquely predicts violence
the following week.
This intensive longitudinal study of high-risk emergency room
patients is among the first attempts to disentangle the complex
relation between psychiatric symptoms and violence over time.
The results indicate that it is crucial to attend to the temporal
relation between mental disorder and violence. Considering the
full 6-month period after a visit to the emergency room, only
high-risk patients with greater average levels of hostility were
more likely to be involved in serious violence. However, during a
given week within that period, individuals with greater general
psychological distress, hostility, anxiety, depression, and TCO
symptoms tended toward co-occurring serious violence.
The temporal relation between symptoms and involvement in
violence, however, is the one of most interest for practice and
policy. This study is the first to indicate that a high-risk patient
with an increased hostility level in 1 week is significantly more
likely to be involved in serious violence in the next week. Hostility
is a dynamic risk factor for proximate violence. This is not true of
the other symptoms examined here, including general psycholog-
ical distress and specific symptom constellations other than hos-
tility (TCO symptoms, anxiety, and depression).
Anger as a Dynamic Risk Factor for High-Risk Patients
Because hostility emerged as the only leading indicator of
violence in this study, it is important to examine how it was
assessed. The BSI Hostility subscale reflects how bothered an
individual has been over the past week by five problems: feeling
easily annoyed or irritated; temper outbursts that he or she could
not control; having urges to beat, injure, or harm someone; having
urges to smash things; and getting into frequent arguments
(Derogatis & Melisaratos, 1983). A review of these items raises
First, the item content raises questions regarding criterion con-
tamination: Is it simply the case that the BSI Hostility subscale
assesses violence this week, which predicts violence next week?
That is, do the present results merely express Meehl’s maxim that
past behavior is the best predictor of the same behavior in the
future? Our results suggest this is not the case (Figure 2). Even
after controlling for the effects of whether an individual was
and hostility. T ? time; O.R. ? odds ratio for binary outcomes, estimates
are significant at p ? .05 level; ? ? standardized coefficient, continuous
Cross-lagged time series structural equation model for violence
and general psychological distress. T ? time; ? ? standardized coefficient,
Cross-lagged time series structural equation model for violence
PSYCHIATRIC SYMPTOMS AND COMMUNITY VIOLENCE
violent during 1 week (past behavior predicts future, like behavior)
and concurrent levels of hostility, hostility scores this week inde-
pendently predicted whether violence would occur next week. BSI
hostility scores, then, tap something unique—that is, not just “like
behavior”—that significantly predicts proximate violence.
Second, an examination of the BSI hostility items indicates that
this subscale may be assessing something more closely related to
anger than hostility. Although anger and hostility often are used as
interchangeable terms in the literature, there are important distinc-
tions between them (see Eckhardt, Norlander, & Deffenbacher,
2004). Hostility is an attitudinal disposition that involves “a de-
valuation of the worth and motives of others, an expectation that
others are likely sources of wrongdoing, a relational view of being
in opposition toward others, and a desire to inflict harm or see
others harmed” (Smith, 1994, p. 26). In contrast, anger may be
defined as “an unpleasant emotion ranging in intensity from irri-
tation or annoyance to fury or rage” (Smith, 1994, p. 25). More
broadly, anger involves “a constellation of specific uncomfortable
subjective experiences and associated cognitions (i.e., thoughts,
beliefs, images, etc.) that have various associated verbal, facial,
bodily, and autonomic reactions” (Kassinove & Sukhodolsky,
1995, p. 11). Models of anger include both angry emotions and the
manner in which they are controlled and expressed (e.g., Novaco,
1994; Siegel, 1986). For example, Spielberger’s (1999) measure is
designed to assess feeling angry, feeling like expressing anger
verbally, and feeling like expressing anger physically.
This construal of anger closely approximates the dynamic risk
factor for violence identified in the present study for high-risk
patients. The BSI Hostility subscale assesses feelings of annoyance
and irritation, a tendency to argue, urges to destroy property or hurt
others, and uncontrollable temper outbursts. Because this subscale
addresses emotional reactivity more than an attitudinal disposition,
Jarvis and Novaco (2006) concluded that it assessed anger rather
than hostility. This conclusion enjoys some empirical support.
Although there is no evidence that the BSI Hostility subscale
relates strongly to other measures of hostility, there is evidence
that this subscale correlates moderately strongly (r ? .40) with
well-validated anger scales, particularly those that assess trait
anger and expressions of anger (Conger, Conger, Edmondson,
Tescher, & Smolin, 2003; see also Suris et al., 2004). The BSI
Hostility subscale also shares its violence-predictive variance with
anger scales (Vanni et al., 2004). Although future research should
examine the relations among the BSI Hostility subscale, traditional
anger scales, and violence in this high-risk group, extant theory and
research suggest that anger is the construct tapped in this study.
