Developing a Screening Instrument and At-Risk Profile for Nonsuicidal

Self-Injurious Behavior in College Women and Men

Hsiu-Lan Cheng

University of Michigan

Brent Mallinckrodt

University of Tennessee

Johanna Soet and Todd Sevig

University of Michigan

Archival data (N

⫽ 1,048 women, 1,136 men) from a mental health survey of college students were used

to investigate incidence of nonsuicidal self-injury (NSSI), including cutting. Significant levels (defined
as 4 –5 lifetime incidents) were found in 9.3% of women and 5.3% of men. The Counseling Center
Assessment for Psychological Symptoms (a global symptom inventory) and an assessment of trauma had
been field tested with this sample. We randomly partitioned half of these data into a holdout sample and
used the remainder to develop an NSSI screening inventory that included (a) 5 women’s screening items,
including 1 item to assess trauma experienced; (b) 11 men’s screening items; and (c) 12 items common
to men and women, including depression, dissociation, anger, unwanted thoughts, nightmares or
flashbacks, and having witnessed trauma. Logistic regression and receiver-operating characteristic curve
analysis suggested the inventory significantly discriminated NSSI men and women in the holdout sample,
p

⬍ .001. Cutoff scores were identified to correctly classify about 48% of the true positive male and

female NSSI cases, with false positive rates of 13.2% and 8.4% for women and men, respectively.

Keywords: self-injury incidence, screening instrument, college students, cutting, affect regulation

Supplemental materials: http://dx.doi.org/10.1037/a0018206.supp

Nonsuicidal self-injury (NSSI), sometimes referred to as cut-

ting, involves deliberate destruction or alteration of body tissue to
inflict self-harm and pain but without conscious intention to com-
mit suicide (Favazza, 1998; Nock & Prinstein, 2004). Although
some research suggests that people with a history of NSSI are at
greater risk for suicide (Hawton, Zahl, & Weatherall, 2003), many
individuals with NSSI deny having suicidal thoughts or making
suicide attempts. Also, considerable research suggests that when
persons with a history of self-injury do attempt suicide, they often
use methods other than those typically seen with NSSI (e.g., drug
overdose), and the attempt usually occurs during periods in which
they are not actively engaging in NSSI (Stanley, Gameroff,
Michalsen, & Mann, 2001). However, unintentional deaths by
NSSI do occur (Kehrberg, 1997). Apart from the life-threatening

risk, the negative physical, emotional, and interpersonal conse-
quences of NSSI are also of grave concern.

The occurrence of NSSI in the general U.S. adult population is

estimated to be about 4%, whereas in psychiatric adult inpatients,
a rate of approximately 21% has been estimated (Klonsky, Olt-
manns, & Turkheimer, 2003). Because NSSI typically begins in
adolescence and lasts for 10 –15 years (Favazza, 1999), the prev-
alence rate may be even higher for adolescents and young adults
than in general samples of adults. Studies show that approximately
15% of high school students report engaging in NSSI (Laye-
Gindhu & Schonert-Reichl, 2005), but other research reported
NSSI rates of 37% and 28% in two samples of privileged adoles-
cents (Yates, Tracy, & Luthar, 2008). The prevalence of NSSI on
college and university campuses may be part of a general trend
toward higher rates of serious mental health issues with each new
student cohort (Gallagher, Zhang, & Taylor, 2003).

However, a search of the literature located only nine empirical

studies (based on only seven samples) that examined the occur-
rence rates of NSSI among college students (Brown, Williams, &
Collins, 2007; Croyle & Waltz, 2007; Favazza, DeRosear, &
Conterio, 1989; Gratz, 2001, 2006; Gratz & Chapman, 2007;
Gratz, Conrad, & Roemer, 2002; Klonsky & Olino, 2008; Whit-
lock, Eckenrode, & Silverman, 2006). In these seven samples,
NSSI incidence rates ranged from 14% (Favazza et al., 1989) to
44% (Gratz & Chapman, 2007). When considering this research, it
is important to distinguish between incidence and prevalence.
Prevalence refers to the proportion of a sample that currently
manifests the symptom or disease. Incidence refers to the propor-

Editor’s Note.

Karen O’Brien selected Helen Neville as the action editor

for this article.

Hsiu-Lan Cheng, Johanna Soet, and Todd Sevig, Counseling and Psy-

chological Services, University of Michigan; Brent Mallinckrodt, Depart-
ment of Psychology, University of Tennessee.

Johanna Soet is now at the Michigan Domestic Violence Prevention and

Treatment Board, Lansing, Michigan.

Correspondence concerning this article should be addressed to Hsiu-Lan

Cheng, 530 South State Street, 3100 Michigan Union, Counseling and
Psychological Services, University of Michigan, Ann Arbor, MI 48109.
E-mail: [e-mail]

Journal of Counseling Psychology

© 2010 American Psychological Association

2010, Vol. 57, No. 1, 128 –139

0022-0167/10/$12.00

DOI: 10.1037/a0018206

128

tion that has exhibited the symptom or developed the disease at
any time during a specified interval, for example, within the last
year. In psychological research, when a question begins with the
phrase “Have you ever,” the incidence period is effectively re-
spondents’ complete lifetime. Thus, lifetime incidence rates can
never be lower than the prevalence rate for the same sample.
Occasionally, psychological research surveys designed to assess
prevalence also include a time frame that includes the present, for
example, “within the past month.” In these cases, the inference is
that any occurrence during this period means that the respondent
has the condition currently.

Brown et al. (2007) reported that prevalence (within the last 12

months) in a small sample of male and female college students of
NSSI was 10%, whereas the incidence rate of past NSSI, defined
by lifetime occurrence but more than 12 months ago, was 18%.
Croyle and Waltz (2007) reported a incidence rate of 20% in a
sample of 290 college students when NSSI was defined as “more
injurious” behavior within the past three years (e.g., cutting, burn-
ing), in contrast to a rate of 31% for “mildly injurious” self-harm
(e.g., fingernail biting, skin picking). In two reports based on the
same sample, Gratz (2001) and Gratz et al. (2002) found a inci-
dence of 38% for college students when NSSI was defined as at
least one lifetime occurrence of self-harm. In still another set of
two reports based on a different small sample, Gratz (2006) re-
ported a lifetime incidence of 37% in women and 44% for men
(Gratz & Chapman, 2007). Klonsky and Olino (2008) found that
25% of their sample of 815 male and female undergraduates
reported at least one previous incident of NSSI. In a sample of
more than 2,800 male and female college students, Whitlock et al.
(2006) found that 17% reported having engaged in at least one
NSSI incident in their lifetime, and approximately 13% engaged in
NSSI more than once.

Among the demographic correlates of NSSI, sex differences

have received the most empirical attention. Studies based on
psychiatric samples suggest that the incidence of NSSI in women
is about 1.5 to 3 times higher than that in men (for a review, see
Yates, 2004), but it should be noted that women are overrepre-
sented in these clinical samples (Claes, Vandereycken, & Vertom-
men, 2007). Among college samples not drawn from students in
counseling, Brown et al. (2007), Gratz et al. (2002), and Whitlock
et al. (2006) found no sex difference when occurrence was defined
as a single NSSI incident. However, Whitlock et al. (2006) re-
ported the rate for repeated NSSI incidents was 1.5 times higher in
college women than men. Given these mixed findings, the first
goal for this study was to evaluate NSSI lifetime incidence rates in
a very large college sample and to do so separately for men and
women.

Our second and most important goal was to identify a subset of

items from a general symptom inventory now routinely adminis-
tered at many counseling centers that might be used to identify
clients who are at risk for NSSI. Early detection allows counselors
to address the factors that contribute to self-injury at the outset of
counseling and, in settings where time limits are flexible, to permit
more sessions at the beginning of counseling for working with
these difficult issues. Given the apparent effectiveness of the
checklist used by Whitlock et al. (2006) that asks directly about
NSSI behavior, it is reasonable to question why a new screening
measure for NSSI is needed. There are two reasons that such a
checklist is not appropriate for use as a counseling center screening

tool. First, studies that present checklists of the type used by
Whitlock et al., which ask directly about NSSI behaviors, always
assure respondents of confidentiality or, more typically, complete
anonymity. However, because of the highly stigmatized nature of
NSSI behavior, if such a checklist were presented to clients at
intake who know their responses will be reviewed by the counselor
they are about to meet, it is highly unlikely these clients would
respond as forthrightly as research subjects promised anonymity.
The inaccuracy of client self-reports at intake can be a concern for
a wide range of presenting problems and symptoms. Unfortu-
nately, no empirical studies could be located that have examined
how willing clients would be to disclose NSSI if asked directly by
a therapist, particularly at intake. However, some data suggest this
problem may be especially severe for NSSI. Of the 490 college
students who reported at least one self-injurious behavior on the
confidential Internet survey used by Whitlock et al. (2006), 36%
reported that no one else was aware of their NSSI behavior.
Although 53% of these 490 students reported being in counseling
for any reason, only 21% of the 490 reported that they had ever
disclosed their NSSI to a therapist. (Unfortunately, Whitlock et al.
did not report a separate breakdown of those who disclosed to a
counselor based on only those students who had been in counsel-
ing.)

