A Comparison of Linear Versus Non-Linear Models of Aversive

Self-Awareness, Dissociation, and Non-Suicidal Self-Injury

Among Young Adults

Michael F. Armey and Janis H. Crowther

Kent State University

Research has identified a significant increase in both the incidence and prevalence of non-suicidal
self-injury (NSSI). The present study sought to test both linear and non-linear cusp catastrophe models
by using aversive self-awareness, which was operationalized as a composite of aversive self-relevant
affect and cognitions, and dissociation as predictors of NSSI. The cusp catastrophe model evidenced a
better fit to the data, accounting for 6 times the variance (66%) of a linear model (9%–10%). These results
support models of NSSI implicating emotion regulation deficits and experiential avoidance in the
occurrence of NSSI and provide preliminary support for the use of cusp catastrophe models to study
certain types of low base rate psychopathology such as NSSI. These findings suggest novel approaches
to prevention and treatment of NSSI as well.

Keywords: non-suicidal self-injury (NSSI), dissociation, cusp catastrophe, non-linear analysis

Recent research (Favazza, 1998; Sansone & Levitt, 2002) has

identified a significant increase in both the incidence and preva-
lence of non-suicidal self-injury (NSSI). NSSI, defined as the
deliberate destruction or alteration of body tissue without con-
scious suicidal intent that results in significant tissue damage or
scarring, has been frequently identified in college-aged samples
(Gratz, Conrad, & Roemer, 2002). Consistent with research dem-
onstrating associations between emotional reactivity, poor emotion
regulation, and NSSI (e.g., Gratz, 2003; Gratz & Roemer, 2004,
Chapman, Gratz, and Brown (2006) have introduced a model of
NSSI, referred to as “deliberate self-harm” by these researchers,
which proposes that when an event occurs that triggers an un-
wanted emotional response, individuals engage in NSSI to escape
from this unpleasant affective state. While their model focuses
predominantly on NSSI as a behavior of “emotional avoidance,”
they recognize that it also may function to reduce other unpleasant
internal states that may be associated with unwanted emotions.
Thus, individuals may engage in NSSI to escape from the negative
self-appraisal of a personally relevant stimulus or event. Taken
together, an emerging body of literature has suggested that aver-
sive self-awareness, defined here as the experience of aversive and
self-relevant emotions and cognitions, often in response to nega-
tive events, predicts the occurrence/frequency of NSSI (e.g., Chap-
man et al., 2006). Thus, to the extent that NSSI reduces or
eliminates aversive self-awareness, it is negatively reinforced as a
means of managing emotionally charged experiences.

As defined here, aversive self-awareness is composed of both

affective and cognitive elements. Consistent with Watson, Clark,

and Tellegen’s (1988) tripartite model of affect, it seems likely that
aversive self-awareness is characterized by high levels of negative
affect (e.g., shame, sadness, fear, and hostility) and low levels of
positive affect. Negative affect might include not only sadness,
fear, and hostility, but also shame. Shame involves the assumption
that a negative situational outcome is caused by personality char-
acteristics of the individual that are open to criticism by others
(Tangney & Fischer, 1995) and thus may be rich with aversive
self-awareness and self-loathing. The cognitive component of
aversive self-awareness might be conceptualized in terms of
Nolen-Hoeksema’s depressive rumination construct. Nolen-
Hoeksema defines rumination as the process of “focusing pas-
sively and repetitively on one’s symptoms of distress and the
meaning of those symptoms without taking action to correct the
problems one identifies” (Nolen-Hoeksema, 1998, p. 216). Rumi-
native self-awareness has been linked to the onset (Just & Alloy,
1997), exacerbation (Kuehner & Weber, 1999), and increased
chronicity of dysphoric affect and depressed mood (Nolen-
Hoeksema, 2000). Taken together, rumination may function to
exacerbate negative affect in individuals who engage in NSSI.

