PFI_12mmX178mm.

pdf + eps format

Journal of Child Psychology and Psychiatry 60:10 (2019), pp 1076–1084 doi:10.1111/jcpp.13072

Does higher-than-usual stress predict nonsuicidal self-

injury? Evidence from two prospective studies in
adolescent and emerging adult females
Adam Bryant Miller,1 Tory Eisenlohr-Moul,2 Catherine R. Glenn,3 Brianna J. Turner,4
Alexander L. Chapman,5 Matthew K. Nock,6 and Mitchell J. Prinstein1
1
Department of Psychology and Neuroscience, University of North Carolina at Chapel Hill, Chapel Hill, NC;
2
Department of Psychiatry, University of Illinois at Chicago, Chicago, IL; 3University of Rochester, Rochester, NY,
USA; 4Department of Psychology, University of Victoria, Victoria, BC; 5Department of Psychology, Simon Fraser
University, Burnaby, BC, Canada; 6Department of Psychology, Harvard University, Cambridge, MA, USA

Background: Nonsuicidal self-injury (NSSI) is highly prevalent among adolescent and emerging adult females. Most
studies examining the relationship between stress and NSSI largely have relied on aggregate self-report measures of
stress and between-person models. Using data from two prospective samples, this manuscript tests the hypothesis
that within-person models of NSSI provide better clinical markers of risk for NSSI than between-person models of
NSSI. Methods: Two samples (Sample 1: 220 high-risk girls, M age = 14.68, SD = 1.36, baseline assessment and 3-
month follow-ups for 18 months; Sample 2: 40 emerging adult females with a history of NSSI, M age = 21.55,
SD = 2.14, 14 days with daily retrospective reports) were followed prospectively and completed validated measures of
stress and NSSI. Models were adjusted for age and depression. Results: In Sample 1, a within-person model
demonstrated that higher-than-usual (but not average) stress levels predicted NSSI within the same 3-month wave.
In Sample 2, results from a within-person model with daily diary assessment data showed that higher-than-usual
stress (but not average daily stress) predicted same-day NSSI. Conclusions: Together, our results suggest that
higher-than-usual stress, relative to one’s typical stress level, but not average stress levels, signals times of enhanced
risk for NSSI. These results highlight the clinical utility of repeated assessments of stress. Keywords: Nonsuicidal
self-injury; stress; within-person designs.

reinforced for NSSI through both relief of distress

Introduction
and unwanted interpersonal demands, as well as
Nonsuicidal self-injury (NSSI) is the direct, inten-
through increases in desirable emotional or cognitive
tional harm to one’s own body performed without the
states and social support following NSSI. Addition-
intent to die. NSSI is a robust predictor of future
ally, theories of NSSI strongly emphasize the emo-
suicidal behavior (Ribeiro et al., 2016). Rates of NSSI
tion-regulation function (Chapman, Gratz, & Brown,
increase sharply in adolescence with a slight decline
2006; Nock & Prinstein, 2004), positing that NSSI
during the late adolescent and emerging adulthood
provides an effective, albeit maladaptive, means of
years (Plener, Schumacher, Munz, & Groschwitz,
reducing distress among youth who experience high
2015), and rates of NSSI are higher in females
stress reactivity and difficulty regulating distress.
relative to males (Fox et al., 2015). Adolescent NSSI
From a developmental perspective, the period from
is associated with a fourfold increase in the odds of
adolescence into emerging adulthood is character-
making a suicide attempt by emerging adulthood
ized by both the highest risk for NSSI (Plener et al.,
(Scott, Pilkonis, Hipwell, Keenan, & Stepp, 2015).
2015) and significant increases in stress (Rudolph,
Together, the period from adolescence through
2002). Thus, stress may be particularly relevant to
emerging adulthood represents the developmental
understanding NSSI during this developmental per-
period of highest risk for NSSI. In this multistudy,
iod. Most prior research tests stress as a between-
prospective investigation, we examined whether ado-
person risk factor for NSSI (Fox et al., 2015; Liu,
lescents and emerging adults are at risk for NSSI
Cheek, & Nestor, 2016), and prevailing models and
when they experience periods of higher-than-usual
extant research have not examined within-person
stress.
dynamics of stress and NSSI.
A core feature across NSSI theories is that youth
Most previous research has conceptualized stress
engage in NSSI as a self-regulatory mechanism to
as a static, between-person NSSI risk factor (Liu
relieve distress secondary to stress, defined in the
et al., 2016). Most frequently, studies utilize
present manuscript as actual stressors or perceived
research designs, assessment strategies, and statis-
stress. For example, the Four-Function Model of
tical indices that are best suited to test whether
NSSI (Nock & Prinstein, 2004) posits that youth are
individuals with high stress (actual or perceived) are
at greater risk for NSSI, compared to others in the
sample. Although helpful for identifying population-
Conflict of interest statement: No conflicts declared.