The second reason measures used in NSSI research could not be

used for client screening is that, in an effort to keep the intake
process as streamlined as possible, few centers would administer
this type of disorder-specific checklist to all of their clients. We
believe that a procedure with the best chance of being adopted at
counseling centers must be based on a general intake inventory
already in use. Thus, we hoped to identify a subset of items from
a generalized distress measure now widely in use that could serve
as a red flag at intake to identify clients at risk for NSSI. These
clients could then be directly asked about self-injurious behavior
either by their intake counselor or later by the counselor assigned
to work with them. The general screening inventory examined in
this study was the Counseling Center Assessment for Psycholog-
ical Symptoms (CCAPS; Sevig, Soet, Malofeeva, & Dowis, 2006).
The CCAPS was an ideal candidate for this study because it has
recently been made available as an option in Titanium Schedule
software (Center for the Study of College Student Mental Health,
2006). This software is now used at more than 400 college coun-
seling centers. The archival data used in the current study was
originally collected as part of a project that used the CCAPS to
survey the general mental health of the entire student body at a
large Midwestern university.

We also hoped that by identifying specific CCAPS items asso-

ciated with NSSI, this study might advance research through a
better understanding of the comorbid symptoms, suggested etiol-
ogy, emotional dynamics, and factors that influence heightened
risk for self-injurious behavior. For example, specific emotional
states, particularly hostility, anger, sadness, anxiety, and deperson-
alization, seem to frequently precede acts of NSSI in both clinical
and nonclinical samples (Brown et al., 2007; Jacobson & Gould,
2007; Ross & Heath, 2003). Consequently, many theorists and
researchers consider NSSI to be part of a maladaptive coping
strategy used in an attempt to regulate affect and reduce tension
(Favazza, 1998; Haines & Williams, 2003). NSSI has also been
associated with posttraumatic stress disorder (Skegg, Nada-Raja,
& Moffitt, 2004; Zlotnick, Mattia, & Zimmerman, 1999). Fortu-

129

SELF-INJURY SCREENING FOR COLLEGE STUDENTS

nately, in addition to the CCAPS items, the original project survey
also included questions about experiences of traumatic events.
Therefore, we examined responses to these non-CCAPS items in
an effort to explore the role that trauma may play in the develop-
ment of NSSI behavior.

We developed screening profiles separately for men and women

because sex differences in expression or motivations for NSSI have
been noted in previous studies. For example, among college students,
women were found to be more likely to cut themselves than were men
(Whitlock et al., 2006). Likewise, in a psychiatric sample (Claes et al.,
2007) and in adolescent samples (Laye-Gindhu & Schonert-Reichl,
2005), women were found to be more likely to engage in cutting,
whereas men were more likely to hit or burn themselves. In terms of
functional aspects, women were more likely to use NSSI to relieve
intrapersonal, punitive, or aversive psychological states (e.g., self-
hatred), whereas men were more likely to engage in NSSI as a way
to communicate or influence other people, including to appear
tough to peers (Laye-Gindhu & Schonert-Reichl, 2005) or to get
attention from others (Claes et al., 2007). Finally, Gratz et al.
(2002) found that although dissociation was the strongest predictor
for NSSI in women, childhood separation was the most important
predictor for NSSI in men. Although we expected that different
sets of unique screening items would emerge in the risk profiles for
women and men, we also expected at least some items to be
common in both profiles.

To sum up, we could locate only one previous study that

estimated NSSI incidence rates from a sample of 2,000 or more
college students who responded to a campuswide random solici-
tation. We believe a sample of this size, drawn from all segments
of campus, is a minimum requirement to effectively study behavior
like NSSI with such a low base rate. Therefore, our first goal in
this study was to corroborate the large-sample, campuswide survey
conducted by Whitlock et al. (2006) by investigating the lifetime
incidence of NSSI with an equally large sample of college students
drawn from a different region of the United States. The few
available studies of college students suggest that single instances
of NSSI may be surprisingly common and— of even more con-
cern— occurrence of multiple, sustained episodes of NSSI are not
rare. Many students with high levels of NSSI apparently do not
seek counseling, and a substantial proportion of those who do, do
not bring these symptoms to the attention of their counselor
(Whitlock et al., 2006). Because of the stigmatized nature of NSSI,
it is very likely that many clients may not self-disclose NSSI at
intake. Although it is possible that some clients would self-disclose
after being asked directly by a counselor about this behavior, no
existing studies have investigated clients’ willingness to disclose
NSSI if asked directly at intake. Thus, the second purpose of this
study was to identify items from a general psychological symptom
inventory now in use on many college campuses that could be
used as a red flag to initially screen for NSSI. Our intention was
not to develop a final, definitive diagnostic instrument. Instead,
using a process similar to the one researchers have used to
identify subsets of Minnesota Multiphasic Personality Inven-
tory (MMPI) items to indicate potential for substance abuse
without asking directly (e.g., Clements & Heintz, 2002), we
hoped to identify subsets of CCAPS items for men and women
that could be used with all clients as a routine first step in a
multistage process to identify those at risk for NSSI. Finally, we
hoped that by identifying specific CCAPS items and perhaps

also experiences of severe trauma that are associated with
NSSI, this study could contribute to an increased understanding
of the etiology and dynamics of this behavior.

Method

Participants

This study used archival data from a project designed to inves-

tigate college students’ mental health, help-seeking behavior, and
recent psychological symptoms. An e-mail invitation was issued to
10,000 randomly selected students at a large public university in
the upper Midwest. Of these, 2,344 (23%) completed the survey
with valid responses and indicated their sex. Of these, 160 students
(6.8%) did not answer the NSSI screening question that asked
whether they had ever engaged in self-injurious behavior. Instead
of assuming that these students had never engaged in NSSI,
following the example of Whitlock et al. (2006), we excluded data
from cases where NSSI status could not be determined. There were
significant departures from univariate normality, as well as 60
multivariate outliers detected using the Mahalanobis distance cri-
terion suggested by Tabachnick and Fidell (2007). Outliers in-
cluded many students who engaged in NSSI. Therefore, we de-
cided not to exclude data from these students.

Thus, data from 1,136 male and 1,048 female participants were

used in the current study. These participants’ mean age was 22.61
years (SD

⫽ 4.98 years, range ⫽ 18–62 years). A total of 199

(9.1%) indicated they were international students, and 11 (0.5%)
did not answer this question. Of the 1,974 (90.4%) U.S. students,
1,239 (62.8%) described themselves as “European American/
Caucasian/White (non-Hispanic),” 213 (10.8%) as “African/
African American,” 19 (1.0%) as “Arab/Arab American/Persian,”
109 (5.5%) as “Asian/Asian American/Pacific Islander,” 206
(10.4%) as “Latino/Latina,” 120 (6.1%) as “multiracial,” 9 (0.5%)
as “Native American,” and 47 (2.4%) as “other”; 12 (0.6%) did not
report their race or ethnicity. The sample contained 796 (36%)
graduate and professional students, with the remainder being un-
dergraduates. In response to a question that asked, “What is your
sexual orientation?” 2,013 (92%) indicated “heterosexual,” 50
(2.3%) indicated “gay male,” 14 (0.6%) indicated “lesbian,” 58
(2.7%) indicated “bisexual,” and 34 (1.6%) indicated “question-
ing”; 15 (0.7%) did not report their sexual orientation. Using data
available from institutional sources, we compared our sample with
the total student body. There were no statistically significant
differences in sex or proportion of graduate students. However, our
sample contained fewer international students (9.2%) compared
with the university population (11.7%),

␹

2

(1, N

⫽ 42,198) ⫽

12.91, p

⬍ .001, and fewer students in the age category of “36

years or older” (2.6%) than in the university population (3.7%),
␹

2

(7, N

⫽ 42,148) ⫽ 18.8, p ⬍ .01. A 2 ⫻ 2 chi-square comparing

proportions of all students who indicated a race or ethnicity other
than “European American/Caucasian/White (non-Hispanic)” indi-
cated that our sample contained more minority students (36.9%)
than did the university population (31.5%),

␹

2

(1, N

⫽ 37,298) ⫽

24.00, p

⬍ .001.