Psychological dissociation is another construct often examined

within the context of NSSI (Allen, 1995; Gratz et al., 2002;
Herpertz, 1995; Suyemoto, 1998). Psychological dissociation is
defined as the failure to associate thoughts, memories, emotions,
perception of the environment, and personal identity into an inte-
grated whole (American Psychiatric Association, 2000). While the
association between psychological dissociation and NSSI has been
attributed to the high rate of past traumatic life experiences in
individuals who engage in NSSI (Gratz, 2003; Suyemoto, 1998),
the function of psychological dissociation for these individuals is
unclear. For example, some individuals who engage in NSSI
describe dissociation as an experience that occurs before and is
terminated by self-injury (Briere & Gil, 1998), while others de-
scribe it as a consequence of the self-injurious act (Suyemoto,

Michael F. Armey and Janis H. Crowther, Department of Psychology,

Kent State University.

Correspondence concerning this article should be addressed to Michael

F. Armey, Department of Psychology, Kent State University, Kent, OH
44242. E-mail: [e-mail]

Journal of Consulting and Clinical Psychology

Copyright 2008 by the American Psychological Association

2008, Vol. 76, No. 1, 9 –14

0022-006X/08/$12.00

DOI: 10.1037/0022-006X.76.1.9

9

1998). Thus, at this point, the nature of the associations among
aversive self-awareness, dissociation, and NSSI is unclear.

Catastrophe Theory

Previous research has explored the phenomenon of NSSI

through traditional statistical procedures that assume linear rela-
tionships between variables. Although these procedures have pro-
vided a wealth of information about correlates and predictors of
NSSI, it has been argued that complex human behaviors may, at
times, be poorly represented by linear models (Ehlers, 1995). In
contrast to traditional linear models, non-linear dynamic systems
theory (NDST) provides a framework for exploring and evaluating
complex systems and behaviors wherein dependent variables such
as NSSI might be better characterized as dichotomous, present or
absent, state measures (Barton, 1994). NDST seeks to identify the
factors that lead to a shift from one behavioral state (i.e., no
self-injury) to another behavioral state (i.e., self-injury).

Catastrophe theory (Thom, 1972/1975) is one application of

NDST that can be used to describe the factors contributing to the
development of NSSI. Catastrophe theory allows for the modeling
of large, “catastrophic,” non-linear changes in behavior that result
from small changes in continuous predictor variables. This type of
catastrophe model is referred to as the “cusp catastrophe” model
(Gilmore, 1981), since a sudden behavioral change is exhibited
once predictor variables cross the “cusp” threshold. In this study,
we would predict that as aversive self-awareness and dissociation
gradually change, individuals exhibit a sudden shift from a state
absent of self-injury to a self-injuring state.

A predicted behavior must possess five qualities to be consid-

ered within the context of catastrophe theory (Zeeman, 1976). The
first is bimodality, defined as the binary presence or absence of the
behavior of interest. With respect to NSSI, although individuals
may self-injure at different rates, they either engage in NSSI or
they do not. The second quality is inaccessibility, a situation
conceptually related to bimodality in which a particular level of
behavior does not exist as a stable state. Within the cusp catastro-
phe model, the combined influence of predictor variables forces an
intermediate level of a behavior to transform itself into one of two
distinct, and binary, outcomes, in this case, either the absence of
self-injury or infrequent self-injury versus chronic and severe
self-injury. Intermediate values are theoretically impossible or, at
the very least, highly improbable and unstable. Third, NSSI has the
quality of divergence, defined as the tendency for relatively small
changes in predictor variables, that is, aversive self-awareness and
dissociation, to be related to large, state changes in the dependent
variable, that is, NSSI. The fourth quality is hysteresis, or the
tendency for the variables responsible for the occurrence of NSSI
to not be responsible for a return to a state free of NSSI. Finally,
although the last quality, abrupt transitions, is somewhat poorly
defined in NDST (Gilmore, 1981), the theory proposes that the
shift from the absence to the presence of a given behavior, such as
NSSI, occurs with a high probability given a set of predictors.