© 2019 Association for Child and Adolescent Mental Health

Published by John Wiley & Sons Ltd, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main St, Malden, MA 02148, USA
doi:10.1111/jcpp.13072 Within-person stress and NSSI 1077

level risk factors, it is an ecological fallacy to assume (Study 1 and 2) over and above individuals’ prior
that between-person risk factors translate to within- NSSI and between-person stress severity. We adjust
person risk. For example, the clinical utility of for the effects of age and depression, given that these
knowing stress is associated with risk for NSSI is factors all are consistently associated with risk for
less helpful for a clinician treating a caseload of NSSI (Fox et al., 2015). Procedures for all studies
chronically stressed adolescents. Increasingly, were approved by human subject review boards at
researchers have advocated for adopting within- the respective institutions.
person approaches to understanding the relation-
ship between stress and NSSI (Hamza & Willoughby,
2014). A handful of studies using ecological momen-
Study 1
tary assessment (EMA) methodologies with adults
Methods
have provided insight into within-person associa-
tions between affective states and NSSI (Rodr
ıguez- Participants and procedures. Participants were
Blanco, Carballo, & Baca-Garc
ıa, 2018), but these 220 adolescent females (M age = 14.69 years,
studies have not yet examined risk for NSSI in the SD = 1.37). Participants were part of an 18-month
context of within-person fluctuations in stress multiwave study examining stress and trajectories of
among adolescents and emerging adults. self-injurious thoughts and behaviors. We have previ-
No two individuals experience stress in exactly the ously examined different predictors of suicidal idea-
same way. The stress-sensitization framework of tion and behaviors in this sample (for details,
depression (Hammen, Henry, & Daley, 2000) asserts including rates of suicidal ideation and behavior see:
individuals have different thresholds for stress that, Eisenlohr-Moul et al., 2018; Miller et al., 2017). Here,
when exceeded, lead to depressive episodes. A sim- we examine NSSI as an outcome. Participants were
ilar process may occur with NSSI. Two individuals, recruited from local inpatient units (33%), outpatient
facing similar actual or perceived stress, may have facilities and practices (12%), local advertisements
very different thresholds that must be crossed before (40%), and mass emails to university employees
each engages in NSSI. Consistent with this sensiti- (15%). Inclusion/exclusion criteria were assessed via
zation framework, we expect that individuals may be caregiver report during a phone screening. Inclusion
at risk for NSSI when they experience higher-than- criteria were as follows: female sex, 12–16 years old,
usual (relative to their own average) stress, regard- caregiver (parent or legal guardian) willing to take part
less of their level of stress compared to others. Data in the study, and a history of at least one mental
from studies with typically developing adolescents health concern in the prior 2 years. Mental health
have begun to demonstrate that within-person vari- concerns (e.g. mood, anxiety disorders) were identified
ation in stress is associated with lower positive mood via caregivers’ reports of previous mental health
and greater negative mood (Bai & Repetti, 2018; treatment, diagnosis, or symptomatology. Exclusion
Timmons & Margolin, 2015), greater systemic criteria were as follows: active psychosis, intellectual
inflammation (Fuligni et al., 2009), and altered disability disorder, and any other developmental dis-
HPA axis functioning (Lippold, Davis, McHale, Bux- order identified via caregiver report on the phone
ton, & Almeida, 2016). Together, these studies screen. Diagnostic presentation at baseline, as
provide compelling evidence that higher-than-usual assessed by caregiver report on the Behavioral Assess-
stress affects both psychological and biological pro- ment System for Children, is presented in Table 1.
cesses. Thus, higher-than-usual stress is a promis- After consenting/assenting at baseline, adoles-
ing candidate predictor of NSSI among adolescents cents and caregivers separately completed clinical
and emerging adults. interviews and a series of questionnaires described
The current study aimed to test whether within- below. Approximately 3, 6, 9, 12, 15, and 18 months
person models of stress are more clinically useful in post-baseline, a trained research assistant re-ad-
identifying periods of increased risk for NSSI com- ministered a structural clinical interview by phone to
pared to between-person models of stress. Here, we assess NSSI and verbally administered question-
present results from two prospective studies of naires to assess depressive symptoms and stress.
adolescent and emerging adult females. We focused Retention across waves was high (see Table 2).
on females given that the rates of NSSI (Fox et al.,
2015) and reported effects of stress (Rose & Rudolph, Measures. Nonsuicidal self-Injury: Adolescents
2006) are higher among females compared to males. completed the Self-Injurious Thoughts and Behav-
These studies allow us to test a within-person, iors Interview (SITBI) (Nock, Holmberg, Photos, &
stress-threshold model of NSSI, which asserts that Michel, 2007) at baseline and each follow-up assess-
youth are at risk for engaging in NSSI during times ment. The SITBI is a structured clinical interview.
they experience increased stress relative to their own The current study focused on the presence (1) or
average stress level. We hypothesized that higher- absence (0) of NSSI (Have you ever purposefully hurt
than-usual stress, relative to an individual’s own yourself without wanting to die?). At each follow-up,
typical stress level, would be associated with con- participants were asked this same question regarding
current risk for NSSI thoughts (Study 2) and acts the previous 3 months. The SITBI has strong

© 2019 Association for Child and Adolescent Mental Health

1078 Adam Bryant Miller et al. J Child Psychol Psychiatr 2019; 60(10): 1076–84

Table 1 Demographic statistics (100% female)