Measures

Self-injurious behaviors.

Participants were presented with a

NSSI screening question that asked, “Have you ever intentionally

130

CHENG, MALLINCKRODT, SOET, AND SEVIG

hurt yourself?” If they answered “yes,” they were directed to
answer a 16-item checklist of NSSI behaviors that Whitlock et al.
(2006) developed from existing NSSI measures, a review of liter-
ature, and interviews with mental health professionals and people
who have deliberately injured themselves without suicidal intent.
After answering these 16 items, participants were asked, “Approx-
imately on how many total occasions have you intentionally hurt
yourself?” The question was followed by seven choices: only once,
2–3 times, 4 –5 times, 6 –10 times, 11–20 times, 21–50 times, more
than 50 times. Whitlock et al. (2006) did not report reliability or
validity information, but NSSI in their study was found to be
associated with history of abuse, suicide ideation, level of psycho-
logical distress, and eating disorder characteristics.

CCAPS.

The CCAPS (Sevig et al., 2006) was developed to

assess college students’ current mental health concerns. The orig-
inal version contains 70 items preceded by the stem “How well
does this item describe you?” for which respondents are to con-
sider a time frame of the past two weeks.

1

Respondents answer

using a 5-point partially anchored response scale (0

⫽ not at all,

4

⫽ extremely well). The CCAPS contains the following nine

subscales: Depression (15 items), Eating Issues (10 items), Sub-
stance Use (eight items), General Anxiety (seven items), Hostility
(six items), Social Role Anxiety (six items), Family of Origin
Concerns (five items), Academic Issues (five items), and Spiritu-
ality Concerns (three items). Five additional items assess (a)
homicidal thoughts, (b) history of abuse in the family, (c)
dissociative symptoms (two items), and (d) cultural and/or
ethnic identity. Soet and Sevig (2006) reported reliability co-
efficients (coefficient alpha) ranging from .80 to .93 for the nine
CCAPS subscales in a sample of 2,155 college students. Con-
struct validity was demonstrated through correlations between
five of the CCAPS subscale scores (Eating Issues, General
Anxiety, Depression, Academic Role, Social Role Anxiety) and
a diagnostic index score (psychiatric diagnosis) received by
clients in a clinical sample of 218 students. Additionally, col-
lege students who were receiving counseling reported signifi-
cantly higher scores on all nine subscales than did those who
were not currently receiving counseling. In the current study,
data from the item “I have thoughts of ending my life” from the
CCAPS Depression subscale was excluded to avoid the possi-
bility of inflating correlation of the Depression subscale with
self-injury behaviors. In this study, internal reliability (coeffi-
cient alpha) for the subscales ranged from .69 (Social Role
Anxiety) to .94 (Spirituality).

Traumatic life events.

A measure of past traumatic experi-

ence adapted from Brack, Brack, and Hill-Carlson (1997) was
included in the archival battery. The measure was developed on the
basis of one of the key Diagnostic and Statistical Manual of
Mental Disorders (4th ed.; DSM–IV; American Psychiatric Asso-
ciation, 1994) criteria for posttraumatic stress disorder (PTSD).
Participants were asked, “Have you ever experienced an event or
events that involved actual or perceived threatened death or serious
injury or a threat to your physical integrity?” They were also
asked, “Have you ever witnessed an event that involved death,
injury, or a threat to the physical integrity of another person and in
response to which you felt helplessness or horror or intense fear?”
If they answered “yes,” to either question, the variable was coded
1; if the response was no or left blank, a 0 was assigned. Brack et

al. (1997) did not report reliability or validity properties for this
instrument.

Procedures

For the original project that generated the archival data used in

the current study, the university registrar generated a random
sample of 10,000 e-mail addresses of currently enrolled students.
Students were sent an e-mail inviting them to participate in an
online survey about college student mental health. In this invita-
tion, students were provided with a Web link that directed them to
the survey at a third-party website. Participants filled out this Web
survey after reading and completing the informed consent. Data
were collected between November 2006 and January 2007. After
completing the survey, participants were asked if they would like
to enter a drawing for ten $50 cash prizes (totaling $500). If they
said yes, they were instructed to send their names and addresses
via e-mail to ensure the confidentiality of their online survey
responses.

Results

Incidence of Self-Injurious Behaviors

Among the 2,184 participants, 119 (10.5%) men and 169

(16.1%) women responded “yes” to the NSSI screening question
indicating that they had engaged in self-injurious behavior, a sex
difference that was statistically significant,

␹

2

(1, N

⫽ 2,184) ⫽

15.20, p

⬍ .001. When the threshold for NSSI was defined as at

least two to three times in one’s lifetime, the incidence rate
decreased to 9.0% (n

⫽ 102) for men and 13.5% (n ⫽ 142) for

women, a sex difference that was also statistically significant,
␹

2

(1, N

⫽ 2,184) ⫽ 11.48, p ⬍ .01. When the threshold was

further raised to engaging in NSSI at least four to five times during
one’s lifetime, the incidence rate was 5.3% (n

⫽ 60) for men and

9.3% (n

⫽ 97) for women, a difference that was also statistically

significant,

␹

2

(1, N

⫽ 2,184) ⫽ 12.90, p ⬍ .001. For those who

had engaged in NSSI, men engaged in a mean of 2.78 different
forms (SD

⫽ 2.36, range ⫽ 1–15), whereas women engaged in a

mean of 2.88 different methods of injuring themselves (SD

⫽ 1.91,

range

⫽ 1–9).

Before screening items could be identified, it was necessary to

develop an operational definition of NSSI from the archival data.
We wanted to set the threshold high enough to exclude isolated
acts that did not constitute an ongoing pattern of NSSI. Thus, we
defined significant NSSI as answering “yes” to the initial screening
question “Have you ever intentionally hurt yourself?” followed by
a response of at least “4 –5 times” to the question that asked about
the number of occasions they had engaged in any of the listed
forms of NSSI. This threshold is as consistent as the archival data
would allow with recommendations for NSSI to be recognized as
a distinct DSM disorder and suggest that a repetitive pattern
consists of five or more incidents (Muehlenkamp, 2005). In the

1

A 62-item version of the CCAPS has recently been developed to

replace the 70-item version for use in the Titanium Schedule software.
Although the full 70-item version of the CCAPS was presented to respon-
dents in the original archival study, we have limited our selection analyses
to only the 62 items that remain available.

131

SELF-INJURY SCREENING FOR COLLEGE STUDENTS

current study, a total of 97 (9.3%) women and 60 (5.3%) men met
these criteria. Persons who reported that NSSI occurred once in
their lifetime (17 men and 27 women) or only two to three times
(42 men and 45 women) were excluded from further analyses.

Of the 976 women who remained in the analyses, 20% indicated

that they had witnessed serious trauma and 18% indicated they had
experienced serious trauma. Of the 1,077 men who were retained for
analysis, 15% had witnessed and 14% had experienced serious
trauma. Women were significantly more likely to report witnessing
trauma,

␹

2

(1, N

⫽ 2,053) ⫽ 7.058, p ⬍ .01, and experiencing trauma,

␹

2

(1, N

⫽ 2,053) ⫽ 6.64, p ⬍ .05. Table 1 shows means, standard

deviations, and results of Pearson correlations between CCAPS
subscales, traumatic events experienced or witnessed, and NSSI
group membership. Because significant NSSI and traumatic expe-
riences were coded 1 versus 0, positive point biserial coefficients
indicate variables positively associated with NSSI or trauma ex-
periences. Results for women are shown above the diagonal and
for men below the diagonal.