This study indicates that high-risk patients who experience an
increase in angry reactivity this week are at greater risk for
violence next week. Given the relatively tight conceptual connec-
tions between anger and aggression (see Novaco, 1994), this
finding makes sense. Nevertheless, this study is the first to suggest
that anger functions as both a status and a state in elevating
violence risk. Although anger is an emotional state, people who
experience anger frequently and intensely are high in status, or
“trait,” anger (Spielberger, 1999). As a group, the high-risk pa-
tients studied here are characterized by high levels of ongoing
anger. Indeed, the actuarial formula used to screen participants into
the study included elevated BSI hostility scores and violence
(Gardner et al., 1996; Skeem et al., 2002). These high-risk patients
often experience angry arousal and are prone to expressing it in
maladaptive ways (see Blackburn, 1993). Past risk status research
has indicated that anger (Gardner et al., 1996; McNiel, Eisner, &
Binder, 2003; Monahan, Steadman, et al., 2001; Novaco, 1994)
and antagonistic traits involving proneness to anger and temper
outbursts (see Skeem, Miller, Mulvey, Tiemann, & Monahan,
2005) are strong risk factors for violence, even among general
psychiatric patients. One retrospective study even suggested that
increases in anger predict sexual aggression among offenders
(Hanson & Harris, 2000).
The present findings are the first to point to a more dynamic
linkage in which both trait and state levels of anger contribute to
violence potential among high-risk patients. Anger is strongly
related to itself over time, suggesting that it is trait-like. Within a
given week, however, anger is also related to violence, reflecting
the moderate overlap between anger and aggression. Nevertheless,
even taking these considerable trait effects into account in a
high-risk and “highly angry” group, a patient’s anger this week
predicts violence next week.
General Psychological Distress and Specific Symptom
Constellations as Risk Markers
In contrast to the results for anger, general psychological dis-
tress and specific symptoms (depression, anxiety, and TCO be-
liefs) did not predict proximate violence. Prior research has pro-
duced mixed results on whether and how psychiatric symptoms
elevate risk of violence (see Douglas & Skeem, 2005). Several
well-designed risk status studies have suggested that mental dis-
order alone is not a particularly strong risk factor for violence
(Lidz et al., 1993), although mental disorder combined with sub-
stance abuse moderately increases risk (Monahan, Steadman, et
al., 2001; F. Swanson et al., 1990). Substance abuse was not
examined in the current analyses, but the present results are in
keeping with this past research.
The present results suggest that anger is more important in
predicting proximate violence than are general psychological dis-
tress and such specific symptom constellations as anxiety and
depression. These results are consistent with past indications from
risk status studies that symptoms that are more intuitively related
to violence (e.g., hostile suspiciousness; excitement) are, in fact,
more strongly related to violence than those that are not (e.g.,
depression, anxiety; see Link & Stueve, 1994; Lowenstein et al.,
1990; McNeil & Binder, 1994; McNiel et al., 2003). Becoming
more generally distressed, depressed, or anxious may indirectly
place one at greater risk for violence by leading to interpersonal
problems with family and friends, reducing one’s coping capacity
for stressors, and the like (Douglas & Skeem, 2005). In fact, in
contrast to the results of this intensive longitudinal study, early
results of a multinational study suggested that increases in anxiety
and depressive symptoms measured at one time point are strongly
predictive of later community violence (Freese, Miotto, & Reback,
2002). However, becoming more angry and irritable may directly
lead to violence given that aggression can be an expression of
anger. It seems that the more directly related a symptom is to
aggression conceptually, the stronger its empirical connection.
Despite the lack of a time-ordered relationship between symp-
toms other than anger and violence, both general psychological
distress and all specific symptom constellations assessed in this
study were likely to co-occur with violence during a given week.
SKEEM ET AL.
This suggests that in a group of high-risk patients, these symptoms
other than anger are risk markers for violence during short time
periods, even if they are not risk factors for proximate violence
(see Kraemer et al., 1997). During tumultuous periods of patients’
lives, various forms of psychiatric symptoms and violence are
more likely to occur. This co-occurrence is not particularly infor-
mative, given that an increase in symptoms may precede, follow,
or have nothing to do with the occurrence of violence during that
week. Other factors, including relationship problems and sub-
stance abuse or the cumulative effect of multiple factors, may be
responsible for both the increase in symptoms and risk of violence.
The TCO Debate
In this study, on the basis of past research, we focused on
psychiatric symptoms that were most likely related to violence.
Some past research has indicated that negative affect (i.e., anger,
anxiety, depression) and particular features of psychosis (i.e., TCO
beliefs) hold the most promise as predictors of proximate violence
(for a review, see Douglas & Skeem, 2005). As noted earlier, this
may be particularly true of active symptoms.