Depression: The Moods and Feelings Question-
Study 1 Study 2 naire (MFQ; Costello & Angold, 1988) is a 33-item
self-report measure designed to assess depressive
N 220 40
Race
symptoms in children and adolescents. Participants
Black 24% 3% rated on a 3-point Likert scale how true (0 = not true,
White 64% 59% 1 = sometimes true, 2 = mostly true) each depressive
Asian 1% 25% symptom (e.g. ‘I didn’t enjoy anything at all.’) was in
Indigenous (Native American/Canadian) 0% 5% the previous 2 weeks. Mean scores were computed
Other/mixed race 10% 5%
Hispanic/Latinx 1% 3%
for baseline and each follow-up time point (Cron-
Age range 12–16 18–25 bach’s a’s = .92–.95).
Mean age 14.69 21.55
Data analytic plan. Data were prepared and ana-
Diagnostic presentation, Clinically elevated lyzed in SAS. NSSI occurrence was modeled in log-
currenta symptoms Diagnoses
link, binary outcome multilevel models in SAS PROC
Depression 10% 25% GLIMMIX, with observations nested within girls
Anxiety 11% 67% across the follow-up time period. All models were fit
Hyperactivity 17% – to the data using restricted maximum-likelihood
Attention 21% –
estimation assuming incomplete data were missing
Conduct 20% –
Substance use – 8% at random (Singer, 1998). Repeated measures of
stress were decomposed into between-person differ-
a
Study 1 = Behavioral Assessment Scale for Children, Study ences in mean stress and within-person fluctuations
2 = Structured Clinical Interview for DSM-IV. in stress relative to one’s longitudinal mean. Between-
person mean (level 2) average stress was defined as a
Table 2 Frequency of NSSI during follow-up periods and girl’s unique sample-standardized person longitudi-
retention rates nal mean across all follow-up assessments. Within-
person deviations in (level 1) stress were calculated as
Nonsuicidal self-Injury (%) Retention (%) a given assessment’s value minus a girl’s unique
Study 1 person mean across all assessments divided by the
Baseline 41 100 girl’s unique standard deviation (i.e. person-stan-
3 months 15 91 dardized). Thus, these within-person deviation vari-
6 months 14 88 ables captured fluctuations in stress at each wave
9 months 11 91
12 months 13 84
around a girl’s mean level of stress across all waves.
15 months 10 84 We ran three separate models predicting same-wave
18 months 8 80 NSSI from current within-person deviations in each
Study 2 form of stress (time-varying). Each model controlled
14-days 61a 73b for within-person passage of time since baseline and
Study 1 baseline reflects lifetime history of NSSI. For each between-person factors including age, sample-stan-
subsequent wave, the time interval is NSSI within the prior dardized mean depression, sample-standardized
three months. mean stress, psychotropic medication use, and
a
b
96% reported NSSI thoughts. reported exposure to abuse1 (see Miller et al., 2017).
Participants completed 88% of possible daily entries. Random intercepts and random effects of time-varying
predictors were examined and included only
convergent validity, inter-rater reliability (K = .90), when doing so improved model fit ( 2 pseudo log-
and test–retest reliability (K = .70) (Nock et al., 2007). likelihood); of note, inclusion of all random effects
regardless of model fit did not substantively alter
fixed-effects results. Results of multilevel models are
Stress: Stress was assessed with the Child Chronic
presented as gamma weights in tables, which are
Strain Questionnaire (Rudolph, Kurlakowsky, &
analogous to unstandardized beta weights in OLS
Conley, 2001). Here, we focused on peer stress (11
logistic regression. Odds ratios and 95% confidence
items, e.g., ‘Do kids at school pick on or tease you?’),
intervals are provided for significant effects. Because
mother–child stress (7 items, e.g., ‘Do you have
time-varying predictors were person-standardized,
trouble getting along with your mom?’), and aca-
gamma coefficients and associated odds ratios can
demic stress (6 items, e.g., ‘Do you need extra help or
be interpreted as the effect of a one person-SD
tutoring with your schoolwork?’). Participants
increase in the predictor relative to one’s person mean.
responded to these questions with a 5-point Likert
scale (1 = Not at all, 2 = A little, 3 = Some, 4 = Much,
5 = Very Much), and mean scores were computed.
Results
Adolescents completed this measure at each follow-
up regarding the previous 3 months (Cronbach’s Descriptive statistics. Descriptive statistics for
a’s = .86–.90). study variables are included in Tables 1 and 2.

© 2019 Association for Child and Adolescent Mental Health

doi:10.1111/jcpp.13072 Within-person stress and NSSI 1079

Bivariate correlations are presented in Table S1. All

12.39, 44.75
0.54, 0.68

1.72, 2.89

1.08, 1.66
95% CI
predictor values were normally distributed. Of those

Covariance parameters

0.24
0.23
reporting NSSI at each wave, most (approximately

SE
Model 3:Mother–child stress
70%) reported engaging in NSSI between 1 and 3

–
–

–
–

–
times. Attrition analyses revealed that individuals
with missing NSSI data at any time point did not

23.55
0.61

2.23

1.34
significantly differ from individuals with complete

OR
NSSI data across all time points on any predictor

–
–

–
–

–
variable (all p’s > .05).

1.39
0.10
0.06
0.33
0.29
0.23
0.13
0.12
0.11
SE

0.13
0.32
Multilevel models. Multilevel models examining
same-wave associations with NSSI are presented in

c
0.50***
Table 3. Overall, the pattern of results across the

0.80***

0.29***
3.16**
0.97
0.10

0.04

0.08
c
three models with different types of stressors was

0.30
similar. In all models, age (as a between-person
factor) was not associated with risk of NSSI engage-

11.94, 39.47
ment across waves. The within-person passage of

0.57, 0.70

1.63, 2.65

1.14, 1.73
95% CI
Outcome: Nonsuicidal Self-Injury Acts
time across waves was associated with declining risk

Covariance parameters
of NSSI. Between-person mean depression was

0.20
SE
associated with increased overall risk of NSSI. Life-

0
Model 2:Academic stress

–
–

–
–

–
time history of NSSI was the strongest predictor of
current NSSI. Between-person differences in stress

21.71
0.63

2.08

1.40
OR
across waves were not significantly associated with
risk of NSSI. However, consistent with hypotheses,

–
–

–
–

–
higher-than-usual stress at a given wave was asso-
1.36
0.09
0.06
0.30
0.22
0.25
0.12
0.12
0.11
SE

ciated with greater risk of NSSI within the same

wave.

.28
0
c
0.46***
3.08***

0.73***

0.34**
Discussion
1.05
0.10

0.02
c

0.35

0.02
Study 1 begins to clarify when an adolescent girl
Table 3 Results of within-person models predicting NSSI Acts across 3-month follow-ups

experiencing stress is at risk for engaging in NSSI.