The Present Study

The major goal of this research was to compare two approaches

to the investigation of aversive self-awareness and dissociation as
predictors of NSSI. Initially, a linear structural equation model was

evaluated to investigate the hypothesis that aversive self-
awareness—which was operationally defined by (a) high levels of
negative affect, including sadness, fear, hostility, and shame; (b)
low levels of positive affect; and (c) high levels of rumination—
and dissociation would predict NSSI, with dissociation mediating
this relationship. Given the non-normal distribution of NSSI, we
then explored the suitability of a non-linear cusp catastrophe
model by using aversive self-awareness, dissociation, and NSSI as
defined above. Here, we hypothesized that subtle changes in
aversive self-awareness and dissociation would result in a dramatic
change from the absence to the presence of NSSI. We hypothe-
sized that a non-linear cusp catastrophe model would better predict
the presence of NSSI than would a linear model since the non-
linear model incorporates more effectively the characteristics of
NSSI as a state variable rather than as a continuous variable. The
optimal statistical approach was determined by comparing the
amount of variance in scores on the Deliberate Self-Harm Inven-
tory (DSHI; Gratz, 2001) explained by both linear and non-linear
models.

Method

Participants

Participants (n

⫽ 225) were male (36.4%) and female (63.6%)

undergraduates (M age

⫽ 19.5, SD ⫽ 3.11) enrolled in general

psychology at a large public university. The ethnic composition
was 85% Caucasian, 10% African American, 2% Asian or Asian
American, 1% Hispanic, 1% of mixed racial heritage, and 1% who
described themselves as Other.

Measures

The Response Styles Questionnaire (Nolen-Hoeksema & Mor-

row, 1991) is a 71-item self-report questionnaire with good con-
struct validity (Nolen-Hoeksema, 2000; Nolen-Hoeksema & Mor-
row, 1991) that assesses an individual’s characteristic tendency to
engage in rumination, distraction, problem solving, or dangerous
behavior when experiencing dysphoria. In this study, the 22-item
Ruminative Responses Subscale (

␣ ⬎ .90; Nolen-Hoeksema &

Morrow, 1991) was scored.

The Positive and Negative Affect Schedule—Expanded Form

(PANAS–X; Watson & Clark, 1994) is a 60-item, self-report
questionnaire that assesses affective states. There is substantial
evidence for the internal consistency and construct validity of the
PANAS and PANAS–X (Watson & Clark, 1994; Watson et al.,
1988). For this study, the Positive Affect (Watson et al., 1988),
Hostility, Sadness, and Fear subscales (

␣s exceeding .82, Watson

& Clark, 1994; Watson et al., 1988) were used. Given our interest
in assessing shame as one component of aversive self-awareness,
a composite SHAME variable was computed by summing the
scores of the “ashamed” and “disgusted with self” items from the
PANAS–X.

The Dissociative Experiences Scale—II (Carlson & Putnam,

1993) consists of 28 questions related to the occurrence of disso-
ciative experiences in everyday life, independent of situations
involving drugs or alcohol. Split-half reliability for the original
Dissociative Experiences Scale is .96 for clinical samples, .71 for

10

ARMEY AND CROWTHER

non-clinical samples, and .95 in college students (Bernstein &
Putnam, 1986).

The DSHI (Gratz, 2001) is a17-item inventory assessing delib-

erate forms of NSSI. The forms of NSSI assessed by the DSHI
(e.g., carving words into the skin) were drawn from clinical ob-
servation and interviews with individuals engaging in these behav-
iors. Individuals respond to each of the 17 questions, reporting
whether or not they have ever engaged in the described behavior.
Scores are summed across items to produce an aggregate measure
of NSSI, with higher scores reflecting greater numbers of endorsed
NSSI behaviors. The DSHI has been shown to identify NSSI in
college samples (Gratz, 2001) and to be significantly associated
with conceptually relevant measures, such as the Difficulties in
Emotion Regulation Scale (Gratz & Roemer, 2004).

Procedure

Following informed consent, participants completed the ques-

tionnaires in small groups. No participants who signed up for the
study refused to participate or chose to discontinue the study. At
the conclusion, participants were verbally debriefed regarding the
true purpose of the study and provided with a listing of campus and
community resources offering psychological counseling or therapy
in case they desired assistance in dealing with dysphoria or NSSI.

