Page 1
�������� ����� ��
Temporal Associations between Disordered Eating and Non-suicidal Self-injury: Examining Symptom Overlap over One Year
Brianna J. Turner, Angelina Yiu, Brianne K. Layden, Laurence Claes,Shannon Zaitsoff, Alexander L. Chapman
PII: S0005-7894(14)00109-9DOI: doi: 10.1016/j.beth.2014.09.002Reference: BETH 510
To appear in: Behavior Therapy
Received date: 11 September 2013Accepted date: 5 September 2014
Please cite this article as: Turner, B.J., Yiu, A., Layden, B.K., Claes, L., Zaitsoff, S.& Chapman, A.L., Temporal Associations between Disordered Eating and Non-suicidalSelf-injury: Examining Symptom Overlap over One Year, Behavior Therapy (2014), doi:10.1016/j.beth.2014.09.002
This is a PDF file of an unedited manuscript that has been accepted for publication.As a service to our customers we are providing this early version of the manuscript.The manuscript will undergo copyediting, typesetting, and review of the resulting proofbefore it is published in its final form. Please note that during the production processerrors may be discovered which could affect the content, and all legal disclaimers thatapply to the journal pertain.
Page 2
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 1
Temporal Associations between Disordered Eating and Non-suicidal Self-injury: Examining
Symptom Overlap over One Year
Brianna J. Turner, M.A.1, Angelina Yiu, B.A.1, Brianne K. Layden, M.A.1, Laurence Claes,
Ph.D.2, Shannon Zaitsoff, Ph.D.1, & Alexander L. Chapman, Ph.D, R.Psych1.
1 Simon Fraser University 2 Catholic University of Leuven
Authors’ Note: We would like to thank the Social Sciences and Humanities Research Council for
providing financial support for this study, and the Canadian Institute for Health Research for
providing doctoral funding to the first author.
Page 3
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 2
Abstract
Disordered eating (DE) and non-suicidal self-injury (NSSI) commonly co-occur. This
study compared several models of the longitudinal relationship between DE and NSSI, including
concurrent and prospective models, and examined the possible moderating roles of self-
objectification, impulsivity, and emotion dysregulation in these relationships. Individuals with
NSSI (N = 197) recruited from online forums completed measures of NSSI and DE every three
months for one year. We tested the associations between NSSI and DE using hierarchical linear
models. Results supported a concurrent relationship, wherein frequency of NSSI positively
covaried with concurrent DE severity. Body surveillance moderated the concurrent relationship
between NSSI and DE. Individuals who engaged in more body surveillance endorsed high levels
of DE pathology, whereas those lower in body surveillance engaged in more frequent NSSI only
at higher levels of DE. In addition, whereas DE did not prospectively predict NSSI, frequency of
NSSI predicted more severe DE three months later. The prospective relationship between DE
and later NSSI was moderated by emotion dysregulation, such that highly dysregulated
individuals had a stronger relationship between DE and later NSSI, whereas this relationship was
not significant among individuals low in emotion dysregulation. These findings add valuable
information regarding the co-occurrence of self-damaging behaviors.
Key words: Non-suicidal self-injury; eating disorder; longitudinal; prospective; self-
objectification; emotion dysregulation.
Page 4
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 3
Temporal Associations between Disordered Eating and Non-suicidal Self-injury: Examining
Symptom Overlap over One Year
Self-damaging behavior, defined as any deliberate behavior with a high potential for
physical self-harm, can take a variety of forms. For example, disordered eating (DE) includes a
range of maladaptive or atypical eating and weight control behaviors such as restricting one’s
food intake, binge eating, and compensatory behaviors (e.g., self-induced vomiting, laxative or
diuretic abuse, over-exercising; Stice, Marti, Shaw, & Jaconis, 2009) that often result in tissue
damage and serious health concerns when they are repeated over time. Other self-damaging
behaviors, such as non-suicidal self-injury (NSSI), result in tissue damage as a direct and
immediate consequence of the behavior. Although these behaviors are often painful (Claes,
Vandereycken, & Vertommen, 2006; Selby et al., 2010) and carry a risk of negative emotional
and social consequences (Leibenluft, Gardener, & Cowdry, 1987), they often serve to reduce
unwanted emotions and other internal experiences (Haynos & Fruzzetti, 2011; Heatherton &
Baumeister, 1991; Klonsky, 2007; Smyth et al., 2007). Current theories posit that the immediate
reduction in negative affect provides powerful reinforcement such that these behaviors are
repeated despite negative consequences that can accumulate over the long term (Chapman,
Gratz, & Brown, 2006; Fairburn, Cooper, & Shafran, 2003; Haynos & Fruzzetti, 2011;
Heatherton & Baumeister, 1991). Given their similar features, it is perhaps not surprising that
there is considerable co-occurrence of DE and NSSI. A substantial proportion of individuals with
DE report that they have engaged in NSSI (32-70%; see Svirko & Hawton, 2007, for a review),
and many individuals with NSSI report engaging in DE (as many as 60%; Darche, 1990; Ross,
Heath, & Toste, 2009).
Shared risk factors may also explain the co-occurrence of NSSI and DE. DE and NSSI
have similar correlates, including childhood abuse (Muehlenkamp, Kerr, Bradley, & Adams-
Page 5
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 4
Larson, 2010; Wonderlich et al., 2007), emotion dysregulation (Muehlenkamp, Peat, Claes, &
Smits, 2012), impulsivity (Peterson & Fischer, 2012), negative body regard (Muehlenkamp,
Swanson, & Brausch, 2005; Tylka & Sabik, 2010), and anxiety and mood symptoms (Hudson et
al., 2007; Touchette et al., 2011). Thus, individuals at risk for developing one self-damaging
behavior may be more likely to engage in the other.
Despite growing awareness and concern about the overlap between self-damaging
behaviors, little is known about the temporal relationship between DE and NSSI. Some evidence
suggests that the presence of both DE and NSSI predicts a more chronic or severe course of
behavior over time. For example, among women with bulimia nervosa, engagement in NSSI
predicted continued engagement in binge eating twelve years later (Fichter, Quadflieg, &
Hedlund, 2008). Conversely, although bulimia symptoms were positively related to lifetime
(retrospective) NSSI frequency among university students with a history of NSSI, they did not
prospectively predict NSSI frequency one year later (Glenn & Klonsky, 2011). A recent
investigation of undergraduate women found that whereas NSSI and purging behavior were not
significantly associated at baseline, NSSI predicted greater engagement in purging eight months
later, and vice versa (i.e. purging predicted more frequent NSSI at 8-month follow-up; Peterson
& Fischer, 2012). These findings suggest that NSSI and DE may be reciprocally related over
time. Whereas NSSI seems to be prospectively associated with DE behavior, findings regarding
the prospective association between DE and NSSI have been more mixed. Further, it is unclear
whether NSSI and DE are independently related after accounting for underlying factors
associated with each behavior, such as psychological distress.
Examination of potential moderators of the concurrent and prospective association of
NSSI and DE may help to clarify previous findings and identify factors potentiating the link
between these two self-damaging behaviors. Extant literature has emphasized the possible role of
Page 6
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 5
individual differences in the link between NSSI and DE, including self-objectification
(Muehlenkamp et al., 2005; Tylka & Sabik, 2010), impulsivity (Peterson & Fischer, 2012), and
emotion dysregulation (Muehlenkamp et al., 2012). Each of these factors could be expected to
increase vulnerability for both NSSI and DE. For instance, self-objectification, or the tendency to
view ones’ body as an object rather than part of oneself, may reduce inhibitions toward actions
that may cause physical damage (i.e. people may be harder on objects they do not perceive a
sense of ownership over and connection to), thus increasing the likelihood that individuals would
engage in more than one self-damaging behavior. Difficulty with impulsivity and poor inhibitory
control may make it difficult for people to resist urges to engage in both NSSI and DE. Indeed,
previous research shows that compared to individuals who engage in NSSI alone, those who
engage in both NSSI and DE tend to score higher on each of these factors (Claes et al., 2013;
Muehlenkamp et al., 2012; Petersen & Fischer, 2012; Svirko & Hawton, 2007). Thus, in a
sample of self-injurers, self-objectification, impulsivity, and emotion dysregulation could be
expected to moderate the relationship between NSSI and DE by strengthening the association
between these two behaviors. Whether these factors strengthen the concurrent (e.g., co-variation
between NSSI and DE within a three-month period) or the prospective (e.g., co-variation in
which NSSI predicts more severe DE three months later) relationship between NSSI and DE has
not been addressed in the literature.