12.12, 44.59

Consistent with study hypotheses, periods of higher-

0.56, 0.70

1.37, 2.50

1.07, 1.70
95% CI

than-usual stress were associated with greatest risk Covariance parameters

0.36
0.23
for engaging in NSSI within the same 3-month wave.
Interestingly, between-person mean stress was not SE
–
–

–
–

–
Model 1:Peer stress

associated with increased risk for engaging in NSSI.

Rather, higher-than-usual stress (compared to one’s
23.24
0.63

1.85

1.35
OR

typical level) was associated with same-wave NSSI.

–
–

–
–

This pattern was consistent across three types of

stressors: peer, mother–child, and academic stress.
1.36
0.22
0.06
0.33
0.22
0.25
0.15
0.15
0.12
SE

Prior work has shown different effects of stressors

0.27
0.29

across these domains on mood (e.g. Timmons &

c

Margolin, 2015). Interestingly, higher-than-usual

0.47***

Bold values are significant, **p < .01; ***p < .001.
3.15***

0.62***

stress across all three domains in the present study

0.30**
0.86
0.12

0.01
c

0.34

0.27

was associated with within-wave NSSI risk.

There was an overall decline in NSSI risk across
time. This likely reflects regression to the mean given
Within-person deviations in stress
Within-person deviations in stress

that our sample was recruited for high risk. We

provided a list of mental health resources to all
families in our study. Thus, the sample may have
Psychotropic medication

had higher rates of treatment engagement overall.

Regardless, girls were more likely to engage in NSSI
during periods of higher-than-usual stress within
Mean depression
Abuse exposure

the same three-month wave. Theories of NSSI posit

Lifetime NSSI

Mean stress

that the link between stress and engaging in NSSI is

Fixed effects
Intercept

Intercept

closer in space and time than months. Thus, we were

Predictor

interested in whether our results held when we

Study 1

Time
Age

examined the effects of stress and NSSI at the daily

level in Study 2. Additionally, while a three-month

© 2019 Association for Child and Adolescent Mental Health

1080 Adam Bryant Miller et al. J Child Psychol Psychiatr 2019; 60(10): 1076–84

retrospective report improves upon prior research, thoughts and acts and perceived stress separately
there is still a concern regarding recall bias between within three predefined periods each day: morning
time points. Study 2 attempts to replicate the results (defined as first waking to noon), afternoon (defined
from Study 1 with a more fine-grained assessment of as noon to 6 p.m.), and evening (defined as 6 p.m.
stress and NSSI. until the diary entry). While this method may intro-
duce recall biases compared to more frequent col-
lections, the day reconstruction method is preferred
to a single daily aggregate rating and may reduce
Study 2
participant burden (Kahneman, Krueger, Schkade,
Methods
Schwarz, & Stone, 2004).
Participants and procedures. Participants for Participants were provided with reminders every
Study 2 were drawn from a larger sample of 60 day at 5 p.m. with a link to complete the daily survey
emerging and young adults (aged 18–35, M as close to bedtime as possible. Participants were
age = 23.25, SD = 4.25, 85% female) with a history given until 11 a.m. the next morning to complete a
of NSSI (Turner, Cobb, Gratz, & Chapman, 2016). missed survey. Participants were also asked to report
Previous studies with the larger sample have any NSSI that happened between submitting the
assessed differences in interpersonal relationship previous evenings’ survey and the next morning’s.
quality between those with and without NSSI Participants completed 735 out of 840 possible
(Turner, Wakefield, Gratz, & Chapman, 2017), dif- daily entries, with an average of 12.10 per person
ferences in engaging in disordered eating and NSSI (SD = 3.39, Range = 1–15).2 Compliance did not vary
(Turner, Yiu, Claes, Muehlenkamp, & Chapman, as a function of age, gender, or psychiatric diagnosis,
2016), and predictors of resisting urges to engage in p’s > .12. Thirty-one participants reported engaging
NSSI (Turner, Baglole, Chapman, & Gratz, 2019). in NSSI on a total of 90 days, consistent with NSSI
Almost half of the participants (48%) reported engag- rates observed in previous microlongitudinal studies
ing in NSSI within 30 days of study participation, (Nock, Prinstein, & Sterba, 2009).
and 40% reported engaging in NSSI within 2 weeks
prior to study participation. Here, we were primarily Measures. NSSI: Consistent with Study 1, NSSI
interested in examining the relationship between thoughts and acts were assessed with two items
stress and NSSI among adolescent and emerging based on the SITBI (Nock et al., 2007): ‘Did you think
adult females; thus, we restricted analyses to females about purposefully injuring yourself without intend-
aged 18–25 (N = 40, M age = 21.55, SD = 2.14). ing to die (NSSI)?’ and ‘Did you engage in NSSI today?’
Sociodemographic characteristics between those
included and excluded did not differ. Detailed proce-
dures have been reported elsewhere (Turner, Cobb Stress: The Perceived Stress Scale (Cohen, Kamar-
et al., 2016). In brief, participants were recruited from ck, & Mermelstein, 1994) is a 10-item measure that
the community via online advertisements and com- assesses how frequently participants experienced
munity flyers. Inclusion criteria were as follows: (a) stress on a scale of 0 (never) to 4 (very often). Example
age 18–35; (b) 10 or more lifetime episodes of NSSI, (c) items include: ‘how often have you found that you
one or more NSSI episode in the past year, (d) at least could not cope with all the things that you had to do?’
1 thought or urge for NSSI in the past 2 weeks, and (e) and ‘how often have you felt difficulties were piling up
regular internet access to complete surveys. Given the so high that you could not overcome them?’ Partici-
transdiagnostic nature of NSSI, exclusion criteria pants provided three retrospective reports of stress for
were limited to conditions that could interfere with the morning, afternoon, and evening. Items were
the daily diary protocol, including current psychotic summed within period and then averaged across
disorders, mania, or substance dependence. Inclu- periods, creating an average daily perceived stress
sion/exclusion criteria were assessed via the Struc- score ranging from 0 (no perceived stress) to 38 (high
tured Clinical Interview for DSM-IV (SCID) (First, perceived stress) (Cronbach’s a = .92).
Spitzer, Gibbon, & Williams, 1997) and the Deliberate
Self-Harm Inventory (Gratz, 2001). Rates of current Depression: Major depressive disorder was diag-
psychiatric disorders for the sample are in Table 2. nosed via the SCID (First et al., 1997) at baseline.