Results

Of the initial sample of 225 participants, only 3 were found to

have incomplete data on at least one subscale and were excluded

from subsequent analyses. Bivariate correlations, using a Bonfer-
roni correction of

␣⬘ ⫽ .002 (␣⬘ ⫽ ␣/28) to correct for family-wise

error, were calculated between study variables. Internal consis-
tency for all variables exceeded

␣ ⫽ .70, with the exception of the

two-item SHAME variable, where the two items correlated at r

⫽

.42. Table 1 summarizes descriptive and correlational data.

Consistent with the relatively low base rate of NSSI in the

general population, the DSHI demonstrated positive skew and
kurtosis. Because the DSHI serves as the primary dependent
variable, the positive skew (3.12) exhibited by the DSHI would
likely hinder the fit of a structural model. For this reason, DSHI
scores were logarithmically

1

transformed to minimize the im-

pact of normality violations on linear analyses (Kline, 1998).
Transformed DSHI scores were used for all subsequent linear
models, although skew remained somewhat elevated (1.72).
The remaining variables met the necessary assumptions for
these analyses.

Structural Equation Model and Model Fit

Initial model fit (EQS, Version 6.1; Multivariate Software

Inc., 2003) was assessed with a

␹

2

procedure in which a

non-significant

␹

2

statistic reflects an acceptable model fit

(Kline, 1998). Given the non-normal distribution of DSHI

1

There were no significant improvements in the skew or kurtosis of the

DSHI with the alternate use of either a square-root or natural log transfor-
mation.

Table 1
Bivariate Correlations, Means, Standard Deviations, and Cronbach’s Alpha

Measure

1

2

3

4

5

6

7

8

45.65

1. DES–II

42.40

.97

18.52

2. RRS

.31

*

12.85

.95

31.41

3. PANAS–PA

⫺.16

⫺.35

*

7.52

.89

11.32

4. PANAS–Fear

.30

*

.47

*

⫺.28

*

4.14

.81

11.49

5. PANAS–Hostility

.36

*

.50

*

⫺.37

*

.63

*

4.26

.82

9.72

6. PANAS–Sadness

.29

*

.56

*

⫺.38

*

.62

*

.65

*

4.04

.84

3.41

7. SHAME

.28

*

.44

*

⫺.31

*

.66

*

.70

*

.63

*

1.56

.42

.69

8. DSHI

.17

.34

*

⫺.26

*

.12

.27

*

.19

.19

1.51

.71

Note.

Means, standard deviations, and Cronbach’s alpha on the diagonal, with standard deviations in italics. DES–II

⫽ Dissociative Experiences

Scale—II; RRS

⫽ Ruminative Responses Subscale; PANAS ⫽ Positive Affect Negative Affect Schedule, PA ⫽ Positive Affect Subscale, Fear ⫽ Fear

Subscale, Sadness

⫽ Sadness Subscale, Hostility ⫽ Positive Affect Hostility Subscale; SHAME ⫽ PANAS Composite Shame Variable; DSHI ⫽ Deliberate

Self-Harm Inventory.

*

p

⬍ .002.

11

SPECIAL SECTION; AVERSIVE SELF-AWARENESS, DISSOCIATION, AND SELF-INJURY

scores, the model was evaluated with fit indices derived from
robust variances, an estimation procedure which minimizes the
impact of non-normal data on model fit (Satorra & Bentler,
1994). Because the magnitude of this

␹

2

statistic may be con-

flated with sample size and thus may overstate the poor fit of a
structural model (Bollen, 1989), a number of additional fit
indices were employed, such as the

␹

2

/df statistic, whose ac-

ceptable value would be less than 3 (Kline, 1998). Additionally,
to minimize both Type I and Type II error rates, Hu and Bentler
(1999) recommended that a cutoff of 0.95 or above the com-
parative fit index (CFI; Bentler, 1990) be combined with a
root-mean-square error of approximation (RMSEA) “close to
0.06” (Hu & Bentler, 1999, p. 1) to assess the fit of observed
data to a given model.