Aims and Hypotheses
This study compared several exploratory models of the interplay between NSSI and DE
symptoms over one year in a sample of self-injurers. Given that NSSI and DE are conceptualized
as serving similar emotion regulatory functions (Haynos & Fruzzetti, 2011; Klonsky, 2007), we
expected that these behaviors would positively covary within concurrent time periods. That is,
during times of greater emotional distress we expected that both behaviors would increase,
Page 7
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 6
whereas during times of less emotional distress the behaviors would both decrease (see Booth et
al., 2010). Thus, hypothesis 1a was that DE at TimeT would be positively, concurrently
associated with NSSI at TimeT and would account for unique and significant within-person
variance in this outcome. Similarly, we expected that NSSI at TimeT would be positively,
concurrently associated with DE at TimeT and would account for significant within-person
variance in DE. Given the prominent role that psychological distress is thought to play in each
behavior (Muehlenkamp et al., 2012), hypothesis 1b was that covarying for distress would
diminish the magnitude of the concurrent association between NSSI and DE.
We also examined whether there was a reciprocal, prospective (time-lagged) association
between NSSI and DE, such that DE symptoms at one time point (TimeT) would be positively
associated with frequency of NSSI at subsequent time points (TimeT+1) and vice versa.
Consistent with previous research (Fichter et al., 2008; Peterson & Fischer, 2012), we expected
that NSSI frequency at TimeT would be positively associated with subsequent DE severity at
TimeT+1 (hypothesis 2a). Although findings regarding the prospective relationship between DE
and later NSSI have been mixed (Glenn & Klonsky, 2011; Peterson & Fischer, 2012), we also
expected a positive relationship between NSSI frequency at TimeT and subsequent DE severity
over a three-month interval (at TimeT+1; hypothesis 2b).
The second aim of this study was to clarify individual differences that may moderate the
relationship between NSSI and DE. We hypothesized that self-objectification (hypothesis 3a),
impulsivity (hypothesis 3b), and emotion dysregulation (hypothesis 3c) would moderate the
concurrent relationship between NSSI and DE, such that individuals who score high on these
traits (i.e. one standard deviation above the mean) would exhibit a stronger, more positive
concurrent relationship between the two behaviors compared to those who scored low on these
traits (i.e. one standard deviation below the mean). Similarly, we expected that each of these
Page 8
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 7
traits would moderate the prospective relationship between NSSI and DE (hypotheses 4a, 4b, and
4c, respectively), such that those with high scores would exhibit a stronger prospective
relationship between the two behaviors compared to those with low scores.
Methods
Participants
The sample was composed of 211 individuals who reported at least one instance of NSSI
in their lifetime. Participants were mostly female (n = 197, 93.4%) and Caucasian (n = 193
91.5%). The mean age was 22.94 (SD = 7.15, range = 16 - 57). Most participants resided in the
United States (n = 109, 51.7%), Canada (n = 37, 17.5%), the United Kingdom (n = 26, 12.3%),
and Australia (n = 15, 7.1%), and the remaining participants resided in Europe, Russia, Mexico,
Israel, New Zealand, Japan, and South Africa. Many participants reported an annual household
income of less than $50,000 per year (n = 104, 49.3%) and had completed either high school (n =
73, 34.6%) or some college/university (n = 95, 45.0%).
Procedures
Participants were recruited from online self-injury forums on social networking websites
including Facebook.com, LiveJournal.com, and DailyStrength.org. Advertisements were posted
on the community forums describing the study. Interested participants who contacted the study
coordinator via email received a copy of the informed consent form and a link and password for
the online questionnaire portal. Of the 211 participants who completed the baseline
questionnaires, 197 indicated that they would be interested in completing follow-up
questionnaires every three months for one year. Participants were emailed one week prior to their
follow-up date (e.g., one week before their 3-month follow-up was due), and were then sent
reminder emails roughly once per week. Participants in the first cohort received a $5 gift
certificate for Amazon.com or PayPal.com for each time point completed and a $20 bonus for
Page 9
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 8
completion of all five time-points. The payment scheme was modified part way through the
study to reflect the length of questionnaires, and participants in the second cohort received $10
for each time point and a $25 bonus for completion of all five time-points.
Measures
Non-suicidal Self-injury. The Questionnaire for Non-suicidal Self-injury (QNSSI;
Kleindienst et al., 2008) is a 34-item self-report measure that assesses the frequency, methods,
and functions of NSSI. Originally in German, the measure was translated into English for
research conducted in our laboratory (see Turner, Layden, & Chapman, 2012). The frequency of
NSSI is measured using a single item. After defining NSSI as any behavior that involves
deliberately injuring oneself, the QNSSI asks: “How often have you hurt yourself on average in
the last 3 months?” Responses are scored as follows: 0 = “I haven’t hurt myself in the last 3
months,” 1 = “Once a month or less often,” 2 = “2-3 times per month,” 3 = “1-2 times per week,”
4 = “3-6 times per week,” and 5 = “Daily or more than once a day.” While the psychometric
properties of the functional scales have been evaluated in previous research (Turner et al., 2012),
to our knowledge no published research has examined the properties of the item assessing NSSI
frequency. In the current study, the test-retest reliability of this item was low to moderate over
three-month (Spearman rs = .38 - .59) and six-month (Spearman rs = .32 - .54) intervals. These
rankings, however, may be expected to have relatively low stability, particularly in non-clinical
samples where NSSI frequency may vary considerably over three-month intervals. The
frequency item was correlated with items assessing the absolute frequency (i.e. a count of
number of NSSI episodes) of NSSI over the lifetime (r = .29, p = .002), and with an ordinal scale
of lifetime NSSI frequency (r = .23, p = .001), although these correlations were small. These
small correlations make theoretical sense, however, given the potential changes in such
behaviors over time. Previous research has reported the test-retest reliability of dichotomous
Page 10
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 9
(Gratz, 2001) and interval-level NSSI scales (Fliege et al., 2006; Glenn & Klonsky, 2011; Gratz,
2001), with reliability estimates generally ranging from .49 to .91. To our knowledge, previous
studies have not reported reliability of ordinal-type scales assessing NSSI frequency specifically.
Although we would expect the stability of NSSI frequency in the past three months to be lower
than the stability of dichotomous and count data regarding past behavior, the possibility of
psychometric limitations with regard to this item cannot be ruled out.
Eating Disorder Symptoms. The Eating Disorder Diagnostic Scale (EDDS; Stice,
Telch, & Rizvi, 2000) is a 22-item measure assessing symptoms of eating disorders as indicated
by the DSM-IV-TR (APA, 2000). Items from the EDDS assess DE-related attitudes and
behaviors over a range of time periods consistent with DSM-IV criteria (e.g., fear of weight gain
during the past three months; average weekly frequency of fasting, excessive exercise, self-
induced vomiting, and laxative/diuretic use during the past three months; average weekly
frequency of binge eating during the past six months). We used a composite, total score to index
overall DE severity, computed by summing the items in their original metric (Stice, Fisher, &
Martinez, 2004). Scores on this measure ranged from 0 to 93. Previous research demonstrates
that the composite score has good internal consistency (α = .89), test-retest reliability over one
week (r = .89), and convergence with other validated self-report and interview-based measures
of DE symptoms (Stice et al., 2000). In this study, the internal consistency of the composite
score was good at each time point (αs = .81 - .87). The EDDS can also be used to screen for
probable eating disorder diagnoses. Diagnoses derived from the EDDS demonstrated good
agreement with interview-based diagnoses (average κ = .83; Stice et al., 2000), and acceptable
sensitivity and specificity (> .77; Stice et al., 2004).
Self-objectification. The Objectified Body Consciousness Scale (OBCS; McKinley &
Hyde, 1996) is a 24-item self-report measure that includes three subscales: body surveillance
Page 11
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 10
(engagement in appearance-related body monitoring, and placing a high value on how the body
looks as opposed to how the body feels), body shame (experiencing shame when the participant
perceives that his or her body does not conform to social standards), and appearance control
beliefs (beliefs that the participant can control his or her weight with sufficient effort, as opposed
to believing that weight is determined by heritable or uncontrollable factors). We used the three
subscale scores to measure aspects of self-objectification; scores on each subscale were
computed as the mean of at least six of the eight items (McKinley & Hyde, 1996). Scores on
each scale range from 1 to 7. The OBCS subscales have acceptable internal consistency (αs = .68
- .89 in undergraduate students, αs = .70 - .76 in middle-aged women), good test-retest reliability
over two weeks (r = .73 - .79), and converge with related constructs, including body
consciousness, body esteem, and eating disorder symptoms (McKinley & Hyde, 1996). In this
study, the OBCS subscales demonstrated good internal consistency (αs = .81 - .83).