Daily diary procedures. Participants completed a Data analytic plan. We used the same multilevel
baseline assessment and then completed daily diary modeling procedures from Study 1. Here, between-
entries for 14 days. To increase reliability and valid- person mean (level 2) perceived stress was defined
ity, a modified day reconstruction method was used as a participant’s unique sample-standardized
to characterize presence or absence of NSSI thoughts longitudinal person mean across all days, while
and acts and perceived stress. Participants complete within-person deviations were person-standardized
a single daily report at the end of the day during deviations around this mean. We ran one model
which they retrospectively report on different ‘epi- predicting same-day NSSI thoughts and one
sodes’ within the day. Participants rated NSSI model predicting same-day NSSI acts from current

© 2019 Association for Child and Adolescent Mental Health

doi:10.1111/jcpp.13072 Within-person stress and NSSI 1081

within-person deviations in daily perceived stress Consistent with hypotheses, higher-than-usual

(time-varying). Each model controlled for the effects perceived stress on a given day was associated with
of within-person passage of time since initial assess- greater risk of NSSI acts on the same day. As
ment and between-person factors, including age, expected, exploratory analyses examining whether
depression diagnosis,3 and standardized mean per- higher-than-usual stress predicted next-day NSSI
ceived stress. Although we expected the association after accounting for next-day higher-than-usual
between higher-than-usual stress to be most stress and NSSI today yielded no statistically sig-
strongly associated with same-day NSSI risk, we nificant prospective effects.
conducted exploratory analyses to examine associa-
tions between higher-than-usual stress today (t) and
Discussion
risk for NSSI tomorrow (t + 1) after accounting for
NSSI today (t) and higher-than-usual stress tomor- Results from Study 2 replicate results from Study 1
row (t + 1). and provide further support for our study hypothesis
that periods of higher-than-usual stress are associ-
ated with NSSI risk. Similar to Study 1, mean levels
Results
of daily stress were not associated with increased
Descriptive statistics. Descriptive statistics can be NSSI risk. Rather, participants were more likely to
found in Tables 1 and 2. Bivariate correlations of think about and engage in NSSI when they reported
between-person, average stress across all days, and higher-than-usual daily perceived stress relative to
any NSSI thoughts or acts across all days were their own average perceived stress.
examined. Higher average perceived stress was asso- Study 2 provides an important extension of Study
ciated with experiencing NSSI thoughts, r = .30, 1 as it demonstrates that this stress-threshold model
p < .001, but not NSSI acts, r = .05, p = .18. NSSI functions similarly in two independent samples.
thoughts were positively associated with NSSI acts, Results suggest that young people may have a
r = .38, p < .001. unique stress threshold wherein an upward shift in
stress may be associated with NSSI risk regardless of
Multilevel models. Models examining same-day mean stress levels relative to others. Further, results
NSSI thoughts and acts are presented in Table 4. are consistent with previous EMA studies that show
Neither age nor depression diagnosis (between-per- how increased within-person negative affect predicts
son factors) was associated with risk of NSSI NSSI over the course of next few hours (Rodr
ıguez-
thoughts across days. Between-person mean per- Blanco et al., 2018).
ceived stress was not associated with increased risk
of NSSI thoughts. However, higher-than-usual per-
ceived stress on a given day was associated with General discussion
greater risk of NSSI thoughts on the same day. Taken together, these two studies provide initial
A similar pattern emerged with regard to NSSI support for a within-person, stress-threshold model
acts. Age (between-person factor) was not associ- of NSSI among adolescents and emerging adults.
ated with greater risk of NSSI acts across days. Previous research has demonstrated consistent con-
Between-person mean perceived stress was not current associations between stress and NSSI. How-
associated with increased risk of NSSI acts. ever, the current studies together demonstrate that

Table 4 Results of within-person models predicting NSSI thoughts and acts across days

Outcome: Nonsuicidal self-injury thoughts Outcome: Nonsuicidal self-injury acts

c SE OR 95% CI c SE OR 95% CI
Fixed effects Fixed effects
Study 2

Intercept 0.58 3.05 – – 7.53** 3.31 – –

Age 0.02 0.14 – – 0.27 0.15 – –
Time 0.06 0.03 – – 0.06 0.04 – –
Depression diagnosis 0.40 0.66 – – 0.55 0.68 – –
Mean perceived stress 0.09 0.06 – – 0.03 0.05 – –
Within-person deviations in perceived stress 1.00*** 0.16 2.72 2.00, 3.70 0.71*** 0.17 2.04 1.46, 2.84

Covariance parameters Covariance parameters

c SE c SE

Intercept 2.70** 0.90 2.34** 0.96

Within-person deviations in stress 0.14 0.19 0 0

Bold values are significant, **p < .01; ***p < .001.