Recall that the patients screened into the present study as at
high risk for violence by our actuarial tool were not actively
thought-disordered at the time of recruitment. Patients with
active delusions or a current diagnosis of schizophrenia (but not
active hallucinations) were excluded. This raises a concern
about whether the sample manifested significant variability in
psychosis over time to detect a relationship between TCO
delusions and violence. Several points help ameliorate this
concern. First, psychotic symptoms (including TCO) are not
limited to patients with schizophrenia and could have occurred
with sufficient regularity in this sample because patients in the
present sample suffered from disorders that include features of
psychosis. One patient in 10 (11%) had an emergency room
diagnosis of a psychotic disorder (e.g., schizoaffective disorder;
psychotic disorder NOS), and 3 in 4 (76%) were diagnosed with
affective disorders that can include psychotic symptoms. Re-
cent studies have indicated that one in five major depressive
episodes includes psychotic features (Ohayon & Schatzberg,
2002), and 2 of 3 patients with bipolar disorder experience
psychotic symptoms (including Schneiderian first-rank symp-
toms) during affective episodes (Keck et al., 2003). Second,
symptoms of psychosis ebb and flow across episodes of a
patient’s illness. Although patients in the present sample did not
manifest active delusions at the time of recruitment, there is
evidence that they were often acutely ill during the 6 months
that they were followed in the community. Nearly half (46%)
became so ill that they were hospitalized during this period. As
a group, they averaged 7 inpatient days (Lidz et al., 2006).
Third, our TCO data suggest that there was variability in the
scale both across individuals (see Table 1) and within individ-
uals over time (M ISD ? 1.5, SD ? 1.2).4Although our findings
may not generalize to “more psychotic” patient groups, our
sample is still capable of representing the relation between TCO
symptoms and violence in this group of repeatedly violent
The primary measure of psychosis included in the present study
was the TCO scale. Although the extent to which this scale
measures the delusional quality of an individual’s perceptions per
se is questionable (Appelbaum et al., 2000), we did not include a
clinical assessment of whether TCO beliefs were actually psy-
chotic. On the basis of past research (Appelbaum et al., 2000),
such an addition would weaken any observed relationship between
apparent TCO delusions and violence. In the present study, TCO
symptoms did not predict proximate violence, suggesting that there
was little need for such an adjustment. These results are consistent
with past major studies indicating that delusions and hallucinations
are not predictive of violence (Appelbaum et al., 2000; Lidz et al.,
1993; Monahan, Steadman, et al., 2001; cf. J. W. Swanson, Borum,
Swartz, & Monahan, 1996), with the possible exception of hallu-
cinations that specifically command violence (Monahan, Stead-
man, et al., 2001).
It is somewhat surprising that the generally hostile, suspicious
cognitive style that the TCO questions may tap (Appelbaum et al.,
2000) was not predictive of violence in the present study. Theo-
retically, this cognitive style would overlap with the anger tapped
by the BSI Hostility subscale, which was predictive of proximate
violence. Although patients’ average scores on the two scales were
strongly correlated (r ? .74), only anger (BSI Hostility) signifi-
cantly increased the risk of proximate violence in this study.
Feeling angry and feeling like smashing things or injuring others
are more relevant to proximate violence than feeling threatened or
controlled by other people or forces.
Limitations and Future Directions
Because we deliberately focused on a small subgroup of partic-
ipants at high risk for repeated violence in this study, we cannot
assume that the findings apply to all psychiatric patients. Changes
in symptoms (other than anger) may be relevant to violence risk
for the general psychiatric population. We cannot underscore this
point heavily enough.
At the same time, we believe that focusing on the high-risk
subgroup studied here has the greatest potential to inform system-
atic violence risk management efforts. Given that the likelihood of
violence in the general psychiatric population is relatively low, it
makes sense from a policy perspective to focus on the small group
of patients who account for the majority of violent incidents. This
raises two questions.
First, are psychiatric patients at high risk for violence a homo-
geneous group that is well represented in the present sample?
Preliminary evidence suggests that high-risk patients are not a
homogeneous group, but the bulk of these patients share charac-
teristics with those of the present sample. In the MacArthur Vio-
lence Risk Assessment Study, investigators applied the actuarial
tool described earlier (Monahan, Steadman, Appelbaum, et al.,
4An ISD, or individual standard deviation, captures intraindividual
variability in TCO scores. Each individual has a mean TCO score over the
observation period (M) and value that characterizes how much they varied
around their mean (ISD). The ISD values demonstrate that people did in
fact vary in their TCO scores over the observation period. Only 11 people
had an ISD score of 0, indicating that they did not deviate from their
average TCO score. Notably, the ISD values for TCO scores were twice
those of hostility scores (M ISD ? 0.7, SD ? 0.3), which manifested a
significant relation to violence over time.
PSYCHIATRIC SYMPTOMS AND COMMUNITY VIOLENCE
2005) to data on more than 1,000 patients to identify 165 patients
as at high risk for violence (Monahan, Steadman, et al., 2001).
These high-risk patients were found to form three theoretically
coherent subgroups (Skeem et al., 2005). Although two of the
subgroups share significant features with those of the present
sample (e.g., negative affect, anger, and substance abuse prob-
lems), the third subgroup was characterized by thought disorder,
including active delusions and command hallucinations. This
thought-disordered subgroup accounted for only 14% of high-risk
patients and 2% of the full patient sample. Despite the small size
of this subgroup, there may be a relationship between such symp-
toms as TCO and violence in this subgroup. This question is open
for future research.