Impulsivity. The Barratt Impulsiveness Scale – 11 (BIS-11; Patton, Stanford, & Barratt,
1995) is a 30-item self-report measure that assesses features of impulsivity. For the purposes of
this study, the total score was used to index overall impulsivity. Total scores range from 30 to
120, and are calculated by summing the ratings on each item; higher scores reflect greater
impulsivity. The total score of the BIS-11 demonstrates good internal consistency (α = .83), test-
retest reliability over one month (r = .83), convergence with clinical problems related to
impulsivity (Stanford et al., 2009), and had good internal consistency in this study (α = .86).
Emotion Dysregulation. The Difficulties in Emotion Regulation Scale (DERS; Gratz &
Roemer, 2004) is a 36-item self-report measure. For this study, we used the total score
(calculated by summing all 36 items) to index overall emotion dysregulation, with higher scores
indicating greater dysregulation. Total scores range from 36 to 180. The DERS total score has
demonstrated excellent internal consistency (α = .93), test-retest reliability over four to eight
Page 12
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 11
weeks (ρ = .88), and convergence with other measures of emotion regulation and experiential
avoidance (Gratz & Roemer, 2004). In this study, the total score demonstrated excellent internal
consistency (α = .91).
Psychological Distress. We used the General Severity Index (GSI) of the Brief Symptom
Inventory (Derogatis, 1993) to assess psychological distress at each time point. The GSI is a
weighted sum of nine symptom dimensions (somatization, obsessive-compulsive, interpersonal
sensitivity, depression, anxiety, hostility, phobic anxiety, paranoid ideation, and psychoticism),
and is used to index overall symptom severity or psychological distress. Scores on the GSI range
from 0 to 212. The GSI has demonstrated excellent internal consistency (α > .90; see Pereda,
Forns, & Pero, 2007 for a review), test-retest reliability over two weeks (r = .90; Derogatis,
1993), and convergence with measures of symptom severity (Derogatis, 1993) and distressing
life events (e.g., unemployment, divorce; Gilbrar & Ben-Zur, 2002). The GSI had excellent
internal consistency in this study (α = .96).
Planned Analyses
We used hierarchical linear modelling (HLM) to test the hypotheses using HLM 7.0
software (Bryk, Raudenbush, & Congdon, 2010). HLM has several advantages over traditional
repeated measures approaches for longitudinal data, including explicitly modelling within- and
between-person variability, and allowing participants with missing data to contribute to
parameter estimates using maximum likelihood estimation (Black, Harel, & Matthews, 2011;
Little & Rubin, 1987). We used random effects models to allow between-person variability in
slopes and intercepts. All variables were converted to z-scores prior to entry into the HLM
models. Level 1 predictors were group-mean centered and Level 2 predictors were grand-mean
centered to facilitate the interpretability of the coefficients and reduce collinearity. Time was
uncentered and modeled as 0 at the intercept, and then the number of days from baseline to
Page 13
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 12
completion of each follow-up questionnaire. Following the recommendations of Singer and
Willett (2003), hypotheses were tested using a series of increasingly complex models, beginning
with unconditional means (Model 1) and growth (Model 2) models, shown below, and ending
with the moderation models.
Model 1
Level 1 Yti = π0i + eti
Level 2 π0i = γ 00 + r0i
Model 2
Level 1 Yti = π0i + π1i(TimeT) + eti
Level 2 π0i = γ00 + r0i
π1i = γ 10 + r1i
Hypothesis 1a was examined by modelling the predictor (e.g., DE symptoms at TimeT) as
a Level 1 time-varying predictor of the outcome (e.g., NSSI at TimeT), with TimeT as a Level 1
covariate, as shown in the following example equation. Participant age was included as a Level 2
covariate in concurrent and prospective models as age was associated with study drop-out, and
therefore was associated with missing data in both concurrent and prospective models (see
Missing Data and Attrition, below).
Level 1 NSSIT = π0i + π1i(TimeT) + π2i(DET) + eti
Level 2 π0i = γ 00 + γ 01(Age) + r0i
π1i = γ 10 + γ 11 + r1i
π2i = γ 20 + γ 21 + r2i
Hypothesis 1b was examined by adding psychological distressT as a Level 1 covariate to
the above models. The proportion of additional variance explained by adding distress as a
Page 14
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 13
covariate was examined using guidelines published by Scientific Software International, Inc.,
wherein the reduction in Level 1 residual variance is modeled as a proportion of the Level 1
residual variance in the model without the covariate.
We next examined the prospective (time-lagged) multilevel models predicting outcomes
over the follow-up period (TimeT + 1) from predictors assessed at prior time points (TimeT).
Prospective analyses examined associations across four lags: from T1 to T2, T2 to T3, T3 to T4,
and from T4 to T5. For hypotheses 2a and 2b, the time-lagged outcome (either DE or NSSI at
TimeT+1) was modelled as a function of lagged time (TimeT +1), the outcome variable at TimeT (to
account for the autocorrelation between the outcome at TimeT and TimeT+1), and the predictor at
TimeT, with participant age as a Level 2 covariate. The effect of Time T + 1 was fixed in all
prospective models. The following is an example of the equation predicting DET +1 from NSSIT
with Age as a Level 2 covariate:
Level 1 DET +1 = π0i + π1i(TimeT +1) + π2i(DET) + π3i(NSSIT) + eti
Level 2 π0i = γ 00 + γ 01(Age) + r0i
π1i = γ 10 + γ 11 + r1i
π2i = γ 20 + γ 21 + r2i
π3i = γ 30 + γ 31 + r3i
To investigate hypotheses 3 and 4, each of the moderators of interest were added as Level
2 moderators of the relevant Level 1 intercepts and slopes in separate models. In these models,
we were primarily interested in whether the cross-level interaction between the Level 2
moderator and the Level 1 slope was significant. Significant interactions at Level 2 were
clarified by examining the simple slopes and regions of significance (using utilities provided by
Sibley, 2008, and Preacher, Curran, & Bauer, 2006).
Page 15
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 14
Results
Sample Characteristics
Participants reported clinically significant levels of NSSI, including engaging in NSSI 2-
3 times per month on average and using multiple methods of NSSI (M = 5.06, SD = 2.51), such
as cutting (95.7%), hitting (56.4%), and scratching until bleeding occurred (56.4%). The majority
of participants (68.2%) had received medical attention for their NSSI. With regard to DE, 14.2%
of participants met the EDDS cut-offs for a DSM-IV eating disorder, and an additional 11.2%
met sub-threshold criteria. Highlighting the clinical severity of this sample, nearly half of
participants (49.2%) reported at least one previous suicide attempt, 53.6% had received
psychotherapy within the past year, and 16.1% had been admitted to a psychiatric unit within the
past year. Please see Table 1 for further details of the NSSI and DE behaviors endorsed by the
participants.
Missing Data and Attrition
Of the 197 individuals who consented to participate in the longitudinal phase of the
study, eight participants provided email addresses that were invalid by the time the first follow-
up was due. Of the eligible participants, 119 (60%) completed at least one follow-up assessment
(see Table 2 for sample sizes at each time point). Participants completed an average of 2.41 (SD
= 1.53) assessments in this study; 15.8% (n = 34) completed all five assessments, 11.6% (n = 25)
completed four assessments, and 12.1% (n = 26) completed three assessments. Further inspection
of missing data suggested that complete data were available for 62.27% of the values of interest.
We used Little’s Missing Completely at Random (MCAR) test (1988) to examine
patterns of missing data in the demographic variables (i.e. age and sex) and primary variables of
interest (i.e. NSSI, DE, OBCS, BIS, DERS, and GSI). Results supported the assumption that the
data were missing at random (χ2[709] = 764.10, p = .07). Further, comparing participants who
Page 16
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 15
provided baseline data only to those who provided at least some follow-up data revealed no
significant differences with respect to sex (χ2[1] = .93, p =.34), DE severity, frequency of NSSI,
self-objectification, impulsivity, emotion dysregulation, or psychological distress (ts = -0.58 -
0.94, ps > .10). Participants who did not provide follow-up data were younger (t[208] = -4.06, p
< .001) than those who did.1 Simulation studies have demonstrated that when data are missing at
random or completely at random, statistical techniques based on maximum likelihood estimation,
such as HLM, produce unbiased parameter estimates (Black, Harel, & Matthews, 2011). This is
particularly true when variables that are associated with missingness are included as covariates;
thus, we included participant age as a covariate in our concurrent and prospective models.