© 2019 Association for Child and Adolescent Mental Health

1082 Adam Bryant Miller et al. J Child Psychol Psychiatr 2019; 60(10): 1076–84

simply knowing how a young person’s stress level important future direction will be to examine mech-
compared to others (between-person) may not pro- anisms linking higher-than-usual stress with risk for
vide enough information to predict within-wave or NSSI.
same-day NSSI risk. Rather, knowledge about when There are several important clinical implications
an adolescent or emerging adult exceeds their own from the present studies. Overall, results under-
typical stress level (within-person) may be more score the potential utility of within-person assess-
useful for informing clinical care. Importantly, we ment of stress and NSSI using evidence-based tools
use the term stress threshold as it best captures the with ongoing, repeated administrations. Clinicians
unique within-person effects we observed. However, may benefit from transitioning from solely assess-
it is important to note that this stress threshold is ing stress during an intake assessment (where risk
not necessarily an absolute, quantifiable threshold is compared relative to the population) to ongoing
within-person, but more so when an individual monitoring of stress in order to capture individual
exceeds their typical stress level. stress fluctuations. This personalized approach
Based on stress-sensitization literature, we does not rely on previously established clinical
hypothesized that examining within-person devia- cutoffs and allows for tailored treatment
tions in stress may provide more clinically useful approaches for a specific patient. This approach
results. Indeed, having an overall higher stress level can enhance safety planning, signal the need for
relative to others was not associated with NSSI, but, increased monitoring for patients during high
consistent with a stress-sensitization framework stress periods, and aid in tracking treatment
(Hammen et al., 2000), increased stress levels rela- progress. Ongoing assessment is consistent with
tive to one’s typical levels were helpful for predicting the routine safety monitoring in dialectical behav-
concurrent NSSI risk. We did not hypothesize that ior therapy (Linehan et al., 2006; McCauley et al.,
higher-than-usual stress would predict NSSI risk in 2018) as well as outcomes monitoring approaches
the next wave (study 1: next 3-month follow-up; to treatment (Boswell, Kraus, Miller, & Lambert,
Study 2: next day) after accounting for that wave’s 2015). Given that our results were consistent in the
stress. Indeed, exploratory analyses from Study 2 sample of late adolescents and emerging adults
confirmed that higher-than-usual stress did not (Study 2), our results also highlight the potential
predict next-day NSSI risk after accounting for importance of helping adolescents strategically
today’s NSSI and next-day stress. Theories of NSSI apply therapeutic tools during periods of higher-
posit that adolescents and emerging adults engage in than-usual stress in order to prevent NSSI during
NSSI to relieve proximal, intolerable distress (Chap- emerging adulthood.
man et al., 2006). Further, studies examining phys- Strengths across both studies include use of
iological arousal following a stressor document longitudinal methods with rigorous analytic meth-
temporary increase in indices of stress response ods. Both studies benefited from the use of multilevel
systems (e.g. cortisol as a marker of HPA axis) that modeling to simultaneously assess between- and
subside within minutes and hours for most (Gunnar within-person effects of stress on NSSI. Retention
& Quevedo, 2007). As a result, we believe that data and compliance were also high across studies.
on the level of minutes and hours, as gathered from Nevertheless, there are important considerations
EMA or passive data collection, are better suited to when interpreting results across studies. Both sam-
examine whether higher-than-usual stress now pre- ples were not representative from the general popu-
dicts NSSI risk in the next few minutes or hours, lation, and it is unclear how results would generalize
similar to work on negative affect (Rodr
ıguez-Blanco to diverse groups of young people, particularly boys.
et al., 2018). Importantly, across all studies, find- Across studies, the relationship between stress and
ings were over and above the effects of prior NSSI NSSI was assessed concurrently. Thus, causality
(Study 1), age, and depression (symptom or diagno- cannot be determined, and effects could be bidirec-
sis; Studies 1 and 2). tional. Given the focus of our study on NSSI, we did
Findings from typically developing samples of not examine differential prediction of NSSI from
adolescents and emerging adults help contextualize other forms of suicidal thoughts and behaviors.
our findings and point to important areas for future Our previous work has found that higher-than-usual
research. Within-person variations in daily stress stress was associated with risk for suicidal ideation
have been associated with both same-day (Bai & with a similar pattern emerging for suicidal behavior
Repetti, 2018) and next-day (Timmons & Margolin, but only among girls exposed to childhood abuse
2015) psychological functioning and altered same- (Miller et al., 2017). Future research may wish to
day biological stress responses (Fuligni et al., 2009; consider examining whether higher-than-usual
Lippold et al., 2016). Interestingly, some of these stress differentially predicts various forms of self-
studies point to important interactions between injurious thoughts and behaviors. Stress is a broad
within-person variations in stress and available term encompassing many subtle differences and
coping strategies (Sladek, Doane, Luecken, & Eisen- important distinctions within this broad category.
berg, 2016) or social support (Lippold et al., 2016) as In the current manuscript, each study captured two
potential buffers against increased stress. An different types of stress, actual stressors and