Second, will the present results replicate with patients identified
as high risk using tools other than the one applied here? This too
is a question open for future research. Given the overlap among
brief violence risk assessment tools in classifying individuals as
high risk (Skeem, 2005), however, the sample studied here should
represent a significant proportion of high-risk patients identified
with other tools. The sample may also better represent patients at
risk for repeated involvement in violence given that tools other
than the one used in this study were not designed to predict this
Assessing Symptoms Differently
In this study of what might be the “most common” type of
high-risk psychiatric patient, we observed no relation between
general distress and most symptoms on the one hand and violence
on the other. These findings do not mean that “symptoms fail to
predict proximate violence” for the typical high-risk patient for
two reasons. First, research cannot prove the null hypothesis.
Second, in this study, we focused on general psychological distress
and a limited number of specific symptom constellations (e.g.,
negative affect, TCO beliefs, and anger) that held promise in
predicting proximate violence. We did so using a measure that
would be feasible for use on a weekly basis. It is possible that
symptom constellations other than those measured here (e.g., ma-
nia, agitation) and measured by means other than those applied
here (e.g., clinician ratings) would produce different results. Be-
cause this is the first study of its kind, it is important to replicate
the results with other measures.
On a related note, it is possible that this study’s weekly mea-
surement was too molar to capture a time-ordered relationship
between changes in symptoms and changes in violence. Recall that
there was a concurrent (within-week) relationship between all
forms of symptoms and violence, whereas only hostility predicted
violence in lagged (across-weeks) analyses. Although this is a
possibility to explore in future research with even more frequent
measurement of symptoms, this explanation seems unlikely. Most
psychiatric symptoms (e.g., a depressive episode) do not “come
and go” on a daily basis. For example, negative affect (Harmon-
Jones, 2000) and mood (Benedict, Dobraski, & Goldstein, 1999)
change over periods of weeks. Moreover, for practical purposes,
this unit of measurement is perhaps the most meaningful as it
mirrors the observation period that a clinical relationship affords.
It is unlikely that clinicians and patients would regularly meet at
A second potential explanation for the observation of a concur-
rent, but not lagged, relationship between most symptoms and
violence is reporting bias. Each week, patients reported how both-
ered they had been by a variety of symptoms during the previous
week. Patients and collateral informants also reported whether the
patient had been violent during the previous week. It is possible
that some patients endorsed having experienced more symptoms
during the previous week because they had been violent the
previous week. We were unable to test this explanation, given
the present data. Nevertheless, if this explanation were accu-
rate, it would suggest only that concurrent relationships that we
identified between symptoms and violence within a week were
artifactual. This would not affect the study’s principal finding
that only increases in anger this week predict violence the
Testing Interaction Effects
In this study, we focused intensively on psychiatric symptoms.
It is possible that these symptoms interact with environmental
(Silver, Mulvey, & Monahan, 1999) and individual (Monahan &
Steadman, 1994) risk factors in more complex patterns that yield
violence. This is particularly important when it comes to substance
abuse. Past risk status studies have indicated that mental disorder
affects risk chiefly when it is combined with substance abuse
disorders (Steadman et al., 1998; F. Swanson et al., 1990). Al-
though the present analyses provide no evidence that psychiatric
symptoms other than anger alone increase risk of proximate vio-
lence, future work should determine whether this finding holds
when symptoms, substance abuse, and other factors are examined
Implications for Research and Practice
These results have important implications for research and prac-
tice. For both professional domains, important lessons have been
learned here about attending to and adequately assessing the time-
ordered relationship between symptoms and violence. When ex-
amined concurrently (during the same week), all forms of symp-
toms are likely to co-occur with violence. This is infinitely less
useful than knowing that a symptom precedes and increases the
risk of proximate violence (Kraemer et al., 1997). When examined
over time (from 1 week to the next), only anger predicted proxi-
mate violence. Thus, with the important exception of anger, this
study suggests that in a sample of high-risk patients, knowing
about increased symptoms studied here during 1 week does not
provide a solid foundation for expecting violence the following
Speaking in practical terms, a clinician who is monitoring an
identified high-risk patient in the community probably would be
incorrect if she assumed that her patient’s increasing general
distress (or depression, or anxiety, or TCO symptoms) this week
signified increased risk of imminent violence. For high-risk pa-
tients (not necessarily actively psychotic patients), the results of
this study challenge the prevailing assumption that deterioration in
psychiatric symptoms increases patients’ risk of imminent vio-
lence. This assumption dominates legal standards for involuntary
inpatient and outpatient treatment as much as it dominates inter-
SKEEM ET AL.
vention strategies for patients viewed as high risk (Mulvey & Lidz,
The present results suggest that merely providing high-risk
patients with “more” or “different” medication or therapy focused
on general psychiatric symptoms is unlikely to reduce violence
risk. Interventions for high-risk patients would do better to focus
on anger reduction and management. Anger, unlike other symp-
toms, is a changeable risk factor for violence for these patients.