Descriptive and Preliminary Data Analysis
We examined the correlations between NSSI frequency and DE severity at each time
point, and their associations with each of the moderators of interest (see Table 2). NSSI and DE
had small, positive relationships with one another at each time point (rs = .17 - .31). Further,
whereas body shame had medium correlations with DE (rs = .40 - .58), the other moderators had
small or inconsistent correlations with DE and NSSI (rs = -.16 - .33).
To clarify the within- and between-person variability in the outcomes of interest, we
fitted an unconditional means model (Model 1) to each of the outcomes and calculated the intra-
class correlation coefficients (ICC). The average initial NSSI score was 1.84, indicating that the
average participant in this study engaged in NSSI 2-3 times per month at the beginning of the
study. The ICC revealed that 51.41% of the variability in NSSI was found within participants.
The unconditional growth model (Model 2) revealed that on average, NSSI frequency decreased
slightly over the course of the study (γ = -.001, p = .003).
With respect to DE severity, the unconditional means model (Model 1) revealed that the
average initial DE score was 30.93 (range = 0 - 93), indicating that the average participant in this
Page 17
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 16
study endorsed moderate symptoms of DE at the beginning of the study. Most variability in DE
was between participants (72.13%). The unconditional growth model (Model 2) revealed that on
average, DE severity decreased over the course of the study (γ = -.001, p < .001).
Concurrent Models
Consistent with hypothesis 1a, DE symptoms significantly and positively co-varied with
NSSI at any given time (γ = .29, p = .004, see Table 3). In other words, an increase in NSSI
frequency within a given three-month period was associated with an increase in DE symptom
severity. Consistent with hypothesis 1b, the magnitude of the effect diminished slightly after
controlling for psychological distress (γ = .22, p = .02), although the relationship between DE
and NSSI remained significant. DE explained 15.94% of the within-person variability in NSSI
(beyond what was explained by the effects of time and age), and adding psychological distress
accounted for an additional 3.63% of the within-person variability in NSSI.
Also consistent with hypotheses 1a and 1b, when we examined DE as the outcome, we
found that NSSI was positively, concurrently associated with DE (γ = .15, p = .003, see Table 3),
even after controlling for psychological distress (γ = .10, p = .04). NSSI frequency explained
28.11% of the within-person variability in DE (beyond what was explained by time and age), and
psychological distress accounted for a further 14.39% of the within-person variability in DE.
Prospective Models
Consistent with hypothesis 2a, DE symptoms at TimeT did not predict subsequent NSSI at
TimeT+1 (γ = .12, p = .37, see Table 3). Contrary to hypothesis 2b, however, NSSI prospectively
predicted DE (γ = .12, p = .009, see Table 3), indicating that more frequent NSSI at one time
point was associated with greater DE severity three months later.
Moderators of the Concurrent NSSI-DE Relationship
Page 18
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 17
Partially supporting hypothesis 3a, body surveillance significantly moderated the
concurrent relationship between DE symptoms and NSSI frequency (γ = -.28, p = .004, see Table
4), whereas body shame (γ = .06, p = .43) and appearance control beliefs (γ = -.05, p = .58) did
not. Simple slope analyses revealed a stronger, more positive association between DE and NSSI
at lower values of body surveillance compared to higher values (see Figure 1). Among
participants who scored one standard deviation below the mean on body surveillance, there was a
positive association between NSSI and DE (γ = .55, SE = .16, t = 3.57, p < .001), whereas among
those who scored one standard deviation above the mean, the relationship between NSSI and DE
was not significant (γ = -.01, SE = .12, t = -.10, p = .92). Similarly, only body surveillance
significantly interacted with NSSI frequency when predicting concurrent DE symptom severity
(surveillance γ = -.14, p < .001; shame γ = -.04, p = .57; control γ = .04, p = .52; see Table 4).
There was a stronger relationship between NSSI and DE among participants who scored one
standard deviation below the mean on body surveillance (γ = .24, SE = .04, t = 5.49, p < .001),
whereas the relationship between NSSI and DE was not significant among those who scored one
standard deviation above the mean (γ = -.05, SE = .07, t = -.63, p = .53). As shown in Figure 1,
participants who scored high on body surveillance exhibited more severe DE symptoms than
those who scored low on body surveillance, regardless of NSSI frequency. Although body shame
did not moderate the association between NSSI and DE, the addition of body shame as a Level 2
covariate explained an additional 44.09% of the between-person variance in DE (see also Table 2
for correlations). Further, models that included body shame improved model fit (deviance =
795.94) compared to models that included other moderators (deviances = 842.34 - 909.58).
Contrary to hypothesis 3b, impulsivity did not moderate the concurrent relationship
between DE and NSSI (γ = -.06, p = .33), or between NSSI and DE (γ = -.05, p = .35, see Table
4). Inconsistent with hypothesis 3c, there was no significant moderating effect of emotion
Page 19
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 18
dysregulation for DE predicting concurrent NSSI (γ = .002, p = .98) or NSSI predicting
concurrent DE (γ = -.04, p = .49, see Table 4).2 Further, models that included impulsivity or
emotion dysregulation explained minimal additional between-person variance in the outcomes
(1.7% - 6.5%), did not explain additional within-person variance in the outcomes, and resulted in
worse fit compared to the models without these moderators.
Moderators of the Prospective NSSI-DE Relationship
Although none of the self-objectification variables (hypothesis 4a; γs = -.09 - .18, ps >
.07) or impulsivity (hypothesis 4b; γ = -.02, p = .83) moderated the prospective relationship
between DET and later NSSIT+1, consistent with hypothesis 4c, emotion dysregulation moderated
this prospective relationship (γ = .27, p < .001; see Table 4). None of the variables of interest
moderated the prospective relationship between NSSIT and later DET+1 (γV = -.08 - .08, ps >
.10; see Table 4). Consistent with our expectations, simple slopes analyses revealed a stronger,
positive relationship between DE and later NSSI at high levels of emotion dysregulation (see
Figure 1). Among participants who scored one standard deviation below the mean on emotion
dysregulation, there was a negative and non-significant association between DE and later NSSIs
(γ = -.16, SE = .14, t = -1.14, p = .26), whereas among those who scored one standard deviation
above the mean the relationship between DE and later NSSI was positive and significant (γ = .38,
SE = .13, t = 2.98, p = .004).
Discussion
This study is one of the first empirical investigations of the temporal relationships
between DE and NSSI in a targeted sample of individuals with NSSI, and thus addresses an
important gap in the literature. Although previous clinical observations suggest that NSSI and
DE may be negatively related (e.g., Washburn, Gebhardt, Styer, Juzwin, & Gottlieb, 2012), with
one behavior increasing as the other decreases, our results demonstrate that NSSI and DE shared
Page 20
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 19
a positive concurrent relationship within three-month intervals in an untreated sample. That is, at
any given time point an increase in NSSI frequency was associated with more severe DE
behavior, and vice versa. Consistent with research underscoring the importance of psychological
distress in accounting for fluctuations in self-damaging behaviors (Booth et al., 2010), the
strength of the relationship between NSSI and DE diminished when distress was taken into
account; however, it remained significant, suggesting that there is a relationship between NSSI
and DE that is independent of psychological distress.
With respect to moderators of the concurrent relationship between NSSI and DE, our
results highlight the importance of self-objectification, and related constructs such as body
surveillance and shame, for understanding the relationship between NSSI and DE. Specifically,
results suggested that self-injurers who engage in high levels of body surveillance exhibited
more severe DE behaviors, regardless of frequency of NSSI. Among those low in body
surveillance, however, more severe DE was observed among those who were engaging in
frequent NSSI, whereas those engaging in less frequent NSSI endorsed less severe DE.