An over-identified structural model explored the relationship

between aversive self-awareness and dissociation with NSSI (Fig-
ure 1). Although the assumption of multivariate normality was
violated due to the positive skew of DSHI scores (Mardia’s coef-
ficient

⫽ 22.70), the structural model provided an acceptable fit to

the data, Satorra–Bentler

␹

2

(19)

⫽ 34.42, p ⫽ .02 (␹

2

/df

⫽ 1.81;

CFI

⫽ 0.97; RMSEA ⫽ 0.06; 90% confidence interval (CI) ⫽

0.03, 0.09), but accounted for only 9% of the total variance in
DSHI scores. With respect to indirect effects (Baron & Kenny,
1986; Ullman, 2006), dissociation did not mediate the relationship
between aversive self-awareness and NSSI (standardized coeffi-
cient for indirect effect

⫽ .025, p ⬎ .05).

Cusp Catastrophe Model

The cusp catastrophe model was fit by using Hartelman’s

(1996) Cuspfit program, which permits the assignment of mul-
tiple independent predictor variables and a single dependent
criterion variable. Within the framework of cusp catastrophe
theory, variations in the predictor (or “control”) variables lead
to catastrophic changes in the dependent (or “state”) variable
(Gilmore, 1981). Consequently, although the dependent vari-

able is initially specified as a continuous variable, cusp catas-
trophe theory assumes that the variable is, essentially, bimod-
ally distributed and that the two modes represent distinct
behavioral states.

All variables were standardized to convert measures to a

common metric and to permit the creation of an aggregate
aversive self-awareness variable. Given the planned aggrega-
tion of predictor variables constituting aversive self-awareness
following standardization, the PANAS–Positive Affect was
multiplied by –1 to convert its negative association with the
proposed latent variable of aversive self-awareness to a positive
association. Finally, an aggregate aversive self-awareness vari-
able was computed by summing the standardized versions of the
Ruminative Responses Subscale, PANAS–Fear, PANAS–
Hostility, PANAS–Sadness, PANAS–Positive Affect, and
SHAME subscales. This aversive self-awareness composite
variable and the standardized version of the Dissociative Ex-
periences Scale—II were used as predictor variables for the
cusp catastrophe model. Standardized DSHI scores served as
the state variables.

Model fit for the cusp catastrophe model was assessed by using

a set of indices appropriate for non-normal data. Cuspfit fits linear,
logistic, and cusp models to the data. The appropriateness of a cusp
catastrophe model is evaluated on the basis of its comparison with
both linear and logistic models by using an R

2

value, the Akaike

information criterion (AIC), and the Bayesian information crite-
rion (BIC) statistics (Browne, 2000; Zucchini, 2000). The model
with the highest R

2

and the lowest AIC and BIC values is consid-

ered to provide the best fit to the data.

Results indicated that the cusp model (R

2

⫽ .66; AIC ⫽

415.50; BIC

⫽ 442.70) provided a superior fit to the data than

could either the logistic (R

2

⫽ .10; AIC ⫽ 616.00; BIC ⫽

633.00) or linear models (R

2

⫽ .10; AIC ⫽ 661.30; BIC ⫽

674.90). These results suggest that linear and logistic models
greatly underestimate the fit of the data to the model, account-

.76*

.06*

.39*

.27*

.84*

-.45*

.80*

.62*

DSHI

DES-II

RRS

PANAS Sadness

PANAS PA

PANAS Host.

PANAS Fear

.80*

Aversive

Self-

Awareness

Shame

Figure 1.

Structural model with dissociation mediating the association between aversive self-awareness and

non-suicidal self-injury. * indicates standardized path coefficient statistically significant at p

⬍ .05. RRS ⫽

Ruminative Responses Subscale; PANAS

⫽ Positive Affect Negative Affect Schedule, Sadness ⫽ Sadness

Subscale, PA

⫽ Positive Affect Subscale, Host. ⫽ Hostility Subscale, Fear ⫽ Fear Subscale; Shame ⫽ PANAS

Composite SHAME Variable; DSHI

⫽ Deliberate Self-Harm Inventory; DES–II ⫽ Dissociative Experiences

Scale—II.