Identification of NSSI Screening Items

We explored group differences for the nine CCAPS subscales

using a separate one-way multivariate analysis of variance
(MANOVA) for women and men. Results suggested a significant
NSSI versus non-NSSI difference for men, F(9, 1050)

⫽ 12.01,

p

⬍ .001, and a significant difference for women, F(9, 960) ⫽

12.37, p

⬍ .001. Univariate follow-up comparisons suggested

women with significant NSSI reported greater symptoms on all
nine CCAPS subscales than did non-NSSI women. Men with
significant NSSI scored higher than did their counterparts on every
CCAPS subscale except spirituality. (To conserve space, these
findings are presented in Appendix B online in the supplemental
materials.) To examine whether there were Sex

⫻ NSSI interaction

effects on the CCAPS subscales, we performed a two-way factorial
MANOVA. No significant interaction effects of Sex

⫻ NSSI were

found, multivariate F(9, 2018)

⫽ 1.185, p ⫽ ns.

For the purposes of developing a screening inventory, binary

logistic regression is more useful than MANOVA because it can
be used with nominal variable predictors (Tabachnick & Fidell,
2007). We performed separate analyses for men and women, using
forward stepwise selection with the maximum likelihood ratio
criterion. Possible predictors were the nine CCAPS subscales, plus
two variables coding traumatic events experienced or witnessed as
1 if yes and 0 if no. The dependent variable was coded 1 for
significant NSSI and 0 for no NSSI. Prior to the analyses, we
performed tests for multicollinearity on all of the independent
variables. Results of condition index, eigenvalues, and variance
portions showed that these predictor variables were adequately
independent of each other for both men and women. Thus, all of
the independent variables were retained. Table 2 shows results of
these analyses. In analyses for women, four variables emerged as
significant predictors of NSSI using the Wald criterion. These
were (in the order selected) the CCAPS subscales of Hostility,
Depression, and Spirituality, followed by experience of serious
trauma. Comparison of the constant-only model with a model that
included these four variables suggested that they provided a sig-
nificant increment in fit,

␹

2

(4, N

⫽ 977) ⫽ 95.66, p ⬍ .001. For

men, three variables emerged as significant predictors. They were
(in the order selected) CCAPS Depression, witnessed a traumatic
event, and CCAPS Hostility. Comparison of the constant-only
model with a model that included these three variables indicated a
significant increment in prediction,

␹

2

(3, N

⫽ 1,077) ⫽ 82.289,

p

⬍ .001. Although both models were statistically significant, an

examination of the classification table for each analysis revealed
that the selected sets of predictors correctly classified only 9.3% of
the 97 women in the significant NSSI group and only 6.7% of the
60 men in the significant NSSI group.

On the basis of these findings, we concluded that CCAPS

subscale scores together with whether a traumatic event was wit-
nessed or experienced could not be the basis for a useful NSSI
screening instrument. Therefore, we turned next to an analysis of
individual CCAPS items, together with witnessing or experiencing

Table 1
Means, Standard Deviations, and Correlations for Women and Men

Measure

M

SD

1

a

2

3

4

5

6

7

8

9

10

11

a

12

a

M

0.10

1.16

0.80

0.97

0.66

1.35

0.99

1.41

2.36

1.10

0.18

0.20

SD

0.30

0.83

0.64

0.74

0.62

0.69

0.73

0.89

1.42

0.76

0.38

0.40

1. NSSI

a

0.06

0.23

—

.13

.14

.23

.28

.20

.18

.15

.12

.25

.16

.11

2. Eating Issues

0.73

0.61

.14

—

.43

.45

.33

.36

.32

.41

.03

.56

.05

.00

3. Substance Use

0.84

0.68

.14

.35

—

.37

.35

.31

.27

.34

.23

.37

.07

.02

4. General Anxiety

0.76

0.64

.22

.45

.42

—

.56

.53

.45

.51

.10

.70

.17

.04

5. Hostility

0.67

0.62

.23

.42

.46

.59

—

.52

.56

.39

.04

.55

.21

.14

6. Social Role Anxiety

1.37

0.71

.20

.35

.31

.59

.51

—

.48

.44

.12

.63

.06

⫺.00

7. Family of Origin Concerns

0.86

0.71

.19

.37

.34

.50

.60

.44

—

.38

.17

.59

.12

.08

8. Academic Issues

1.33

0.87

.14

.42

.36

.52

.41

.45

.40

—

.07

.66

.06

.05

9. Spirituality Concerns

2.50

1.38

.05

.01

.17

.03

.03

.09

.11

.09

—

.08

.08

⫺.01

10. Depression

0.96

0.69

.28

.47

.40

.74

.57

.68

.60

.63

.11

—

.12

.09

11. Trauma experienced

a

0.14

0.35

.09

.09

.08

.15

.15

.10

.14

.10

.01

.14

—

.28

12. Trauma witnessed

a

0.15

0.36

.12

.05

.06

.08

.14

.07

.08

.03

⫺.03

.09

.20

—

Note.

Results for woman (n

⫽ 976) are shown above the diagonal, for men (n ⫽ 1,077) below the diagonal. NSSI ⫽ Nonsuicidal self-injury, coded as

0

⫽ none, 1 ⫽ significant NSSI. Variables 2 through 10 are Counseling Center Assessment of Psychological Symptoms subscales. Trauma experienced

and witnessed coded as 0

⫽ no, 1 ⫽ yes.

a

Point biserial correlation.

132

CHENG, MALLINCKRODT, SOET, AND SEVIG

serious trauma. Because 50 subjects per predictor has been sug-
gested as a minimum for logistic regression (Wright, 1995), this
method would not be appropriate for simultaneous analysis of all
CCAPS items. Instead, we began by randomly dividing the sample
into an exploratory sample and a holdout 50% sample for both men
and women. In the exploratory sample of women (n

⫽ 486) and

men (n

⫽ 530), the CCAPS 5-point partially anchored response

scale was recoded into a dichotomy with scale points 3 or 4
(extremely well) recoded as 1 and scale points 0 (not at all), 1, or
2 recoded as 0. Recoding the response scale allowed us to create
a series of 2

⫻ 2 contingency tables for each of the items as

predictors of NSSI status and then to test these associations with
Fisher’s exact test (two-sided). This procedure is recommended
when marginal probabilities are markedly imbalanced or when cell
sizes are less than 10 in a 2

⫻ 2 contingency table (Mehta & Patel,

1997).

An item was selected for the final pool only if responses were

significantly different between NSSI and non-NSSI groups of
women at the p

⬍ .007 level. Note that because we were not

testing hypotheses, we were not concerned about Type I errors in
these multiple tests. Instead, we set a p level that resulted in an
initial pool of the best 20 or fewer items that discriminated be-
tween the NSSI and non-NSSI groups. After inspecting the results,
we selected the .007 p level because it yielded a final pool of 15
items for women. For men, we set the criteria even more strin-
gently, at p

⬍ .005, although nevertheless 22 CCAPS items re-

mained in the pool at that level. We decided not to impose even
more stringent criteria because at the next increment ( p

⬍ .002),

too many items were removed. (SPSS output clustered items tied
at the same low levels of p

⬍ .007, p ⬍ .005, and p ⬍ .002)

Although the item “I have thoughts of ending my life” met the
criteria for both men and women, it was excluded because it might
represent a confound in associations with NSSI. In the supplemen-
tal materials, Appendix A shows the items selected through this
process. (Note that the CCAPS presented in the original project
used somewhat different instructions than are now used in the
Titanium software.) The items “I feel comfortable around other
people,” “I am enthusiastic about life,” and “I like myself” were
reverse scored. Appendix A in the supplemental materials shows

that 11 screening items were common to men and women, four
items were selected only for women, and 11 items only for the
men. However, most items selected for only one subscale on the
basis of the exploratory half sample at the p

⬍ .005 level for men

and p

⬍ .007 level for women were actually significant predictors

for the other sex at least at the p

⬍ .05 level in the full sample.

Although Items 12–26 were used for only one of the two measures,
it would be incorrect to characterize items designated with an a
superscript as unique predictors for only one sex and not the other.

The 15 items for women and 22 items for men were used to

calculate NSSI screening scores. Scores on the women’s screening
inventory could range from 0 – 60. Actual scores in the full sample
ranged from 0 –54 (M

⫽ 12.23, SD ⫽ 9.36, Mdn ⫽ 10). Not

surprisingly for a screening instrument, there was high positive
skew (Z

⫽ 13.82). Kurtosis was also significant (Z ⫽ 7.39).

Coefficient alpha for the 15 items was .88. For the men’s screening
inventory, scores could range from 0 – 88. Actual scores in the full
sample ranged from 0 –77 (M

⫽ 19.50, SD ⫽ 14.42, Mdn ⫽ 16).