Similarly, when NSSI was examined as the outcome, results suggested that frequent NSSI
occurred mainly among those who were low in body surveillance but who were engaging in
severe DE. Together, these findings suggest that individuals who are low in body surveillance
but who are nonetheless engaging in severe DE may represent a particularly high-risk group who
are likely to rely on multiple self-damaging behaviors, possibly prompting more frequent NSSI
in this population. In reconciling this finding with previous literature that generally points to high
self-objectification as being a risk factor for NSSI (Muehlenkamp, 2012), it is important to note
that this study focused on the maintenance of NSSI among individuals with an established
history of this behavior, whereas much of the previous work has focused on initiation or presence
of NSSI (Muehlenkamp, 2012). It is possible that whereas greater body surveillance might
Page 21
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 20
contribute to the initial acts of NSSI, higher body surveillance may in some cases protect against
more frequent NSSI once the behavior has been established. Body surveillance may bring
attention to the negative physical consequences of NSSI, such as scarring and infection.
Alternatively, individuals who attend less to how their body looks and more to how it feels (i.e.
those who are low in surveillance) may be more aware of internal cues for distress, resulting in
greater urges for self-damaging behaviors. Future research should investigate specific
mechanisms (e.g., concern about scarring, interoceptive awareness) that may explain this pattern,
as well as individual differences that can account for variability in the developmental trajectories
following initiation of NSSI.
The results of this study also underscored the important role of body shame in accounting
for between-person variability in DE severity among those who engage in NSSI. Our results
suggest that body shame may function as an individual difference factor that distinguishes
individuals with NSSI who do versus do not engage in DE, with body shame accounting for 44%
of the between-person variability in DE severity. This finding is in line with a robust body of
literature suggesting that body shame and other negative attitudes toward the body are key risk
factors for DE (Muehlenkamp et al., 2005; Tylka & Sabik, 2010). It also underscores the need to
attend to body-related attitudes to further our understanding of why some individuals may
engage in multiple self-damaging behaviors, whereas others “specialize” in a single behavior.
In addition to examining concurrent associations between NSSI and DE, this study
investigated prospective relationships between NSSI and DE over three-month lagged intervals.
Consistent with other studies (Fichter et al., 2008; Peterson & Fischer, 2012), our results
demonstrated a prospective relationship between NSSI and later DE, such that greater frequency
of NSSI predicted more severe DE three months later. One possible explanation for this finding
is that the negative emotions that are often triggered by NSSI over the long term (Leibenluft et
Page 22
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 21
al., 1987) may result in an increased need for alternative emotion regulation behaviors, such as
DE. Alternatively, it may be that NSSI becomes decreasingly effective at reducing negative
emotions over time, prompting engagement in other maladaptive behaviors. Indeed, given that
NSSI is prospectively related to suicidal behavior (Asarnow et al., 2011; Whitlock et al., 2013),
some researchers have posited that NSSI in particular may function as a “gateway” to more
versatile self-damaging behavior (Whitlock et al., 2013). Consistent with this notion, Joiner’s
(2005) interpersonal theory of suicide suggests that experience harming oneself may
progressively diminish inhibitions against doing so while providing experience with the
potentially reinforcing qualities of such behaviors, thereby increasing an individual’s willingness
to try other methods. This emerging body of literature suggesting that NSSI may increase risk for
and severity of other self-damaging behaviors underscores the clinical importance of identifying
and reducing this behavior.
Perhaps helping to understand why previous findings regarding the prospective
relationship from DE to later NSSI have been inconsistent (Glenn & Klonsky, 2011; Peterson &
Fischer, 2012), our results suggested that DE predicted later NSSI only among those who
reported high levels of emotion dysregulation. This finding is consistent with previous work
suggesting that emotion dysregulation may function as an underlying vulnerability for using
multiple self-damaging behaviors (Muehlenkamp et al., 2012). The combination of severe DE
(which itself may be associated with physical, nutritional, and cognitive consequences that
increase dysregulation) and emotion dysregulation may strengthen the prospective relationship
between DE and later NSSI. On its own, however, DE may be less likely to prospectively predict
engagement in other self-damaging behaviors such as NSSI in those who are not vulnerable to
emotion dysregulation. Indeed, among those who are low in emotion dysregulation, engagement
Page 23
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 22
in severe DE was associated with less frequent NSSI, possibly suggesting that a single self-
damaging behavior may suffice to regulate emotions in this population.
Contrary to our expectations, this study did not find a moderating role of impulsivity in
the concurrent or prospective relationships between NSSI and DE. In addition, self-reported
impulsivity demonstrated inconsistent zero-order associations with NSSI and DE across time
points. This unexpected finding may be understood in light of research suggesting that the
strength of the relationship between NSSI and impulsivity differs depending on how impulsivity
is assessed (e.g., behavioral tasks versus self-report; Glenn & Klonsky, 2010; Janis & Nock,
2009). Previous research also suggests that some aspects of impulsivity, such as negative
urgency (i.e. hasty decision making in the presence of intense negative emotions), are important
in accounting for NSSI, whereas other aspects of impulsivity (e.g., lack of perseverance, lack of
premeditation, sensation seeking) are less consistently related to NSSI (Glenn & Klonsky, 2010;
Peterson & Fischer, 2012). The relationship between impulsivity and DE is similarly nuanced.
Whereas urgency and sensation seeking are prominent among individuals with bulimia nervosa
(BN), individuals with restrictive anorexia nervosa have more difficulty with premeditation and
perseverance (Claes, Vandereycken, & Vertommen, 2005). Impulsivity does not differentiate
individuals with BN who engage in multiple self-damaging behaviors, including NSSI, from
those who do not (Newton, Freeman, & Munro, 1993). Understanding the interplay between
impulsivity, NSSI, and DE may require multimodal and more comprehensive assessments of
impulsivity.
We believe these findings may have implications for clinical work with populations who
engage in multiple self-damaging behaviors. These findings underscore the importance of
assessing both NSSI and DE when working with patients who engage in self-damaging behavior,
given the positive temporal relationships between these behaviors. Fewer than 50% of treatment
Page 24
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 23
providers assess for NSSI in ED patients (Peebles, Wilson, & Lock, 2011) and, to our
knowledge, few studies have examined rates of assessment for ED in patients with NSSI.
Routine and repeated use of structured assessments of NSSI and DE can help therapists identify
and target important maladaptive behaviors, and to track how they covary over time. Future work
examining the interplay between these behaviors among individuals receiving psychotherapy is
necessary before conclusions can be drawn about implications for treatment.
Despite the strengths and novelty of this study, several limitations warrant consideration.
First, although three-month intervals provide insight into the broad strokes of the relationship
between NSSI and DE, future research should examine these associations over other intervals of
time (e.g., hours and days, or years). For example, by examining these constructs weekly, we
might find that individuals are more likely to engage in NSSI one week and DE the next week,
providing evidence for a “symptom-swapping” profile that was not observable across three-
month increments. It also is possible that within a shorter timeframe, the relationship among
behaviors may differ. Second, the frequency of NSSI was assessed using a single item. Single-
item assessments often suffer from limited reliability, and the psychometric properties of this
item in the current sample were marginal. It is unclear from this study whether the marginal
performance of this item is related to the qualities of the behavior it assesses (e.g., that NSSI
frequency is not particularly stable over three-month intervals, and may not be correlated with
lifetime frequency counts) or with psychometric limitations of the item. Future investigations
should use multi-item measures assessing NSSI frequency to provide more specific and reliable
information about NSSI. Third, the use of a self-selected sample of individuals who use online
self-injury forums limits the generalizability of our findings. Such individuals may be younger,
more technologically savvy, and more willing to disclose their NSSI and DE behavior compared
to individuals who do not use such forums. Future research examining the relationship between
Page 25
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 24
NSSI and DE in clinical and community samples will clarify the generalizability of our findings.
Fourth, although the analytic methods we used can flexibly handle instances of missing data, the
pattern of findings may have differed if we had had higher participant retention across
assessment points, and thus replication in larger samples is warranted. Fifth, the frequencies of
behaviors in the current study were aggregated across three-month intervals by asking
participants how often on average per week they engaged in NSSI and DE, which may restrict
the ranges of these variables. Despite these limitations, we believe this study provides an
important first step toward understanding the temporal associations between two clinically
relevant behaviors, NSSI and DE, as they unfold in an untreated community sample. Although
the present study does not speak directly to clinical assessment or treatment, future research in
this vein could illuminate important avenues and opportunities for intervention.
Page 26
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 25
References
Asarnow, J. R., Porta, G., Spirito, A., Emslie, G., Clarke, G., Wagner, K.D., … & Brent, D. A.