12

ARMEY AND CROWTHER

ing for less than one-sixth of the variance explained by a cusp
catastrophe model.

2

Discussion

These results suggest the superiority of a cusp catastrophe

model over a traditional linear model when investigating aversive
self-awareness and dissociation as predictors of NSSI in an under-
graduate sample. Although NSSI has never been observed as a
normally distributed variable in community or college-aged sam-
ples, this is, to our knowledge, the first study to employ non-linear
modeling techniques to the prediction of self-injurious behavior.
Given that a cusp catastrophe model explains a greater proportion
of the variance in NSSI than does a more traditional linear model,
even when the latter used estimation procedures that minimize the
impact of non-normal data on model fit, non-linear modeling
techniques may represent a more appropriate match of the statis-
tical technique to the characteristics of the data in question. De-
spite such a benefit, non-linear modeling techniques have some
limitations. For example, structural equation modeling may be
used to test mediation and moderation, while cusp catastrophe
modeling cannot. Moreover, cusp models should be tested only
when the criterion variable of interest can be conceptualized as a
“state” variable, such as the presence or absence of NSSI, and
possesses the characteristics of bimodality, divergence, hysteresis,
inaccessibility, and abrupt transitions (Gilmore, 1981; Zeeman,
1976). Even so, future research should consider the use of non-
linear modeling to understand the development and occurrence of
other low base rate behaviors, such as suicide, or of diagnoses,
such as PTSD, in order to ultimately identify individuals who may
be at risk for such forms of psychopathology.

Consistent with past theory and research (e.g., Chapman et al.,

2006; Gratz, 2003), findings from both models suggest that indi-
viduals experiencing aversive self-awareness are more susceptible
to NSSI. Interestingly, the results of the structural equation model
may help refine our understanding of dissociation, as it was asso-
ciated with both aversive self-awareness and its constituent ele-
ments but was poorly associated with NSSI and did not emerge as
a mediator. While additional research is needed regarding the
nature and timing of the dissociative experience in relation to
NSSI, it may be that both dissociation and NSSI function to
regulate aversive experiences, but consistent with the experiential
avoidance model described by Chapman et al. (2006), dissociation
and NSSI interact to do so.

Of potential clinical significance is the application of the un-

derlying theoretical assumptions of catastrophe theory to our un-
derstanding of how psychopathology such as NSSI may develop.
Instead of conceptualizing the development of NSSI in terms of a
linear combination of predictors, it may be beneficial to examine
how small changes in multiple predictors interact to render an
individual at risk for engaging in NSSI. Possible future avenues of
clinical research include the development and refinement of inter-
ventions targeted at helping individuals at risk for engaging in
NSSI to remain below the cusp catastrophe threshold, in line with
many of the strategies integral to Dialectical Behavior Therapy
(Linehan, 1993). Moreover, assuming that hysteresis represents a
core feature of NSSI, we might consider the notion that changes in
the factors contributing to the occurrence of NSSI may not be
directly responsible for the termination of NSSI behaviors. Thus,

although increases in negative affect, rumination, and dissociation
may combine to trigger NSSI, the simple reduction of these factors
may be insufficient to provide complete relief from NSSI behav-
ior; this possibility may necessitate the development of additional
interventions, used to augment current treatments, that may not be
directly related to the risk factors involved in the development of
NSSI. However, given the preliminary nature of these findings,
additional research is needed to support these conclusions.