Similar to the women’s screening instrument, there was high
positive skew (Z

⫽ 13.05). Kurtosis was also significant (Z ⫽

4.80). Coefficient alpha for the 22 items was .92. Using only the
holdout validation sample of women (n

⫽ 490) and men (n ⫽

547), logistic regression analyses were used to examine the screen-
ing inventory as a predictor of NSSI status. For women, compar-
ison of the constant-only model with a model that included the
screening score indicated a significant increment in prediction,
␹

2

(1, N

⫽ 490) ⫽ 49.81, p ⬍ .001. The increment for men was

also significant,

␹

2

(1, N

⫽ 547) ⫽ 45.58, p ⬍ .001. Thus, it

appears that the two screening inventories developed from the
exploratory sample were also significant predictors of NSSI status
in the validation sample.

Next, we examined whether self-reports of trauma could serve

to improve the predictive value of the CCAPS items alone. Step-
wise logistic regression analyses were conducted using the full
sample. In the first step, the CCAPS screening score was entered.
In the next step, forward stepwise selection based on the Wald
criteria was used to examine the two dichotomous variables of
serious trauma experienced and witnessed. Results are shown in
Table 3. For women, self-reported trauma experienced and trauma

Table 2
Logistic Regression Predicting Nonsuicidal Self-Injury (NSSI) From CCAPS Instrument Subscales

Predictor

B

SE

Wald

Odds ratio

95% confidence interval

Lower

Upper

Women (n

⫽ 976)

Constant

⫺4.515

.359

158.620

ⴱⴱⴱ

0.011

CCAPS Hostility

0.742

.186

16.321

ⴱⴱⴱ

2.099

1.465

3.008

CCAPS Depression

0.575

.157

13.438

ⴱⴱⴱ

1.777

1.307

2.417

CCAPS Spirituality Concerns

0.284

.090

10.042

ⴱⴱⴱ

1.328

1.114

1.583

Trauma experienced

0.745

.252

8.765

ⴱⴱ

2.107

1.286

3.451

Men (n

⫽ 1,077)

Constant

⫺4.851

.326

221.568

ⴱⴱⴱ

0.008

CCAPS Depression

1.087

.197

30.583

ⴱⴱⴱ

2.966

2.017

4.360

Trauma witnessed

0.856

.314

7.434

ⴱⴱ

2.354

1.272

4.357

CCAPS Hostility

0.482

.218

4.889

ⴱⴱ

1.620

1.056

2.484

Note.

Dependent variable coded as 0

⫽ no NSSI, 1 ⫽ significant NSSI. CCAPS ⫽ Counseling Center Assessment for Psychological Symptoms.

ⴱⴱ

p

⬍ .01.

ⴱⴱⴱ

p

⬍ .001.

133

SELF-INJURY SCREENING FOR COLLEGE STUDENTS

witnessed together resulted in a significant increase in prediction
of NSSI,

␹

2

(2, N

⫽ 976) ⫽ 13.63, p ⬍ .01. For men, only

self-reported trauma witnessed significantly increased prediction,
␹

2

(1, N

⫽ 1,077) ⫽ 6.22, p ⬍ .05. Despite the significant incre-

ments, classification results were once again disappointing, with
less than 10% of the women and 8% of the men with NSSI
correctly classified. When the a priori cut values for the classifi-
cation analysis were changed from the default of 50/50 to a 90/10
split to more closely approximate the actual proportion of NSSI
occurring in the sample, 63% of the women and 52% of the men
with NSSI were correctly classified.

In binary classification terminology, these percentages of true

positives are the sensitivity rates for a predictor. The rates of true
negatives (i.e., persons classified as having no NSSI when in fact
they did not) were 99.5% for women and 99.8% for men. These are
the specificity rates. For the purposes of developing most screen-
ing instruments, the sensitivity rate is much more important than
specificity, because it is usually more important to detect true
positives than to accurately rule out persons without the condition.
(Note that by simply classifying the entire sample as “no NSSI,” an
overall correct identification rate of 90% could be achieved for
women and 94% for men in this sample, albeit at a cost of 0%
sensitivity.) Because logistic regression gives equal weight to
classifying false positives and false negatives, if detecting true
positives is a priority, results from logistic regression are not
helpful for determining a screening cutoff score that produces the
most favorable tradeoff between sensitivity and specificity.

To find the best cutting scores, we used the results of the logistic

regression analyses shown in Table 3 as a starting point to incor-
porate information about trauma. Cohen, Cohen, West, and Aiken
(2003, p. 493) explained that odds ratios can be used as coeffi-
cients to weight raw scores to create a composite. Odds ratios from
Table 3 were rounded to the nearest whole point to mirror the kind
of quick calculations that would be performed in an intake situa-
tion. Thus, for women, NSSI screen

⫽ (sum of 15 CCAPS

items)

⫹ 2 points if trauma experienced ⫹ 2 points if trauma

witnessed. For men, NSSI screen

⫽ (sum of 22 CCAPS items) ⫹

2 points if trauma witnessed. Note that the sum of CCAPS items
was based on the 0 – 4-point response scale.

To identify useful cutoff values for these composite scores, we

calculated receiver-operating characteristic (ROC) curves for men

and women. Figure 1 (see the supplemental materials, Appendix
C) shows the ROC curve for women. The amount of shift of the
curve upward and to the left of the diagonal indicates the degree
that scores can be used to discriminate individuals with a greater
than chance probability of engaging in NSSI. The shape of the
curve defines trade-offs between detecting true positive cases at
the cost of increasing false positive cases. The null hypothesis of
chance classification corresponds to the diagonal, an area under the
curve, or discriminative index (DI) equal to .5. The DI for perfect
prediction is equal to 1.0. The ROC curve analysis indicated DI

⫽

.755 for the entire sample of women, with a 95% confidence
interval of .702–.808 (SE

⫽ .027, p ⬍ .001). Figure 2 (see the

supplemental materials, Appendix C) shows the ROC curve for all
men. The shift upward and to the left included .815 of the curve
area with a 95% confidence interval of .759 –.870 (SE

⫽ .028, p ⬍

.001).

Table 4 reports selected coordinates of the ROC curve for

women in the range of possible cutoff scores between 20 and 40.
The second column indicates the proportion of true positive NSSI
cases (i.e., sensitivity) that could be expected at a particular cutoff
score. The third column indicates the proportion of false positives
(i.e., 1 – specificity) that would be identified at the same cutoff
score. Note that lowering the cutoff score results in the identifi-
cation of a higher proportion of true positives (i.e., sensitivity is
increased by lowering the cutoff score), but at the cost of including
more false positive cases (i.e., specificity is decreased by lowering
the cutoff score). Table 4 suggests that a cutoff score greater than
22 could be expected to identify about 48.5% of the women in the
significant NSSI group, with a false positive rate of only 13.2%.
However, note that 13.2% of 879 true NSSI-negative women
corresponds to about 116 incorrectly classified cases. Increasing
the cutoff score to be greater than 27 would still correctly identify
36.1% of the true NSSI-positive women at a cost of only 5.3%
false positives. The last two columns present coordinates for cutoff
scores based only on CCAPS items for settings in which informa-
tion about witnessing or experiencing trauma is not available.

Table 5 reports ROC curve coordinates for the range of possible

cutoff scores between 30 and 50 for men. Again, the second
column shows the sensitivity for a given cutoff score, and the third
column shows the proportion of false positives (i.e., 1 – specific-
ity) at this cutoff score. Thus, a cutoff score for men that is greater

Table 3
Final Model Logistic Regression Predicting Nonsuicidal Self-Injury (NSSI) From Screening Instruments and Trauma

Predictor

B

SE

Wald

Odds ratio

95% confidence interval

Lower

Upper

Women (n

⫽ 976)

Constant

⫺3.907

0.247

245.286

ⴱⴱⴱ

0.021

CCAPS screening score

0.090

0.011

65.549

ⴱⴱⴱ

1.094

1.070

1.118

Trauma experienced

0.671

0.258

6.761

ⴱⴱ

1.956

1.180

3.244

Trauma witnessed

0.526

0.262

4.047

ⴱ

1.693

1.014

2.826

Men (n

⫽ 1,077)

Constant

⫺4.912

0.329

233.293

ⴱⴱⴱ

0.007

CCAPS screening score

0.070

0.008

68.923

ⴱⴱⴱ

1.073

1.055

1.091

Trauma witnessed

0.816

0.314

6.726

ⴱⴱ

2.261

1.220

4.187

Note.