© 2019 Association for Child and Adolescent Mental Health

doi:10.1111/jcpp.13072 Within-person stress and NSSI 1083

perceived stress. While this is a limitation in that engage in NSSI when stress increases above their
results are not comparable in an absolute sense, it own typical levels.
could be considered a strength showing that results
may generalize between actual stressors (Study 1)
and perceived stress (Study 2). Nevertheless, future Supporting information
work may benefit from testing differences between Additional supporting information may be found online
actual stressors or perceived stress. Despite in the Supporting Information section at the end of the
increasingly short assessment periods for each article:
study, recall bias is a concern. All studies assessed Table S1. Correlations among Study 1 variables.
the presence or absence of NSSI. While this is a
coarse way to examine prospective risk, this is a
consistent challenge in the NSSI literature given the Acknowledgements
low base rates of NSSI (relative to other health risk This work was funded by the following grants:
behaviors). Regardless, assessing the link between F32MH108238 (A.M.), K01MH116325 (A.M.),
R01MH085505 (M.P.; M.N.), a Canada Graduate Schol-
stress and NSSI in two, prospective investigations
arship Doctoral award from the Canadian Institutes of
significantly contributes to the literature, given the
Health Research (B.T.), and a Canadian Institutes of
paucity of prospective studies in this domain (Fox Health Research Operating Grant (A.C.). The authors
et al., 2015). have declared that they have no competing or potential
Ultimately, our study results encourage a shift in conflicts of interest.
NSSI research to assess not only between-person
risk factors, but also within-person changes in risk
factors. Such work shifts from focusing on who is at Correspondence
risk for NSSI to when a specific at-risk person will Adam Bryant Miller, Department of Psychology and
engage in NSSI. Our study finds that, while between- Neuroscience, University of North Carolina at Chapel
person differences in stress are not robustly associ- Hill, 235 E. Cameron Ave, Chapel Hill, NC 27599, USA;
ated with NSSI risk, young people are most likely to Email: adam.miller@unc.edu

Key points

• Theoretical models and past research link life stress with risk for nonsuicidal self-injury (NSSI).
• Yet, recent meta-analytic evidence suggests that life stress inconsistently predicts NSSI.
• This might be because most work focuses on between-person differences in life stress.
• Here, we investigate whether a person is at risk for NSSI when they exceed their own typical levels of life stress.
• Our findings suggest that clinicians may be better able to predict NSSI by monitoring for periods when their
adolescent and emerging adult patients experience higher-than-usual life stress.

Notes Chapman, A.L., Gratz, K.L., & Brown, M.Z. (2006). Solving
the puzzle of deliberate self-harm: The experiential avoid-
ance model. Behaviour Research and Therapy, 44, 371–
1. Abuse did not predict NSSI, and the within- 394.
person effect of stress remained significant in models Cohen, S., Kamarck, T., & Mermelstein, R. (1994). Perceived
with or without abuse as a covariate. stress scale. Measuring Stress: A Guide for Health and Social
2. One participant continued diary participation for Scientists, 235–283.
an extra day. Costello, E.J., & Angold, A. (1988). Scales to assess child and
adolescent depression: Checklists, screens, and nets. Jour-
3. We did not have a daily measure of depression
nal of the American Academy of Child and Adolescent
symptoms. Psychiatry, 27, 726–737.
Eisenlohr-Moul, T.A., Miller, A.B., Giletta, M., Hastings, P.D.,
Rudolph, K.D., Nock, M.K., & Prinstein, M.J. (2018). HPA
References axis response and psychosocial stress as interactive predic-
Bai, S., & Repetti, R.L. (2018). Negative and positive emotion tors of suicidal ideation and behavior in adolescent females:
responses to daily school problems: Links to internalizing A multilevel diathesis-stress framework. Neuropharmacol-
and externalizing symptoms. Journal of Abnormal Child ogy, 43, 2564–2571.
Psychology, 46, 423–435. First, M.B., Spitzer, R.L., Gibbon, M., & Williams, J.B. (1997).
Boswell, J.F., Kraus, D.R., Miller, S.D., & Lambert, M.J. Structured clinical interview for DSM-IV clinical version (SCID-
(2015). Implementing routine outcome monitoring in clinical I/CV). Washington DC: American Psychiatric Press.
practice: Benefits, challenges, and solutions. Psychotherapy Fox, K.R., Franklin, J.C., Ribeiro, J.D., Kleiman, E.M., Bentley,
Research, 25, 6–19. K.H., & Nock, M.K. (2015). Meta-analysis of risk factors for