Meta-analytic (Beck & Fernandez, 1998; Edmondson & Conger,
1996) and other reviews (Deffenbacher, Oetting, & DiGiuseppe,
2002) have suggested that anger declines as a result of such anger
management and related treatment. The next step is to determine
the extent to which reducing high-risk patients’ anger or changing
its typical mode of expression reduces their involvement in
These results also inform dynamic risk assessment efforts by
indicating what symptoms to monitor over time after patients have
been identified as high risk. When clinicians are treating a high-
risk patient in the community, they should monitor changes in his
or her degree of anger. Training programs for violence risk assess-
ment and management may need to shift emphasis from general
psychiatric symptoms to anger, violence, and other constructs that
relate more directly to risk of proximate violence.
More generally, the results of this study challenge researchers
and practitioners to examine closely the assumption that psychi-
atric deterioration leads directly to patient violence. We are un-
aware of compelling evidence that this is the case, and the findings
presented here cast doubt on the assumption of a strong link. Given
the amount of mental health policy and practice that rests on this
assumption, further thought and rigorous research are needed to
resolve this issue.
5Although one might argue that our sample was not prototypic of those
eligible for civil commitment, civil committees are a heterogeneous group.
Our sample suffered from serious mental disorder and was at high risk for
violence, which goes to the heart of the “dangerousness” language in most
civil commitment statutes. At the same time, our sample excluded patients
with active psychosis at the time of their emergency room visit, and these
patients also were eligible for civil commitment. Comparing our sample
(17% committed) with the larger sample of general psychiatric patients
from which they were drawn (18% committed), we found no significant
difference in the likelihood of involuntary commitment at the time of the
index visit to the emergency room, ?2(1, N ? 3,297) ? 0.7, ns. Our sample
was as eligible for civil commitment as the larger group from which it was
Allison, P. (2005). Fixed effects regression methods for longitudinal data
using SAS. Cary, NC: SAS Press.
Appelbaum, P., Robbins, P., & Monahan, J. (2000). Violence and delu-
sions: Data from the MacArthur Risk Assessment Study. American
Journal of Psychiatry, 157, 566–572.
Appelbaum, P. S. (2001). Thinking carefully about outpatient commitment.
Psychiatric Services, 52, 347–350.
Banks, S., Robbins, P. C., Silver, E., Vesselinov, R., Steadman, H. J.,
Monahan, J., et al. (2004). A multiple-models approach to violence risk
assessment among people with mental disorder. Criminal Justice and
Behavior, 31, 324–340.
Beck, R., & Fernandez, E. (1998). Cognitive-behavioral self-regulation of
the frequency, duration, and intensity of anger. Journal of Psychopa-
thology and Behavioral Assessment, 20, 217–229.
Benedict, R. H., Dobraski, M., & Goldstein, M. Z. (1999). A preliminary
study of the association between changes in mood and cognition in a
mixed geriatric psychiatry sample. Journals of Gerontology Series B:
Psychological Sciences & Social Sciences, 54B, P94–P99.
Benishek, L., Hayes, C., Bieschke, K., & Stoffelmayr, B. (1998). Explor-
atory and confirmatory factor analyses of the Brief Symptom Inventory
among substance abusers. Journal of Substance Abuse, 10, 103–114.
Blackburn, R. (1993). The psychology of criminal conduct: Theory, re-
search and practice. Oxford, England: Wiley.
Bonta, J., Law, M., & Hanson, K. (1998). The prediction of criminal and
violent recidivism among mentally disordered offenders: A meta-
analysis. Psychological Bulletin, 123, 123–142.
Bonynge, E. (1993). Unidimensionality of SCL-90R scales in adult and
adolescent crisis samples. Journal of Clinical Psychology, 49, 212–215.
Boulet, J., & Boss, M. (1991). Reliability and validity of the Brief Symp-
tom Inventory. Psychological Assessment, 3, 433–437.
Brophy, C., Norvell, N., & Kiluk, D. (1988). An examination of the factor
structure and convergent and discriminant validity of the SCL-90R in an
outpatient clinic population. Journal of Personality Assessment, 52,
Carscaddon, D. M., George, M., & Wells, G. (1990). Rural community
mental health consumer satisfaction and psychiatric symptoms. Commu-
nity Mental Health Journal, 26, 309–318.
Conger, J. C., Conger, A. J., Edmondson, C., Tescher, B., & Smolin, J.
(2003). The relationship of anger and social skills to psychological
symptoms. Assessment, 10, 248–258.
Deffenbacher, J. L., Oetting, E. R., & DiGiuseppe, R. A. (2002). Principles
of empirically supported interventions applied to anger management.
Counseling Psychologist, 30, 262–280.
Derogatis, L., & Melisaratos, N. (1983). The brief symptom inventory: An
introductory report. Psychological Medicine, 13, 595–605.