Dependent variable coded as 0

⫽ no NSSI, 1 ⫽ significant NSSI. Trauma experienced and Trauma witnessed coded 0 ⫽ no, 1 ⫽ yes.

ⴱ

p

⬍ .05.

ⴱⴱ

p

⬍ .01.

ⴱⴱⴱ

p

⬍ .001.

134

CHENG, MALLINCKRODT, SOET, AND SEVIG

than 39 could be expected to identify 48.3% of the significant
NSSI cases, with a false positive rate of 8.4%. Raising the cutoff
score to 45 could still be expected to capture one third of the true
positive cases of men with significant NSSI, with the false positive
rate dropping to 4.7%. The last two columns of the table provide
cutting score information if the experience of witnessing trauma
cannot be assessed. Inclusion of the trauma items increased the
area under the ROC from .745 to .755 for women and from .811
to .815 for men.

Discussion

Our two goals in this study were to investigate the incidence of

NSSI at a large Midwestern public university and to develop
screening instruments to identify men and women with a serious
level of NSSI. Because Whitlock et al. (2006) appears to be the
only study to date that collected a large (N

⬎ 2,000) sample with

all students on campus eligible randomly for selection, the first
goal of this study was to provide a second estimate of NSSI
incidence from a campus in a different region of the United States.
Our findings indicated that 16.1% of women and 10.5% of men
reported at least one NSSI incident in their lifetime, a rate that is
significantly lower than the 17% reported by Whitlock et al. using
the same criterion,

␹

2

(1, N

⫽ 5,059) ⫽ 13.89, p ⬍ .001. (Whitlock

et al., 2006, did not provide a separate breakdown by sex.) How-
ever, if two or more lifetime incidents are considered, the rates for
Whitlock et al. and this study are 12.0% and 11.2%, respectively,
a difference that is not statistically significant,

␹

2

(1, N

⫽ 5,059) ⫽

0.64, p

⫽ ns.

Other studies using different NSSI measures in much smaller

samples reported occurrence rates ranging from 14% to 44%

(Brown et al., 2007; Croyle & Waltz, 2007; Favazza et al., 1989;
Gratz, 2001, 2006; Gratz & Chapman, 2007; Gratz et al., 2002;
Klonsky & Olino, 2008). Perhaps the discrepancies can be ex-
plained, in part, by differences in definitions of NSSI, more selec-
tive sampling procedures, and different survey instructions to
respondents.

Our results indicated that women were significantly more likely

to engage in NSSI than were men, regardless of whether it was a
single lifetime instance or the more stringent criterion of at least
four to five occurrences (9.3% for women vs. 5.3% for men).
Writers have speculated that sociocultural stressors such as sexual
violence and sexism place tremendous coping burdens on women
(Alexander & Clare, 2004) at the same time traditional feminine
socialization gives women fewer options than men for expressing
negative emotions, thus leaving them more at risk for adopting
maladaptive strategies such as self-injury as a means of coping and
possibly also self-expression. Clearly more studies with represen-
tative nonclinical samples are needed to explore sex differences in
NSSI (Claes et al., 2007; Yates, 2004).

Turning now to the second purpose of this study, we were able to

identify two sets of CCAPS items that could be used as a red flag to
identify men and women who might have significant NSSI. Although
the discriminating power of the scales is modest, the DIs of .755 for
women and .815 for men in this study compare favorably to DIs for
other indirect assessment measures, that is, measures that do not ask
directly about the condition of interest. For example, the MacAndrew
Alcoholism Scale—Revised and the Addiction Potential Scale are
special scale subsets of MMPI–2 items used to signal the possibility
of substance abuse. Clements and Heintz (2002) reported ROC curve
DIs for these widely used indirect scales of .71 and .68, respectively.

Table 4
Selected Receiver-Operating Characteristic Curve Coordinates for Women’s Nonsuicidal
Self-Injury Screening Instrument

Cutoff score

With trauma items included

CCAPS items only

Sensitivity

1

⫺ specificity

Sensitivity

1

⫺ specificity

20.5

.536

.171

.495

.115

21.5

.515

.152

.464

.137

22.5

.485

.132

.443

.117

23.5

.454

.107

.423

.099

24.5

.423

.086

.392

.078

25.5

.392

.075

.340

.067

26.5

.381

.063

.309

.050

27.5

.361

.053

.299

.043

28.5

.309

.041

.258

.034

29.5

.278

.034

.227

.030

30.5

.247

.028

.216

.026

31.5

.216

.028

.206

.025

32.5

.206

.025

.206

.020

33.5

.196

.023

.196

.019

34.5

.186

.020

.175

.017

35.5

.175

.019

.165

.015

36.5

.155

.015

.134

.009

37.5

.134

.013

.103

.008

38.5

.113

.009

.093

.008

39.5

.103

.006

.072

.005

40.5

.082

.003

.062

.003

Note.

Sensitivity

⫽ proportion of true positives; 1 ⫺ specificity ⫽ proportion of false positives; CCAPS ⫽

Counseling Center Assessment for Psychological Symptoms.

135

SELF-INJURY SCREENING FOR COLLEGE STUDENTS

The subscales identified in this study are not intended to provide

a definitive diagnosis. Instead, they are best used as the first stage
of a two-step process. At counseling centers that routinely admin-
ister the CCAPS at intake, subscale scores could be calculated
fairly easily. Clients with scores above a predetermined value
could be designated for more detailed follow-up assessment by
their assigned counselor. This process is similar to the two-stage
diagnostic screening procedure common for many medical condi-
tions. An initial, relatively inexpensive test (e.g., mammography
for women or prostate-specific antigen test for men) is routinely
administered to all patients at risk for the disease. However, this
initial test is much less accurate than other, more expensive pro-
cedures. Rather than base a diagnosis on the results of the first
screening alone, the physician sets an indicator threshold that he or
she knows will result in a great number of false positives, but the
initial result is used only as an indication to order the more
expensive (and accurate) diagnostic procedure (e.g., a CAT scan).
Similarly, we believe that scores on the subscales we have iden-
tified should be used only as an indication that further assessment
is warranted.

The specific items identified that differentiated students who

engage in NSSI from others suggest important clues about the
nature of self-injury as a psychological symptom. In the supple-
mental materials, Appendix A shows the 11 CCAPS items com-
mon for identifying both men and women with significant levels of
NSSI (i.e., four to five or more lifetime incidents). These common
items tap symptoms of depression (e.g., feeling helpless, sad all the
time) and also suggest a theme of self-loathing (e.g., disliking
oneself), together with anxiety (e.g., heart racing) and anger,
mixed with themes of losing control (e.g., difficulty controlling

temper, feeling an urge to break or smash things). Common items
also included symptoms of PTSD such as unwanted thoughts,
sleep difficulties, nightmares, flashbacks, or dissociation (e.g.,
feeling disconnected from oneself). These 11 common items are
consistent with previous research that suggests emotional states
such as hostility, anger, sadness, anxiety, and depersonalization are
correlates or antecedents of NSSI (Jacobson & Gould, 2007; Ross
& Heath, 2003). The findings are also consistent with previous
studies reporting that NSSI is associated with mood disorder,
anxiety disorder, and PTSD (Skegg et al., 2004; Zlotnick et al.,
1999). Persons who self-injure, compared with those who do not,
report higher levels of negative mood both in their general daily
life and during a specific NSSI event (Chapman, Gratz, & Brown,
2006) and a coping style that relies heavily on behavioral disen-
gagement and avoidance strategies (Andover, Pepper, & Gibb,
2007; Brown et al., 2007). Overall, our findings are also consistent
with previous studies that reported correlations between NSSI and
interpersonal, familial, and career difficulties (Evans et al., 2005;
Hawton & Harriss, 2007).