© 2019 Association for Child and Adolescent Mental Health

1084 Adam Bryant Miller et al. J Child Psychol Psychiatr 2019; 60(10): 1076–84

nonsuicidal self-injury. Clinical Psychology Review, 42, 156– Plener, P.L., Schumacher, T.S., Munz, L.M., & Groschwitz,
167. R.C. (2015). The longitudinal course of non-suicidal self-
Fuligni, A.J., Telzer, E.H., Bower, J., Cole, S.W., Kiang, L., & injury and deliberate self-harm: A systematic review of the
Irwin, M.R. (2009). A preliminary study of daily interper- literature. Borderline Personality Disorder and Emotion Dys-
sonal stress and C-reactive protein levels among adolescents regulation, 2, 2.
from Latin American and European backgrounds. Psycho- Ribeiro, J.D., Franklin, J.C., Fox, K.R., Bentley, K.H., Kleiman,
somatic Medicine, 71, 329–333. E.M., Chang, B.P., & Nock, M.K. (2016). Self-injurious
Gratz, K.L. (2001). Measurement of deliberate self-harm: thoughts and behaviors as risk factors for future suicide
Preliminary data on the deliberate self-harm inventory. ideation, attempts, and death: A meta-analysis of longitudi-
Journal of Psychopathology and Behavioral Assessment, nal studies. Psychological Medicine, 46, 225–236.
23, 253–263. Rodr
ıguez-Blanco, L., Carballo, J.J., & Baca-Garc
ıa, E. (2018).
Gunnar, M., & Quevedo, K. (2007). The neurobiology of stress Use of Ecological Momentary Assessment (EMA) in Non-
and development. Annual Review of Psychology, 58, 145–173. Suicidal Self-Injury (NSSI): A systematic review. Psychiatry
Hammen, C., Henry, R., & Daley, S.E. (2000). Depression and Research, 263, 212–219.
sensitization to stressors among young women as a function Rose, A.J., & Rudolph, K.D. (2006). A review of sex differences
of childhood adversity. Journal of Consulting and Clinical in peer relationship processes: Potential trade-offs for the
Psychology, 68, 782–787. emotional and behavioral development of girls and boys.
Hamza, C.A., & Willoughby, T. (2014). A longitudinal person- Psychological Bulletin, 132, 98–131.
centered examination of nonsuicidal self-injury among uni- Rudolph, K.D. (2002). Gender differences in emotional
versity students. Journal of Youth and Adolescence, 43, 671– responses to interpersonal stress during adolescence. Jour-
685. nal of Adolescent Health, 30(4, Supplement 1), 3–13.
Kahneman, D., Krueger, A.B., Schkade, D.A., Schwarz, N., & Rudolph, K.D., Kurlakowsky, K.D., & Conley, C.S. (2001).
Stone, A.A. (2004). A survey method for characterizing daily Developmental and social-contextual origins of depressive
life experience: The day reconstruction method. Science, control-related beliefs and behavior. Cognitive Therapy and
306, 1776–1780. Research, 25, 447–475.
Linehan, M.M., Comtois, K.A., Murray, A.M., Brown, M.Z., Scott, L.N., Pilkonis, P.A., Hipwell, A.E., Keenan, K., & Stepp,
Gallop, R.J., Heard, H.L., . . . & Lindenboim, N. (2006). Two- S.D. (2015). Non-suicidal self-injury and suicidal ideation as
year randomized controlled trial and follow-up of dialectical predictors of suicide attempts in adolescent girls: A multi-
behavior therapy vs therapy by experts for suicidal behav- wave prospective study. Comprehensive Psychiatry, 58, 1–10.
iors and borderline personality disorder. Archives of General Singer, J.D. (1998). Using SAS PROC MIXED to fit multilevel
Psychiatry, 63, 757–766. models, hierarchical models, and individual growth models.
Lippold, M., Davis, K.D., McHale, S.M., Buxton, O., & Almeida, Journal of Educational and Behavioral Statistics, 23, 323–
D.M. (2016). Daily stressor reactivity during adolescence: 355.
The buffering role of parental warmth. Health Psychology: Sladek, M.R., Doane, L.D., Luecken, L.J., & Eisenberg, N.
Official Journal of the Division of Health Psychology, Amer- (2016). Perceived stress, coping, and cortisol reactivity in
ican Psychological Association, 35, 1027–1035. daily life: A study of adolescents during the first year of
Liu, R.T., Cheek, S.M., & Nestor, B.A. (2016). Non-suicidal self- college. Biological Psychology, 117, 8–15.
injury and life stress: A systematic meta-analysis and theo- Timmons, A.C., & Margolin, G. (2015). Family conflict, mood,
retical elaboration. Clinical Psychology Review, 47, 1–14. and adolescents’ daily school problems: Moderating roles of
McCauley, E., Berk, M.S., Asarnow, J.R., Adrian, M., Cohen, J., internalizing and externalizing symptoms. Child Develop-
Korslund, K., . . . & Linehan, M.M. (2018). Efficacy of dialec- ment, 86, 241–258.
tical behavior therapy for adolescents at high risk for suicide: Turner, B.J., Baglole, J.S., Chapman, A.L., & Gratz, K.L.
A randomized clinical trial. JAMA Psychiatry, 75, 777–785. (2019). Experiencing and resisting nonsuicidal self-injury
Miller, A.B., Eisenlohr-Moul, T., Giletta, M., Hastings, P.D., thoughts and urges in everyday life. Suicide and Life-
Rudolph, K.D., Nock, M.K., & Prinstein, M.J. (2017). A Threatening Behavior, in press, https://doi.org/10.1111/
within-person approach to risk for suicidal ideation and sltb.12510
suicidal behavior: Examining the roles of depression, stress, Turner, B.J., Cobb, R.J., Gratz, K.L., & Chapman, A.L. (2016).
and abuse exposure. Journal of Consulting and Clinical The role of interpersonal conflict and perceived social
Psychology, 85, 712–722. support in nonsuicidal self-injury in daily life. Journal of
Nock, M.K., Holmberg, E.B., Photos, V.I., & Michel, B.D. Abnormal Psychology, 125, 588–598.
(2007). Self-injurious thoughts and behaviors interview: Turner, B.J., Wakefield, M.A., Gratz, K.L., & Chapman, A.L.
Development, reliability, and validity in an adolescent sam- (2017). Characterizing interpersonal difficulties among
ple. Psychological Assessment, 19, 309–317. young adults who engage in nonsuicidal self-injury using a
Nock, M.K., & Prinstein, M.J. (2004). A functional approach to daily diary. Behavior Therapy, 48, 366–379.
the assessment of self-mutilative behavior. Journal of Con- Turner, B.J., Yiu, A., Claes, L., Muehlenkamp, J.J., & Chap-
sulting and Clinical Psychology, 72, 885. man, A.L. (2016). Occurrence and co-occurrence of nonsui-
Nock, M.K., Prinstein, M.J., & Sterba, S.K. (2009). Revealing cidal self-injury and disordered eating in a daily diary study:
the form and function of self-injurious thoughts and behav- Which behavior, when? Psychiatry Research, 246, 39–47.
iors: A real-time ecological assessment study among ado-
lescents and young adults. Journal of Abnormal Psychology, Accepted for publication: 8 April 2019
118, 816–827. First published online: 3 May 2019

© 2019 Association for Child and Adolescent Mental Health