Douglas, K., & Skeem, J. (2005). Violence risk assessment: Getting
specific about being dynamic. Psychology, Public Policy, and Law, 11,
Eckhardt, C., Norlander, B., & Deffenbacher, J. (2004). The assessment of
anger and hostility: A critical review. Aggression and Violent Behavior,
Edmondson, C. B., & Conger, J. C. (1996). A review of treatment efficacy
for individuals with anger problems: Conceptual, assessment, and meth-
odological issues. Clinical Psychology Review, 16, 251–275.
Ferrer, E., & McArdle, J. J. (2003). Alternative structural models for
multivariate longitudinal data analysis. Structural Equation Modeling,
Freese, T. E., Miotto, K., & Reback, C. J. (2002). The effects and conse-
quences of selected club drugs. Journal of Substance Abuse Treatment,
Gardner, W., Lidz, C., Mulvey, E., & Shaw, E. (1996). A comparison of
actuarial methods for identifying repetitively violent patients with men-
tal illness. Law and Human Behavior, 20, 35–48.
Gardner, W., Mulvey, E. P., & Shaw, E. C. (1995). Regression analyses of
counts and rates: Poisson, overdispersed Poisson, and negative binomial
models. Psychological Bulletin, 118, 392–404.
Gavazzi, S., Julian, T., & McKenry, P. (1996). Utilization of the Brief
Symptom Inventory to discriminate between violent and nonviolent
male relationship partners. Psychological Reports, 79, 1047–1056.
Hafkenscheid, A. (1993). Psychometric evaluation of the Symptom Check-
list (SCL-90) in psychiatric inpatients. Personality and Individual Dif-
ferences, 14, 751–756.
Hanson, R. K., & Harris, A. J. R. (2000). Where should we intervene?
Dynamic predictors of sexual assault recidivism. Criminal Justice and
Behavior, 27, 6–35.
Harmon-Jones, E. (2000). Cognitive dissonance and experienced negative
PSYCHIATRIC SYMPTOMS AND COMMUNITY VIOLENCE
affect: Evidence that dissonance increases experienced negative affect
even in the absence of aversive consequences. Personality and Social
Psychology Bulletin, 26, 1490–1501.
Harris, G. T., Rice, M. E., & Quinsey, V. L. (1993). Violent recidivism of
mentally disordered offenders: The development of a statistical predic-
tion instrument. Criminal Justice and Behavior, 20, 315–335.
Hayes, J. (1997). What does the Brief Symptom Inventory measure in
college and university counseling center clients? Journal of Counseling
Psychology, 44, 360–367.
Heinrich, R., & Tate, D. (1996). Latent variable structure of the Brief
Symptom Inventory in a sample of persons with spinal cord injuries.
Rehabilitation Psychology, 41, 131–147.
Hodgins, S. (1992). Mental disorder, intellectual deficiency, and crime:
Evidence from a birth cohort. Archives of General Psychiatry, 49,
Holden, R., Starzyk, K., McLeod, L., & Edwards, M. (2000). Comparisons
among the Holden Psychological Screening Inventory (HPSI), the Brief
Symptom Inventory (BSI), and the Balanced Inventory of Desirable
Responding (BIDR). Assessment, 7, 163–175.
Jarvis, K. L., & Novaco, R. W. (2006). Post-shelter adjustment of children
from violent families. Journal of Interpersonal Violence, 21, 1046–
Joreskog, K. G. (1970). Estimation and testing of simplex models. British
Journal of Mathematical and Statistical Psychology, 23, 121–145.
Joreskog, K. G., & Sorbom, D. (1979). Advances in factor analysis and
structural equation models. Cambridge, MA: Abt Books.
Kassinove, H., & Sukhodolsky, D. G. (1995). Anger disorders: Basic
science and practice issues. In H. Kassinove (Ed.), Anger disorders:
Definitions, diagnosis, and treatment (pp. 2–27). Washington, DC: Tay-
lor & Francis.
Keck, P. E., Jr., McElroy, S. L., Havens, J. R., Altshuler, L. L., Nolen,
W. A., Frye, M. A., et al. (2003). Psychosis in bipolar disorder: Phe-
nomenology and impact on morbidity and course of illness. Compre-
hensive Psychiatry, 44, 263–269.
Kraemer, H. C., Kazdin, A. E., Offord, D. R., & Kessler, R. C. (1997).
Coming to terms with the terms of risk. Archives of General Psychiatry,
Lidz, C., Mulvey, E., & Gardner, W. (1993). The accuracy of predictions
of violence to others. Journal of the American Medical Association, 269,
Lidz, C., Mulvey, E., Schubert, C., Banks, S., Gardner, W., Skeem, J., et
al. (2006). The effectiveness of civil commitment in a new clinical
environment. Unpublished manuscript.
Link, B., Andrews, H., & Cullen, F. (1992). The violence and illegal
behavior of mental patients reconsidered. American Sociological Re-
view, 57, 275–292.
Link, B., & Stueve, A. (1994). Psychotic symptoms and the violent/illegal
behavior of mental patients compared to community controls. In J.