(2011). Suicide attempts and non-suicidal self-injury in the treatment of resistant
depression in adolescents: findings from the TORDIA study. Journal of the American
Academy of Child & Adolescent Psychiatry, 50(8), 772-81.
Black, A. C., Harel, O., & Matthews, G. (2012). Techniques for analyzing intensive longitudinal
data with missing values. In M. R. Mehl & T. S. Conner, Handbook of Research Methods
for Studying Daily Life, p. 339-356. New York, NY: The Guilford Press.
Booth, B. M., Curran, G., Han, X., Wright, P., Frith, S., Leukefeld, C., Falck, R., & Carlson, R.
G. (2010). Longitudinal relationship between psychological distress and multiple
substance use: results from a three-year multisite natural-history study of rural stimulant
users. Journal of Studies on Alcohol & Drugs, 71(2), 258-67. PMID: 20230724.
Bryk, A.S, Raudenbush, S.W., & Congdon, R. (2010). HLM 7 for Windows [Computer
software]. Chicago, IL: Scientific Software International, Inc.
Chapman, A. L., Gratz, K. L., & Brown, M. (2006). Solving the puzzle of deliberate self-injury:
The experiential avoidance model. Behavior Research & Therapy, 44, 371-394.
Claes, L., Fernández-Aranda, F., Jimenez-Murcia, S., Botella, C., Casanueva, F. F., de la Torre,
R., Menchón, J. M. (2013). Co-occurrence of non-suicidal self-injury and impulsivity in
extreme weight conditions. Personality and Individual Differences, 54(1), 137–140.
doi:10.1016/j.paid.2012.07.035
Claes, L., Vandereycken, W., & Vertommen, H. (2006). Pain experience related to self-injury in
eating disorder patients. Eating Behaviors, 7, 204-213.
Page 27
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 26
Claes, L., Vandereycken, W., & Vertommen, H. (2005). Impulsivity-related traits in eating
disorder patients. Personality and Individual Differences, 39(4), 739-749.
http://dx.doi.org/10.1016/j.paid.2005.02.022
Darche, M. A. (1990). Psychological factors differentiating self-mutilating and non-self-
mutilating adolescent inpatient females. Psychiatric Hospital, 21(1), 31-35.
Derogatis, L. R. (1993). Brief Symptom Inventory: Administration, scoring, and procedures
manual (4rd ed.). Minneapolis, MN: National Computer Systems.
Fairburn, C.G., Cooper, Z., Shafran, R., 2003. Cognitive behavior therapy for eating disorders: A
“transdiagnostic” theory and treatment. Behavior Research and Therapy, 41, 509–528.
Fliege, H., Kocalevent, R-D., Walter, O. B., Beck, S., Gratz, K. L., Gutierrez, P. M., & Klapp,
B. F. (2006). Three assessment tools for deliberate self-harm and suicide behavior:
Evaluation and psychopathological correlates. Journal of Psychosomatic Research, 61(1),
113-121. doi.org/10.1016/j.jpsychores.2005.10.006
Fichter, M. M., Quadflieg, N., & Hedlund, S. (2008). Long-term course of binge eating disorder
and bulimia nervosa: relevance for nosology and diagnostic criteria. The International
Journal of Eating Disorders, 41(7), 577–86. doi:10.1002/eat.20539
Gilbrar, O., & Ben-Zur, H. (2002). Adult israeli community norms for the brief symptom
inventory (BSI). International Journal of Stress Management, 9(1), 1-10.
doi:10.1023/A:1013097816238
Glenn, C. R., & Klonsky, E. D. (2010). A multimethod analysis of impulsivity in non-suicidal
self-injury. Personality Disorders: Theory, Research, and Treatment, 1, 67–75.
Glenn, C. R., & Klonsky, E. D. (2011). Prospective prediction of non-suicidal self-injury: a 1-
year longitudinal study in young adults. Behavior Therapy, 42(4), 751–
62.doi:10.1016/j.beth.2011.04.005
Page 28
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 27
Gratz, K. L., & Roemer, L. (2004). Multidimensional assessment of emotion regulation and
dysregulation: Development, factor structure, and initial validation of the difficulties in
emotion regulation scale. Journal of Psychopathology and Behavioral Assessment, 26(1),
41-54. doi: 10.1023/B:JOBA.0000007455.08539.94
Gratz, K. L., & Roemer, L. (2008). The relationship between emotion dysregulation and
deliberate self-harm among female undergraduate students at an urban commuter
university. Cognitive and Behavior Therapy, 37, 14–25.
Haynos, A. F., & Fruzzetti, A. E. (2011). Anorexia nervosa as a disorder of emotion
dysregulation: Theory, evidence, and treatment implications. Clinical Psychology:
Science and Practice, 18(3), 183–202.
Heatherton, T. F., & Baumeister, R. F. (1991). Binge eating as escape from self-awareness.
Psychological Bulletin, 110(1), 86-108.
Hudson, J. I., Hiripi, E., Pope, H. G., & Kessler, R. C. (2007). The prevalence and correlates of
eating disorders in the National Comorbidity Survey Replication. Biological psychiatry,
61(3), 348–58. doi:10.1016/j.biopsych.2006.03.040
Janis, I. B., & Nock, M. K. (2009). Are self-injurers impulsive?: Results from two behavioral
laboratory studies. Psychiatry Research, 169(3), 261-267. doi:
10.1016/j.psychres.2008.06.041
Joiner, T. E. (2005). Why people die by suicide. Cambridge, MA: Harvard University Press.
Kleindienst, N., Bohus, M., Ludäscher, P., Limberger, M. F., Kuenkele, K., Ebner-Priemer, U., .
…Schmahl, C. (2008). Motives for non-suicidal self-injury among women with
borderline personality disorder. Journal of Nervous and Mental Disease, 196(3), 230-
236. doi: 10.1097/NMD.0b013e3181663026
Page 29
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 28
Klonsky, E. D. (2007). The functions of deliberate self-injury: A review of the evidence. Clinical
Psychology Review, 27(2), 226-239.
Kumar, G., Pepe, D., & Steer, R. (2004). Adolescent psychiatric inpatients' self-reported reasons
for cutting themselves. Journal of Nervous and Mental Disease, 192(12), 830-836.
Leibenluft, E., Gardner, D., & Cowdry, R. (1987). The inner experience of the borderline self-
mutilator. Journal of Personality Disorders, 1, 317–324.
Little, R. J. A. (1988). A test for missing completely at random for multivariate data with
missing values. Journal of the American Statistical Association, 88, 1198-1202.
McKinley, N. M., & Hyde, J. S. (1996). The Objectified Body Consciousness Scale:
Development and validation. Psychology of Women Quarterly, 20(2), 181–215.
Muehlenkamp, J. J. (2012). Body Regard in non-suicidal self-injury: Theoretical explanations
and treatment directions. Journal of Cognitive Psychotherapy, 26(4), 331–347.
Muehlenkamp, J. J., Kerr, P. L., Bradley, A., & Adams-Larson, M. (2010). Abuse subtypes and
non-suicidal self-injury: Preliminary evidence of complex emotion regulation patterns.
Journal of Nervous and Mental Disease, 198, 258–263.
Muehlenkamp, J. J., Peat, C. M., Claes, L., & Smits, D. (2012). Self-injury and disordered
eating: expressing emotion dysregulation through the body. Suicide & Life-threatening
Behavior, 42(4), 416–25. doi:10.1111/j.1943-278X.2012.00100.x
Muehlenkamp, J. J., Swanson, J. D., & Brausch, A. M. (2005). Self-Objectification, risk taking,
and self-harm in college women. Psychology of Women Quarterly, 29, 24–32.
Newton, J. R., Freeman, C. P., & Munro, J. (1993). Impulsivity and dyscontrol in bulimia
nervosa: Is impulsivity an independent phenomenon or a marker of severity? Acta
Psychiatrica Scandinavica, 87(6), 389-394. doi:10.1111/j.1600-0447.1993.tb03393.x
Page 30
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 29
Overton, A., Selway, S., Strongman, K., & Houston, M. (2005). Eating disorders – the regulation
of positive as well as negative emotion experience. Journal of Clinical Psychology in
Medical Settings, 12(1), 39-56.