In addition to the common items, we identified four CAPS items

that were relatively better at identifying women with NSSI and 11
CAPS items that were relatively better at identifying men with
NSSI. However, as noted previously, it would be an error to
characterize the items designed with an a superscript in Appendix
A (found in the supplemental material) as unique predictors of
NSSI in one sex but not the other. Although these items were
assigned to only one subscale, all of those designated with a
superscript were also significant predictors at least at the p

⬍ .05

level for the other sex. Many of the same themes appear in
somewhat different forms for each subscale: for example, personal

Table 5
Selected Receiver-Operating Characteristic Curve Coordinates for Men’s Nonsuicidal Self-Injury
Screening Instrument

Cutoff score

With trauma items included

CCAPS items only

Sensitivity

1

⫺ specificity

Sensitivity

1

⫺ specificity

30.5

.650

.191

.650

.185

31.5

.633

.174

.617

.168

32.5

.617

.157

.600

.152

33.5

.600

.148

.600

.146

34.5

.600

.137

.583

.135

35.5

.533

.128

.517

.124

36.5

.517

.116

.517

.114

37.5

.517

.106

.517

.101

38.5

.483

.099

.467

.094

39.5

.483

.084

.467

.081

40.5

.450

.076

.417

.071

41.5

.450

.066

.417

.064

42.5

.400

.061

.383

.061

43.5

.367

.055

.350

.052

44.5

.350

.053

.333

.049

45.5

.333

.047

.317

.045

46.5

.283

.042

.283

.040

47.5

.267

.033

.233

.031

48.5

.250

.029

.217

.028

49.5

.217

.028

.217

.027

50.5

.217

.028

.217

.027

Note.

Sensitivity

⫽ proportion of true positives; 1 ⫺ specificity ⫽ proportion of false positives; CCAPS ⫽

Counseling Center Assessment for Psychological Symptoms.

136

CHENG, MALLINCKRODT, SOET, AND SEVIG

isolation for women (“I feel comfortable around other people”
reverse scored) and for men (“I don’t enjoy being around people as
much as I used to”). Again, we emphasize it would be an error to
conclude that men with NSSI did not experience panic attacks or
that women with NSSI did not experience racing thoughts.

Two CCAPS items that significantly predicted NSSI for men

but not women were “My family gets on my nerves” and “I have
thoughts of hurting others.” These findings are an ominous indi-
cation that some men prone to self-harm also appear to struggle
with the impulse to harm others. Research has long suggested that
among persons with marked depression, those who self-injure
report more extrapunitive, outward hostility than do depressed
persons with no NSSI (Bennum & Phil, 1983). In a study of
adolescents, self-injuring boys were found to exhibit more extra-
punitive acting out hostility than were self-injuring girls (Ross &
Heath, 2003). Certainly more research is needed to examine inter-
connections between anxiety, depression, anger, self-harm, and
danger to others in both young men and young women.

Perhaps the most striking difference between men and women

was that experiencing a traumatic event was an NSSI indicator
only for women, whereas witnessing trauma was an indicator for
both men and women. Witnessing or experiencing traumatic
events are both diagnostic criteria of PTSD in the DSM–IV (Amer-
ican Psychiatric Association, 2000). Clinically, therefore, it is
important to assess whether men and women who have witnessed
or experienced traumatic events engage in NSSI as part a larger
pattern of PTSD, because some PTSD symptoms (e.g., anhedonia,
psychic numbness) may be associated with increased risk of NSSI
(Nock & Prinstein, 2005). For example, research suggests that two
PTSD symptoms, reexperiencing the trauma and numbing/
avoidance, mediate the relationship between childhood sexual
trauma and NSSI (Weierich & Nock, 2008). Gratz el al. (2002)
found that sexual abuse was significant in predicting NSSI for
women but not for men after controlling for the effects of other
hypothesized self-injury risk factors. Whitlock et al. (2006) found
that both female and male self-injurers reported having experi-
enced more sexual, physical, and/or emotional abuse than did
non-NSSI individuals. Sex differences in our study may be due to
differences in base rates for various traumas in the lives of women
and men.

The current study has several limitations. The online survey

compliance rate of about 23% (although not uncommon in
Web-based survey research) is much lower than ideal and limits
generalizability. For a study of this type, we believe that a
sample drawn from the entire campus population offers advan-
tages, despite a low return rate, over small samples drawn from
a limited segment of students. Whitlock et al. (2006) appears to
be the only study thus far to sample more than 2,000 students.
The return rate in their study was 37%. Although our sample
represented the general student population well at this univer-
sity, it did contain fewer international students and more mi-
nority students. The study was conducted on only one college
campus, further limiting generalizability. Until a nationwide
epidemiology study can be conducted, results of this study must
join those of other surveys of college students conducted on one
or two campuses.

All data were collected at one time point using self-report

measures. Participants might not have been able to accurately
recall the number and frequency of their self-injurious behaviors;

thus, reporting errors might have confounded the findings. In this
study, the criterion (i.e., frequency of NSSI) was measured with an
initial gateway question, “Have you ever intentionally hurt your-
self?” Only students who answered “yes” were directed to com-
plete the full self-report checklist. More accurate reporting of NSSI
incidence might result from dispensing with the gateway question
and presenting the checklist to every respondent, with the 16 items
preceded by the stem “Have you ever intentionally . . .” Alterna-
tive data reporting methods such as behavioral observation or a
daily log of NSSI may also generate more reliable data. The
cross-sectional nature of this study precludes inferences about
causality. Thus, the temporal relation between factors (e.g., expe-
rience of traumatic events) and NSSI cannot be determined. Lon-
gitudinal research is necessary to confirm the chronological pre-
cedence of mental health correlates, comorbid diagnoses, and
psychological distress to facilitate a better understanding of the
development of NSSI. Some of our measurement instruments
have limited information regarding their psychometric proper-
ties and may pose concerns about whether certain subscales
actually measure what they claim to assess. Archival data
precluded the possibility of including the best available mea-
sures of presenting symptoms and traumatic experiences. For
example, whereas lifetime incidence of NSSI was assessed,
CCAPS instructions asked about symptoms only in the past two
weeks. Finally, we developed the screening inventories primar-
ily for use at college counseling centers, but participants were
drawn from the general student population. It remains uncertain
how well these scales will perform with clients.

Despite these limitations, findings of the current study may have

important clinical implications. Counselors can be in a good po-
sition to detect NSSI and intervene with college students. How-
ever, Whitlock et al. (2006) found that only about 21% of college
students who self-injure had ever disclosed or discussed their
NSSI with a therapist. Results of ROC curve analyses suggest
that for the subscales identified in this study, cutoff scores can
be set to identify one third to one half of the true positive cases
without screening in too many false positives. When used only
to red flag clients, the cost of a false positive screening result is
slight, namely, only that a student who does not engage in NSSI
is nevertheless asked about these behaviors. Regardless of
whether a formal screening tool is used, our findings suggest
that counselors should be alert to the possibility of NSSI in
college students who exhibit a combination of depression, hos-
tility, dissociation and depersonalization, anxiety, isolation,
and— especially in men— outwardly directed anger. Recent re-
search reported that adolescents with NSSI, relative to their
counterparts, exhibited higher physiological reactivity, lower
tolerance for distress, and deficit in problem-solving skills
(Nock & Mendes, 2008). This research suggests possible ave-
nues for intervention. Further research is needed to investigate
the etiology and development of NSSI as well as the temporal
sequence among the correlates and risk factors to advance
prevention and treatment. Integrative models that address how
vulnerabilities at the personal, interpersonal, and environmental
levels interact to influence the phenomenology of NSSI are
especially needed in future research (Prinstein, 2008). More
research is needed to establish psychometric properties of reli-
ability and validity for the screening instruments when used
with actual clients, as well as to further examine the effective-

137

SELF-INJURY SCREENING FOR COLLEGE STUDENTS

ness of the scoring system to identify individuals with NSSI
problems at optimal hit rates.

References

Alexander, N., & Clare, L. (2004). You still feel different: The experience

and meaning of women’s self-injury in the context of a lesbian or
bisexual identity. Journal of Community and Applied Social Psychology,
14, 70 – 84.

American Psychiatric Association. (2000). Diagnostic and statistical man-

ual of mental disorders (4th ed., text rev.). Washington, DC: Author.

Andover, M. S., Pepper, C. M., & Gibb, B. E. (2007). Self-mutilation and

coping strategies in a college sample. Suicide and Life-Threatening
Behavior, 37, 238 –243.

Bennum, I., & Phil, M. (1983). Depression and hostility in self-mutilation.

Suicide and Life-Threatening Behavior, 13, 71– 84.

Brack, G., Brack, C., & Hill-Carlson, M. (1997, August). Trauma associ-

ated spiritual experiences: Theory and application. Paper presented at
the 105th Annual Convention of the American Psychological Associa-
tion, Chicago, IL.

Brown, S. A., Williams, K., & Collins, A. (2007). Past and recent delib-

erate self-harm: Emotion and coping strategy differences. Journal of
Clinical Psychology, 63, 791– 803.