Monahan & H. Steadman (Eds.), Violence and mental disorder: Devel-
opments in risk assessment (pp. 137–159). Chicago: University of Chi-
Link, B., Stueve, A., & Phelan, J. (1998). Psychotic symptoms and violent
behaviors: Probing the components of “threat/control override” symp-
toms. Social Psychiatry and Psychiatric Epidemiology, 33, S55–S60.
Lowenstein, M., Binder, R., & McNiel, D. (1990). The relationship be-
tween admission symptoms and hospital assaults. Hospital and Commu-
nity Psychiatry, 41, 311–313.
McArdle, J. J., & Bell, R. Q. (2000). An introduction to latent growth
models for developmental data analysis. In T. D. Little & K. U. Schnabel
(Eds.), Modeling longitudinal and multilevel data: Practical issues,
applied approaches, and specific examples (pp. 69–107, 269–281).
Mahwah, NJ: Erlbaum.
McNiel, D. E., & Binder, R. L. (1994). The relationship between acute
psychiatric symptoms, diagnosis, and short-term risk of violence. Hos-
pital and Community Psychiatry, 45, 133–137.
McNiel, D. E., Eisner, J. P., & Binder, R. L. (2003). The relationship
between aggressive attributional style and violence by psychiatric pa-
tients. Journal of Consulting and Clinical Psychology, 71, 399–403.
Monahan, J. (1996). Violence prediction: The past twenty and the next
twenty years. Criminal Justice and Behavior, 23, 107–120.
Monahan, J., Bonnie, R. J., Appelbaum, P. S., Hyde, P. S., Steadman, H. J.,
& Swartz, M. S. (2001). Mandated community treatment: Beyond out-
patient commitment. Psychiatric Services, 52, 1198–1205.
Monahan, J., Steadman, H., Silver, E., Appelbaum, P., Robbins, P., Mul-
vey, E., et al. (2001). Rethinking violence risk assessment: The
MacArthur Study of Mental Disorder and Violence. New York: Oxford
Monahan, J., & Steadman, H. J. (1994). Violence and mental disorder:
Developments in risk assessment. Chicago: University of Chicago Press.
Monahan, J., Steadman, H. J., Appelbaum, P. S., Grisso, T., Mulvey, E. P.,
Roth, L. H., et al. (2005). Classification of violence risk: Professional
manual. Lutz, FL: Psychological Assessment Resources.
Monahan, J., Steadman, H. J., Robbins, P. C., Appelbaum, P. S., Banks, S.,
Grisso, T., et al. (2005). An actuarial model of violence risk assessment
for persons with mental disorders. Psychiatric Services, 56, 810–815.
Mulvey, E. P., & Lidz, C. W. (1998). Clinical prediction of violence as a
conditional judgment. Social Psychiatry and Psychiatric Epidemiology,
Muthe ´n, B., & Muthe ´n, L. K. (2003). Mplus (Version 3.1) [Computer
software]. Los Angeles: Author.
Norko, M., & Baranoski, M. V. (2005). The state of contemporary risk
assessment research. Canadian Journal of Psychiatry, 50, 18–26.
Novaco, R. W. (1994). Anger as a risk factor for violence among the
mentally disordered. In J. Monahan & H. J. Steadman (Eds.), Violence
and mental disorder: Developments in risk assessment (pp. 21–59).
Chicago: University of Chicago Press.
Ohayon, M. M., & Schatzberg, A. F. (2002). Prevalence of depressive
episodes with psychotic features in the general population. American
Journal of Psychiatry, 159, 1855–1861.
Pekarik, G. (1983). Follow-up adjustment of outpatient dropouts. American
Journal of Orthopsychiatry, 53, 501–511.
Piersma, H., Boes, J., & Reaume, W. (1994). Unidimensionality of the
Brief Symptom Inventory (BSI) in adult and adolescent inpatients.
Journal of Personality Assessment, 63, 338–344.
Piersma, H., Reaume, W. M., & Boes, J. L. (1994). The Brief Symptom
Inventory (BSI) as an outcome measure for adult psychiatric inpatients.
Journal of Clinical Psychology, 50, 555–563.
Ruiperez, M., Ibanez, M., Lorente, E., Moro, M., & Ortet, G. (2001).
Psychometric properties of the Spanish version of the BSI. European
Journal of Psychological Assessment, 17, 241–250.
Schubert, C. A., Mulvey, E. P., Lidz, C. W., Gardner, W. P., & Skeem, J. L.
(2005). Weekly interviews with high-risk participants. Journal of Inter-
personal Violence, 20, 632–646.
Siegel, J. M. (1986). The Multidimensional Anger Inventory. Journal of
Personality and Social Psychology, 51, 191–200.
Silver, E., Mulvey, E. P., & Monahan, J. (1999). Assessing violence risk
among discharged psychiatric patients: Toward an ecological approach.
Law and Human Behavior, 23, 237–255.