Patton, J. H., Stanford, M. S., & Barratt, E. S. (1995). Factor structure of the barratt
impulsiveness scale. Journal of Clinical Psychology, 51(6), 768-774. doi: 10.1002/1097-
4679(199511)51:6<768::AID-JCLP2270510607>3.0.CO;2-1
Peebles, R., Wilson, J. L., & Lock, J. D. (2011). Self-injury in adolescents with eating disorders:
Correlates and provider bias. Journal of Adolescent Health, 48, 310-313.
doi:10.1016/j.jadohealth.2010.06.017
Pereda, N., Forns, M., & Peró, M. (2007). Dimensional structure of the Brief Symptom
Inventory with Spanish college students. Psicothema 19, 634- 639.
Perez, J., Venta, A., Garnaat, S., & Sharp, C. (2012). The difficulties in emotion regulation scale:
Factor structure and association with non-suicidal self-injury in adolescent inpatients.
Journal of Psychopathology and Behavioral Assessment, 34(3), 393-404. doi:
10.1007/s10862-012-9292-7
Peterson, C. M., & Fischer, S. (2012). A prospective study of the influence of the UPPS model of
impulsivity on the co-occurrence of bulimic symptoms and non-suicidal self-injury.
Eating Behaviors, 13(4), 335–41. doi:10.1016/j.eatbeh.2012.05.007
Preacher, K. J., Curran, P. J., & Bauer, D. J. (2006). Computational tools for probing interaction
effects in multiple linear regression, multilevel modeling, and latent curve
analysis. Journal of Educational and Behavioral Statistics, 31, 437-448.
Ross, S., Heath, N. L., & Toste, J. R. (2009). Non-suicidal self-injury and eating pathology in
high school students. The American journal of orthopsychiatry, 79(1), 83–92.
doi:10.1037/a0014826
Page 31
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 30
Selby, E. As, Smith, A. R., Bulik, C. M., Olmsted, M. P., Thornton, L., McFarlane, T. L., …
Joiner, T. E. (2010). Habitual starvation and provocative behaviors: Two potential routes
to extreme suicidal behavior in anorexia nervosa. Behavior research and therapy, 48(7),
634–45. doi:10.1016/j.brat.2010.03.016
Sibley, C. G. (2008). Utilities for examining interactions in multiple regression [computer
software]. University of Auckland.
Smyth, J. M., Wonderlich, S. a, Heron, K. E., Sliwinski, M. J., Crosby, R. D., Mitchell, J. E., &
Engel, S. G. (2007). Daily and momentary mood and stress are associated with binge
eating and vomiting in bulimia nervosa patients in the natural environment. Journal of
Consulting and Clinical Psychology, 75(4), 629–38. doi:10.1037/0022-006X.75.4.629
Stanford, M. S., Mathias, C. W., Dougherty, D. M., Lake, S. L., Anderson, N. E., & Patton, J. H.
(2009). Fifty years of the Barratt Impulsiveness Scale: An update and review. Personality
and Individual Differences, 47(5), 385-395. doi: 10.1016/j.paid.2009.04.008
Stice, E., Marti, C., Shaw, H., & Jaconis, M. (2009). An 8-year longitudinal study of the natural
history of threshold, subthreshold, and partial eating disorders from a community sample
of adolescents. Journal of Abnormal Psychology, 118, 587–597
Stice, E., Fisher, M., & Martinez, E. (2004). Eating disorder diagnostic scale: Additional
evidence of reliability and validity. Psychological Assessment, 16(1), 60-71. doi:
10.1037/1040-3590.16.1.60
Stice, E., Telch, C. F., & Rizvi, S. L. (2000). Development and validation of the eating disorder
diagnostic scale: A brief self-report measure of anorexia, bulimia, and binge-eating
disorder. Psychological Assessment, 12(2), 123-131. doi: 10.1037/1040-3590.12.2.123
Page 32
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 31
Svirko, E., & Hawton, K. (2007). Self-injurious behavior and eating disorders: The extent and
nature of the association. Suicide & Life-threatening Behavior, 37(4), 409–
21.doi:10.1521/suli.2007.37.4.409
Touchette, E., Henegar, A., Godart, N. T., Pryor, L., Falissard, B., Tremblay, R. E., & Côté, S.
M. (2011). Subclinical eating disorders and their comorbidity with mood and anxiety
disorders in adolescent girls. Psychiatry Research, 185(1-2), 185-192. doi:
10.1016/j.psychres.2010.04.005
Turner, B. J., Layden, B. K, & Chapman, A. L. (2012). Intrapersonal and interpersonal functions
of non-suicidal self-injury: Associations with emotional and social functioning. Suicide
and Life-threatening Behavior, 42, 36-55.
Tylka, T. L., & Sabik, N. J. (2010). Integrating social comparison theory and self-esteem within
objectification theory to predict women’s disordered eating. Sex Roles, 63(1-2), 18–31.
doi:10.1007/s11199-010-9785-3
Washburn, J. J., Gebhardt, M., Styer, D. M., Juzwin, K. R., & Gottlieb, L. (2012). Co-occurring
disorders in the treatment of non-suicidal self-injury: An evidence-informed approach.
Journal of Cognitive Psychotherapy: An International Quarterly, 26(4), 348-364.
doi:10.1891/0889-8391.26.4.348.
Whitlock, J., Muehlenkamp, J., Eckenrode, J., Purington, A., Baral Abrams, G., Barreira, P., &
Kress, V. (2013). Non-suicidal self-injury as a gateway to suicide in young
adults. Journal of Adolescent Health, 52(4), 486-492.
doi:10.1016/j.jadohealth.2012.09.010
Wonderlich, S. A., Rosenfeldt, S., Crosby, R. D., Mitchel, J. E., Engel, S. C., Smyth, J., et al.
(2007). The effects of childhood trauma on daily mood lability and comorbid
psychopathology in bulimia nervosa. Journal of Traumatic Stress, 20, 7–87.
Page 33
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 32
Footnotes
1 We compared participants who did and did not complete each time point, and obtained similar
results: only age consistently differed between participants who did or did not complete any given time
point. The exception was that participants who did not complete the second time point were more
impulsive (t[207] = 2.35, p =.02) and psychiatrically severe (t[209] = 2.07, p =.04) than those who did.
These differences were not found for any other time point. Similar results were also obtained using a
logistic regression to predict drop-out status: only age uniquely predicted drop-out status (Model : χ2[9] =
16.56, Nagelkerke R2 = .29, p =.05, Age: OR = 1.14, 95% CI of OR = 1.02 - 1.29, p = .02). Thus, age was
retained as a covariate in concurrent and prospective models.
2 We also examined each of the three secondary BIS subscales, and the six DERS subscales to
examine whether there were significant moderating effects for any specific aspects of impulsivity or
emotion dysregulation, but did not find any significant interaction effects (γs =-.001 - .07, ps = .10 - .99).
More details on these results are available from the first author.
3 Given the small number of males in the sample, we repeated all of our analyses with the females
only. The pattern of results remained consistent, and thus analyses for the full sample are presented.
Page 34
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 33
Table 1: NSSI, Disordered Eating and Eating Disorder Diagnoses in the Present Sample
NSSI Characteristics at Baseline
Mean (SD)
NSSI Number of Methods 5.06 (2.51)
NSSI Age of Onset 13.14 (4.86)
Rates of Disordered Eating at Baseline
Percentage (n)
Binge Eating 31.8% (67)
Weekly Compensatory Behaviors 36.5% (77)
Vomiting 13.7% (29)
Laxative/Diuretic Misuse 8.1% (17)
Fasting 31.8% (67) Excessive Exercise 18.5% (39)
Full and Subthreshold Eating Disorders at Baseline
Percentage (n)
Anorexia Nervosa (AN) 0% (0)
Bulimia Nervosa (BN) 10.9% (23)
Binge Eating Disorder (BED) 3.3% (7)
Subthreshold AN 3.8% (6)
Subthreshold BN 6.7% (12)
Subthreshold BED 0.9% (2)
NSSI and DE Over the Course of One Year
Min Max Mean SD Skew Kurtosis
T1 NSSI 0 5 1.99 1.49 0.29 -0.92
T2 NSSI 0 5 1.72 1.28 0.44 -0.23
T3 NSSI 0 5 1.69 1.32 0.71 -0.28
T4 NSSI 0 5 1.70 1.35 0.73 -0.09
T5 NSSI 0 5 1.53 1.43 0.78 -0.22
T1 EDDS 0 80 32.22 17.74 0.10 -0.32
T2 EDDS 0 93 31.31 17.99 0.42 0.50
T3 EDDS 0 74 28.01 15.34 0.03 0.09
T4 EDDS 0 77 27.67 16.37 0.31 0.11
T5 EDDS 0 58 28.18 14.66 -0.01 -0.73
Note. NSSI: QNSSI frequency; EDDS: Eating disorder diagnostic scale composite score.