Despite these findings, several factors limit the generalizability

of these results. First, the present study utilized a sample of college
students with limited ethnic diversity. While dissociative experi-
ences (Bernstein & Putnam, 1986) and NSSI (Gratz et al., 2002)
are prevalent in college student samples, these findings should be
considered preliminary until they are replicated in a more diverse
clinical sample of individuals who engage in NSSI. Moreover, this
sample included both individuals who do and do not engage in
NSSI; clearly, further investigations of this statistical model are
warranted in clinical samples in order to more fully understand the
potential clinical utility of this theoretical approach. However, the
observed lifetime prevalence rate of NSSI in our sample (30%) is
comparable with the rates of other studies (37%) using the DSHI
in samples of college students (Gratz, 2006). Second, data collec-
tion relied on a retrospective, cross-sectional, methodology that is
unable to assess changes over time inherent to the underlying
catastrophe theory. Consequently, although the present study ex-
amines a potentially useful statistical technique for the analysis of
these NSSI data, these findings would be significantly strength-
ened through replication in a study using a longitudinal design,
such as ecological momentary assessment. Third, the data were
aggregated prior to the cusp analysis, which may limit our ability
to directly relate those findings to the structural-equation-modeling
(SEM) model; however, it should also be noted that the linear SEM
model fit using EQS accounted for a similar amount of variance
(9%) as did the linear model fit through Cuspfit (10%). While this
study represents an important step toward understanding the nature
of the associations between aversive self-awareness, dissociation,
and NSSI within the context of both linear and non-linear models,
future studies of NSSI would benefit from research methodologies
assessing individuals’ in vivo experiences of emotion, cognition,
and NSSI.

2

Although the linear model in the SEM analysis and the linear model in

the Cuspfit analysis accounted for approximately the same percentage of
variance in DSHI scores (9% and 10%, respectively), it is possible that the
different ways that the data were handled (the Cuspfit analysis used
standardized scores, while the SEM analysis did not) may have impacted
the findings. Thus, we reran the SEM analysis by using standardized scores
on all measures. This reanalysis yielded virtually identical findings,
Satorra–Bentler

␹

2

(19)

⫽ 33.26, p ⫽ .02 (␹

2

/df

⫽ 1.75; CFI ⫽ .97;

RMSEA

⫽ .06; 90% CI ⫽ .02, .09). This model accounted for 8% of the

variance.

References

Allen, C. (1995). Helping with deliberate self-harm: Some practical guide-

lines. Journal of Mental Health, 4, 243–250.

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

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

13

SPECIAL SECTION; AVERSIVE SELF-AWARENESS, DISSOCIATION, AND SELF-INJURY

Baron, R. M., & Kenny, A. (1986). The moderator–mediator variable

distinction in social psychological research: Conceptual, strategic, and
statistical considerations. Journal of Personality and Social Psychology,
51, 1173–1182.

Barton, S. (1994). Chaos, self-organization, and psychology. American

Psychologist, 49, 5–14.

Bentler, P. M. (1990). Comparative fit indexes in structural models. Psy-

chological Bulletin, 107, 238 –246.

Bernstein, E., & Putnam, F. (1986). Development, reliability, and validity

of a dissociation scale. Journal of Nervous and Mental Disease, 174,
727–735.

Bollen, K. A. (1989). Structural equations with latent variables. New

York: Wiley.

Briere, J., & Gil, E. (1998). Self-mutilation in clinical and general popu-

lation samples: Prevalence, correlates, and functions. American Journal
of Orthopsychiatry, 68, 609 – 620.

Browne, M. W. (2000). Cross-validation methods. Journal of Mathemati-

cal Psychology, 44, 108 –132.

Carlson, E. B., & Putnam, W. (1993). An update on the Dissociative

Experiences Scale. Dissociation: Progress in the Dissociative Disor-
ders, 6, 16 –27.

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

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

Ehlers, C. L. (1995). Chaos and complexity: Can it help us understand

mood and behavior? Archives of General Psychiatry, 52, 960 –964.

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

Nervous and Mental Disease, 186, 259 –268.

Gilmore, R. (1981). Catastrophe theory for scientists and engineers. New

York: Wiley.

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

on the Deliberate Self-Harm Inventory. Journal of Psychopathology &
Behavioral Assessment, 23, 253–263.

Gratz, K. L. (2003). Risk factors for and functions of deliberate self-harm:

An empirical and conceptual review. Clinical Psychology: Science and
Practice, 10, 192–205.