Skeem, J. (2003, April). Dually diagnosed individuals at high risk for
violence: Treatment needs, preferences, and practices. Paper presented
at the 3rd annual conference of the International Association of Forensic
Mental Health Services, Miami, FL.
Skeem, J. (2005, March). Identifying patients at risk for repeated involve-
ment in violence: Self-perceptions perform better than systematic scales.
In J. Skeem (Chair), Identifying high-risk patients and isolating their
dynamic risk factors for violence. Symposium conducted at the Ameri-
can Psychology Law Society conference, San Diego, CA.
SKEEM ET AL.
Skeem, J., & Mulvey, E. (2002). Monitoring the violence potential of Download full-text
mentally disordered offenders being treated in the community. In A.
Buchanan (Ed.), Care of the mentally disordered offender in the com-
munity (pp. 111–142). New York: Oxford University Press.
Skeem, J., Mulvey, E., Appelbaum, A., Banks, S., Grisso, T., Silver, E., et
al. (2004). Identifying subtypes of civil psychiatric patients at high risk
for violence. Criminal Justice and Behavior, 31, 392–437.
Skeem, J. L., Miller, J. D., Mulvey, E., Tiemann, J., & Monahan, J. (2005).
Using a five-factor lens to explore the relation between personality traits
and violence in psychiatric patients. Journal of Consulting and Clinical
Psychology, 73, 454–465.
Skeem, J. L., Mulvey, E. P., Lidz, C., Gardner, W., & Schubert, C. (2002).
Identifying psychiatric patients at risk for repeated involvement in
violence: The next step toward intensive community treatment pro-
grams. International Journal of Forensic Mental Health, 1, 155–170.
Smith, T. W. (1994). Concepts and methods in the study of anger, hostility,
and health. In A. W. Siegman & T. W. Smith (Eds.), Anger, hostility,
and the heart (pp. 23–42). Hillsdale, NJ: Erlbaum.
Spielberger, C. D. (1999). STAXI-2: State–Trait Anger Expression
Inventory-2 professional manual. Odessa, FL: Psychological Assess-
Steadman, H. J., Mulvey, E. P., Monahan, J., Robbins, P. C., Appelbaum,
P. S., Grisso, T., et al. (1998). Violence by people discharged from acute
psychiatric inpatient facilities and by others in the same neighborhoods.
Archives of General Psychiatry, 55, 393–401.
Straus, M., & Gelles, R. (1990). Physical violence in American families.
Brunswick, NJ: Transaction.
Suris, A., Lind, L., Emmett, G., Borman, P. D., Kashner, M., & Barratt,
E. S. (2004). Measures of aggressive behavior: Overview of clinical and
research instruments. Aggression and Violent Behavior, 9, 165–227.
Swanson, J. W., Holzer, C., Ganju, V., & Jono, R. (1990). Violence and
psychiatric disorder in the community: Evidence from the Epidemio-
logic Catchment Area Surveys. Hospital and Community Psychiatry, 41,
Swanson, J. W., Borum, R., Swartz, M. S., & Hiday, V. (1999). Violent
behavior preceding hospitalization among persons with severe mental
illness. Law and Human Behavior, 23, 185–204.
Swanson, J. W., Borum, R., Swartz, M. S., & Monahan, J. (1996). Psy-
chotic symptoms and disorders and the risk of violent behaviour in the
community. Criminal Behaviour and Mental Health, 6, 309–329.
Swanson, J. W., Estroff, S., Swartz, M., Borum, R., Lachicotte, W.,
Zimmer, C., et al. (1997). Violence and severe mental disorder in clinical
and community populations: The effect of psychotic symptoms, comor-
bidity and lack of treatment. Psychiatry, 60, 1–22.
Taylor, P. (1998). When symptoms of psychosis drive serious violence.
Social Psychiatry and Psychiatric Epidemiology, 33, S47–S54.
Vanni, A., Biancosino, B., Marmai, L., Zucconi, M., Peron, L., Cavrini, L.,
et al. (2004). Psychobiological concomitants of aggressive behaviour
among psychiatric inpatients: A control study. Rivista Di Psichiatria, 39,
Webster, C. D., Douglas, K. S., Eaves, D., & Hart, S. D. (1997). Assessing
risk of violence to others. In C. D. Webster & M. A. Jackson (Eds.),
Impulsivity: Theory, assessment, and treatment (pp. 251–277). New
York: Guilford Press.
Wessely, S., Castle, D., Douglas, A. J., & Taylor, P.J. (1994). The criminal
careers of incident cases of schizophrenia. Psychological Medicine, 24,
Yu, C. Y., & Muthe ´n, B. (2002). Evaluation of model fit indices for latent
variable models with categorical and continuous outcomes (Tech. Rep.).
Los Angeles: University of California at Los Angeles, Graduate School
of Education and Information Studies.
Received October 22, 2005
Revision received May 22, 2006
Accepted May 31, 2006 ?
PSYCHIATRIC SYMPTOMS AND COMMUNITY VIOLENCE