Baseline data on the EDDS were only available for 119 participants, so frequencies are reported
for this subsample.
Page 35
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 34
Table 2: Correlations Among Moderators and Predictors of Interest within each time point
Time 1
N = 211
Time 2
n = 95
Time 3
n = 79
Time 4
n = 69
Time 5
n = 54 Intercorrelations
NSSI DE NSSI DE NSSI DE NSSI DE NSSI DE Surveil Shame Control BIS DERS
Body Surveillance .13 .34 .02 .28 -.06 .41 .19 .24 .08 .28 - .47 .25 .03 .14
Body Shame .18 .57 -.06 .42 -.04 .58 .06 .49 -.003 .40 - .05 .02 .37
Body Control .05 .18 .03 .24 .12 .24 -.10 .19 .09 .19 - -.10 -.08
BIS-11 Total .02 .19 .03 .10 -.13 -.02 .10 .09 .04 .24 - .33
DERS Total .19 .21 .01 .33 -.11 -.09 -.16 .26 .02 .19 -
NSSI Frequency - .17 - .17 - .31 - .23 - .24 - - - - -
Note. BIS-11 Total: total score from the Barratt Impulsiveness Scale; DERS total: total score from the Difficulty in Emotion Regulation
Scale.
Page 36
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 35
Table 3: HLM Models Investigating the Concurrent and Prospective Relationships
Predicting Concurrent NSSI T Predicting Concurrent DE T
Fixed Effects Fixed effects
Coefficient
Estimate (SE) t (df) p
Coefficient
Estimate (SE) t (df) p
Intercept .03 (.08) .35 (162) .73 Intercept .15 (.08) 1.79 (162) .08
Age * Int -.02 (.007) -2.31 (162) .02 Age * Int .01 (.009) 1.31 (162) .19
Time T -.0003 (.0003) -1.15 (163) .25 Time T -.0007 (.0002) -3.09 (163) .002
DE T .29 (.10) 2.89 (163) .004 NSSI T .15 (.05) 3.03 (163) .003
Predicting Prospective NSSI T+1 Predicting Prospective DE T+1
Fixed Effects Fixed effects
Coefficient
Estimate (SE) t (df) p
Coefficient
Estimate (SE) t (df) p
Intercept .01 (.14) .11 (85) .92 Intercept .16 (.13) 1.23 (87) .22
Age * Int -.01 (.007) -1.80 (85) .08 Age * Int .02 (.01) 1.68 (87) .10
Time T+1 -.0003 (.0005) -.59 (209) .56 Time T+1 -.002 (.0004) -4.16 (214) <.001
NSSI T -.25 (.07) -3.43 (86) <.001 DE T -.32 (.07) -4.48 (88) <.001
DE T .12 (.13) .90 (86) .37 NSSI T .12 (.05) 2.67 (88) .009
Note. All fixed effects are presented with robust standard errors. Age * Int = covariate for participant age.
Page 37
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 36
Table 4: Moderating effects of Self-Objectification, Impulsivity and Emotion Dysregulation
Predicting Concurrent NSSI Predicting Concurrent DE
Coefficient
Estimate (SE) t (df) p
Coefficient
Estimate (SE) t (df) p
Surveil (γ02) .04 (.07) .52 (155) .60 Surveil (γ02) .34 (.07) 5.17 (155) <.001
Surveil * DE (γ31) -.28 (.10) -2.90 (156) .004 Surveil * NSSI (γ31) -.14 (.04) -3.46 (156) <.001
Shame (γ02) .06 (.08) .78 (147) .43 Shame (γ02) .54 (.06) 8.69 (147) <.001
Shame * DE (γ31) -.05 (.09) -.63 (148) .53 Shame * NSSI (γ31) -.04 (.07) -.58 (148) .57
Control (γ02) .03 (.07) .48 (148) .63 Control (γ02) .21 (.07) 2.84 (148) .005
Control * DE (γ31) -.05 (.10) -.56 (149) .58 Control * NSSI (γ31) .04 (.05) .64 (149) .52
Impulsivity (γ02) .08 (.06) 1.33 (160) .19 Impulsivity (γ02) .13 (.07) 1.80 (160) .07
Impulse * DE (γ31) -.06 (.06) -.98 (161) .33 Impulse * NSSI (γ31) -.05 (.05) -.93 (161) .35
Emo Dysreg. (γ02) .12 (.07) 1.79 (160) .08 Emo Dysreg. (γ02) .24 (.07) 3.38 (160) <.001
EmoDys * DE (γ31) .002 (.08) .03 (161) .98 EmoDys * NSSI (γ31) -.04 (.06) -.69 (161) .49
Predicting Prospective NSSI Predicting Prospective DE
Coefficient
Estimate (SE) t (df) p
Coefficient Estimate (SE)
t (df) p
Surveil (γ02) -.05 (.10) -.54 (80) .59 Surveil (γ02) .28 (.08) 3.42 (82) <.001
Surveil * DE (γ31) .01 (.10) .11 (81) .91 Surveil * NSSI (γ31) .08 (.06) 1.42 (83) .16
Shame (γ02) -.08 (.11) -.73 (79) .47 Shame (γ02) .47 (.08) 5.94 (81) <.001
Shame * DE (γ31) .18 (.10) 1.86 (80) .07 Shame * NSSI (γ31) .04 (.06) .72 (82) .47
Control (γ02) -.009 (.09) -.10 (77) .92 Control (γ02) .13 (.10) 1.37 (77) .17
Control * DE (γ31) -.09 (.13) -.67 (78) .50 Control * NSSI (γ31) .04 (.05) .95 (78) .35
Impulsivity (γ02) -.03 (.08) -.43 (84) .67 Impulsivity (γ02) .12 (.10) 1.19 (86) .24
Impulse * DE (γ31) -.02 (.12) -.21 (85) .83 Impulse * NSSI (γ31) -.007 (.06) -.11 (87) .91
Emo Dysreg. (γ02) -.10 (.10) -1.08 (84) .29 Emo Dysreg. (γ02) .22 (.10) 2.20 (86) .03
EmoDys * DE (γ31) .27 (.08) 3.43 (85) <.001 EmoDys * NSSI (γ31) -.08 (.05) -1.55 (87) .13
Note. All fixed effects are presented with robust standard errors. Surveil = OBCS body surveillance subscale; Shame = OBCS body shame subscale; Control =
OBCS appearance control beliefs subscale; Impulsivity = BIS-11 total score; EmoDysreg = DERS total score.
Page 38
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 37
Figure 1: Surveillance Moderates the Concurrent Relationships between DE and NSSI.
-0.60
-0.40
-0.20
0.00
0.20
0.40
0.60
0.80
Low DE (-1 SD) High DE (+1 SD)
NS
SI
Freq
uen
cy
(z s
cores)
Low Surveillance (-1 SD)
High Surveillance (+1 SD)
-0.60
-0.40
-0.20
0.00
0.20
0.40
0.60
0.80
Low NSSI (-1 SD) High NSSI (+1 SD)
DE
Sy
mp
tom
Sev
erit
y (
z s
co
res)
Low Surveillance (-1 SD)
High Surveillance (+1 SD)
Page 39
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 38
Figure 2: Emotion Dysregulation Moderates the Prospective Relationship between DE and Later NSSI.
-0.60
-0.50
-0.40
-0.30
-0.20
-0.10
0.00
0.10
0.20
0.30
0.40
Low DE (-1 SD) High DE (+1 SD)
NS
SI
Freq
uen
cy
(z s
co
res)
Low Emotion Dysregulation (-1 SD)
High Emotion Dysregulation (+1 SD)
Page 40
ACC
EPTE
D M
ANU
SCR
IPT
ACCEPTED MANUSCRIPT
Longitudinal Relationship of DE and NSSI 39
Highlights
• NSSI and DE were assessed every three months for one year.
• NSSI and DE are positively, contemporaneously related within three-month intervals.
• Frequency of NSSI predicted more severe DE three months later.
• The contemporaneous association of NSSI and DE was moderated by body surveillance.
• The prospective relationship of NSSI and DE was moderated by emotion dysregulation.