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

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

Gratz, K. L., & Roemer, L. (2004). Multidimensional assessment of emo-

tion regulation and dysregulation: Development, factor structure, and
initial validation of the Difficulties in Emotion Regulation scale. Journal
of Psychopathology & Behavioral Assessment, 26, 41–54.

Hartelman, P. (1996). Cuspfit program. [Computer software]. Retrieved

October 24, 2007, from http://users.fmg.uva.nl/hvandermaas/cata.html

Herpertz, S. (1995). Self-injurious behavior: Psychopathological and noso-

logical characteristics in subtypes of self-injurers. Acta Psychiatrica
Scandinavica, 91, 57– 68.

Hu, L., & Bentler, P. M. (1999). Cutoff criteria for fit indices in covariance

structure analysis: Conventional criteria versus new alternatives. Struc-
tural Equation Modeling, 6, 1–55.

Just, N., & Alloy, L. B. (1997). The response styles theory of depression:

Tests and an extension of the theory. Journal of Abnormal Psychology,
106, 221–229.

Kline, R. B. (1998). Principles and practice of structural equation mod-

eling. New York: Guilford Press.

Kuehner, C., & Weber, I. (1999). Responses to depression in unipolar

depressed patients: An investigation of Nolen-Hoeksema’s response
styles theory. Psychological Medicine, 29, 1323–1333.

Linehan, M. (1993). Cognitive-behavioral treatment of borderline person-

ality disorder. New York: Guilford Press.

Multivariate Software Inc. (1998). EQS 6.1 [Computer software]. Encino,

CA: Author.

Nolen-Hoeksema, S. (1998). The other end of the continuum: The costs of

rumination. Psychological Inquiry, 9, 216 –219.

Nolen-Hoeksema, S. (2000). Further evidence for the role of psychosocial

factors in depression chronicity. Clinical Psychology: Science and Prac-
tice, 7, 224 –227.

Nolen-Hoeksema, S., & Morrow, J. (1991). A prospective study of depres-

sion and posttraumatic stress symptoms after a natural disaster: The
1989 Loma Prieta earthquake. Journal of Personality and Social Psy-
chology, 61, 115–121.

Sansone, R. A., & Levitt, J. L. (2002). Self-harm behaviors among those

with eating disorders: An overview. Eating Disorders: The Journal of
Treatment and Prevention, 10, 205–213.

Satorra, A., & Bentler, P. M. (1994). Corrections to test statistics and

standard errors on covariance structure analysis. In A. von Eye & C. C.
Clogg (Eds.), Latent variable analysis (pp. 399 – 419). Thousand Oaks,
CA: Sage.

Suyemoto, K. L. (1998). The functions of self-mutilation. Clinical Psy-

chology Review, 18, 531–554.

Tangney, J. P., & Fischer, K. W. (1995). Self-conscious emotions: The

psychology of shame, guilt, embarrassment, and pride. New York:
Guilford Press.

Thom, R. (1972/1975). Stabilite´ structurelle et morphogene`se [Structural

stability and morphogenesis] (H. Fowler, Trans.). Reading, England:
Benjamin.

Ullman, J. B. (2006). Structural equation modeling. In B. G. Tabachnick &

L. S. Fidell (Eds.), Using multivariate statistics (pp. 607– 675). Boston:
Pearson.

Watson, D., & Clark, L. (1994). The PANAS–X: Manual for the positive

and negative affect schedule—Expanded form. Unpublished manuscript.

Watson, D., Clark, L., & Tellegen, A. (1988). Development and validation

of brief measures of positive and negative affect: The PANAS scales.
Journal of Personality and Social Psychology, 54, 1063–1070.

Zeeman, E. C. (1976). Catastrophe theory. Scientific American, 23, 65– 83.
Zucchini, W. (2000). An introduction to model selection. Journal of

Mathematical Psychology, 44, 41– 61.

Received January 31, 2007

Revision received September 5, 2007

Accepted September 11, 2007

䡲

14

ARMEY AND CROWTHER