Center for the Study of College Student Mental Health. (2006, November

10). CCAPS integration in Titanium Schedule. Retrieved September 20,
2008, from http://www.sa.psu.edu/caps/research_center_materials.shtml

Chapman, A. L., Gratz, K. L., & Brown, M. Z. (2006). Solving the puzzle

of deliberate self-harm: The experiential avoidance model. Behaviour
Research and Therapy, 44, 371–394.

Claes, L., Vandereycken, W., & Vertommen, H. (2007). Self-injury in

female versus male psychiatric patients: A comparison of characteristics,
psychopathology and aggression regulation. Personality and Individual
Differences, 42, 611– 621.

Clements, R., & Heintz, J. M. (2002). Diagnostic accuracy and factor

structure of the AAS and APS scales of the MMPI–2. Journal of
Personality Assessment, 79, 564 –582.

Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied multiple

regression/correlation analysis for the behavioral sciences (3rd ed.).
Mahwah, NJ: Erlbaum.

Croyle, K., & Waltz, J. ( 2007). Subclinical self-harm: Range of behaviors,

extent, and associated characteristics. American Journal of Orthopsy-
chiatry, 77, 332–342.

Evans, E., Hawton, K., & Rodham, K. (2005). In what ways are adoles-

cents who engage in self-harm or experience thoughts of self-harm
different in terms of help-seeking, communication and coping strategies?
Journal of Adolescence, 28, 573–587.

Favazza, A. R. (1998). The coming of age of self-mutilation. Journal of

Nervous and Mental Disease, 186, 259 –268.

Favazza, A. R. (1999). Self-mutilation. In D. G. Jacobs (Ed.), The Harvard

Medical School guide to suicide assessment and intervention (pp. 125–
145). San Francisco, CA: Jossey-Bass.

Favazza, A. R., DeRosear, L., & Conterio, K. (1989). Self-mutilation and

eating disorders. Suicide and Life-Threatening Behavior, 19, 352–361.

Gallagher, R. P., Zhang, B., & Taylor, R. (2003). National survey of

counseling directors, 2003. Alexandria, VA: International Association
of Counseling Centers.

Gratz, K. L. (2001). Measurement of deliberate self-harm: Preliminary data

on the deliberate self-harm inventory. Journal of Psychopathology and
Behavioral Assessment, 23, 253–263.

Gratz, K. L. (2006). Risk factors for deliberate self-harm among female

college students: The role and interaction of childhood maltreatment,
emotional inexpressivity, and affect intensity/reactivity. American Jour-
nal of Orthopsychiatry, 76, 238 –250.

Gratz, K. L., & Chapman, A. L. (2007). The role of emotional responding

and childhood maltreatment in the development and maintenance of
deliberate self-harm among male undergraduates. Psychology of Men
and Masculinity, 8, 1–14.

Gratz, K. L., Conrad, S. D., & Roemer, L. (2002). Risk factors for

deliberate self-harm among college students. American Journal of Or-
thopsychiatry, 72, 128 –140.

Haines, J., & Williams, C. L. (2003). Coping and problem solving of

self-mutilators. Journal of Clinical Psychology, 59, 1097–1106.

Hawton, K., & Harriss, L. (2007). Deliberate self-harm in young people:

Characteristics and subsequent mortality in a 20-year cohort of patients
presenting to hospital. Journal of Clinical Psychiatry, 68, 1574 –1583.

Hawton, K., Zahl, D., & Weatherall, R. (2003). Suicide following delib-

erate self-harm: Long-term follow-up of patients who presented to a
general hospital. British Journal of Psychiatry, 182, 537–542.

Jacobson, C. M., & Gould, M. (2007). The epidemiology and phenome-

nology of non-suicidal self-injurious behavior among adolescents: A
critical review of the literature. Archives of Suicide Research, 11, 129 –
147.

Kehrberg, C. (1997). Self-mutilating behavior. Journal of Child and Ado-

lescent Psychiatric Nursing, 10, 35– 40.

Klonsky, E. D., & Olino, T. M. (2008). Identifying clinically distinct

subgroups of self-injurers among young adults: A latent class analysis.
Journal of Consulting and Clinical Psychology, 76, 22–27.

Klonsky, E. D., Oltmanns, T. F., & Turkheimer, E. (2003). Deliberate

self-harm in a non-clinical population: Prevalence and psychological
correlates. American Journal of Psychiatry, 160, 1501–1508.

Laye-Gindhu, A., & Schonert-Reichl, K. A. (2005). Nonsuicidal self-harm

among community adolescents: Understanding the “whats” and “whys”
of self-harm. Journal of Youth and Adolescence, 34, 447– 457.

Mehta, C. R., & Patel, N. R. (1997). Exact inference in categorical data.

Biometrics, 53, 112–117.

Muehlenkamp, J. J. (2005). Self-injurious behavior as a separate clinical

syndrome. American Journal of Orthopsychiatry, 75, 324 –333.

Nock, M. K., & Mendes, W. B. (2008). Physiological arousal, distress

tolerance, and social problem-solving deficits among adolescent self-
injurers. Journal of Consulting and Clinical Psychology, 76, 28 –38.

Nock, M. K., & Prinstein, M. J. (2004). A functional approach to the

assessment of self-mutilative behavior. Journal of Consulting and Clin-
ical Psychology, 72, 885– 890.

Nock, M. K., & Prinstein, M. J. (2005). Contextual features and behavioral

functions of self-mutilation among adolescents. Journal of Abnormal
Psychology, 114, 140 –146.

Prinstein, M. J. (2008). Introduction to the special section on suicide and

nonsuicidal self-injury: A review of unique challenges and important
directions for self-injury science. Journal of Consulting and Clinical
Psychology, 76, 1– 8.

Ross, S., & Heath, N. L. (2003). Two models of adolescent self-mutilation.

Suicide and Life-Threatening Behavior, 33, 277–287.

Sevig, T. D., Soet, J. E., Malofeeva, E., & Dowis, J. (2006). Development

and validation of the Counseling Center Assessment of Psychological
Symptoms. Unpublished manuscript, University of Michigan, Ann Ar-
bor, MI.

Skegg, K., Nada-Raja, S., & Moffitt, T. E. (2004). Minor self-harm and

psychiatric disorder: A population-based study. Suicide and Life-
Threatening Behavior, 34, 187–196.

Soet, J. E., & Sevig, T. D. (2006). Counseling Center Assessment of

Psychological Symptoms (CCAPS) manual for administration. Unpub-
lished manual, University of Michigan, Ann Arbor, MI.

Stanley, B., Gameroff, M. J., Michalsen, V., & Mann, J. J. (2001). Are

suicide attempters who self-mutilate a unique population? American
Journal of Psychiatry, 158, 427– 432.

Tabachnick, B. G., & Fidell, L. S. (2007). Using multivariate statistics (5th

ed.). Boston, MA: Pearson Education.

Weierich, M. R., & Nock, M. K. (2008). Posttraumatic stress symptoms

138

CHENG, MALLINCKRODT, SOET, AND SEVIG

mediate the relation between childhood sexual abuse and nonsuicidal
self-injury. Journal of Consulting and Clinical Psychology, 76, 39 – 44.

Whitlock, J., Eckenrode, J., & Silverman, D. (2006). Self-injurious behav-

iors in a college population. Pediatrics, 117, 1939 –1948.

Wright, R. E. (1995). Logistic regression. In L. G. Grimm & P. R. Yarnold

(Eds.), Reading and understanding multivariate statistics (pp. 217–244).
Washington, DC: American Psychological Association.

Yates, T. M. (2004). The developmental psychopathology of self-injurious

behavior: Compensatory regulation in posttraumatic adaptation. Clinical
Psychology Review, 24, 35–74.

Yates, T. M., Tracy, A. J., & Luthar, S. S. (2008). Nonsuicidal self-injury

among “privileged” youths: Longitudinal and cross-sectional approaches
to developmental process. Journal of Consulting and Clinical Psychol-
ogy, 76, 52– 62.

Zlotnick, C., Mattia, J. I., & Zimmerman, M. (1999). Clinical correlates of

self-mutilation in a sample of general psychiatric patients. Journal of
Nervous and Mental Disease, 187, 296 –301.

Received December 4, 2008

Revision received September 9, 2009

Accepted September 21, 2009

䡲

139

SELF-INJURY SCREENING FOR COLLEGE STUDENTS