Content uploaded by Payton Jones
Author content
All content in this area was uploaded by Payton Jones on Jul 16, 2019
Content may be subject to copyright.
PREPRINT VERSION
1
Helping or Harming? The Effect of Trigger Warnings on Individuals with Trauma Histories
Payton J. Jones
Benjamin W. Bellet
Richard J. McNally
Harvard University, Cambridge, MA
Correspondence concerning this article should be addressed to Payton J. Jones, Department of
Psychology, Harvard University, 33 Kirkland St., Cambridge, MA 02138. E-mail:
payton_jones@g.harvard.edu
PREPRINT VERSION
2
Abstract
Objective: Trigger warnings alert trauma survivors about potentially disturbing forthcoming
content. However, most empirical studies on trigger warnings indicate that they are either
functionally inert or cause small adverse side effects. These evaluations have been limited to
either trauma-naïve participants or mixed samples. Accordingly, we tested whether trigger
warnings would be psychologically beneficial in the most relevant population: survivors of
serious trauma.
Method: Our experiment was a preregistered replication and extension of a previous one (Bellet,
Jones, & McNally, 2018); 451 trauma survivors were randomly assigned to either receive or not
receive trigger warnings prior to reading potentially distressing passages from world literature.
They provided their emotional reactions to each passage; self-reported anxiety was the primary
dependent variable.
Results: We found no evidence that trigger warnings were helpful for trauma survivors, for
those who self-reported a PTSD diagnosis, or for those who qualified for probable PTSD, even
when survivors' trauma matched the passages’ content. We found substantial evidence that
trigger warnings countertherapeutically reinforce survivors' view of their trauma as central to
their identity. Regarding replication hypotheses, the evidence was either ambiguous or
substantially favored the hypothesis that trigger warnings have no effect.
Conclusions: Trigger warnings are not helpful for trauma survivors. It is less clear whether
trigger warnings are explicitly harmful. However, such knowledge is unnecessary to adjudicate
whether to use trigger warnings – because trigger warnings are consistently unhelpful, there is no
evidence-based reason to use them.
Keywords: trigger warning, trauma, PTSD, resilience, pre-registered replication
PREPRINT VERSION
3
Helping or Harming? The Effect of Trigger Warnings on Individuals with Trauma Histories
1
2
Giving a trigger warning means providing prior notification about forthcoming content
3
that may be emotionally disturbing (Boysen, 2017). In this sense, trigger warnings are similar to
4
PG-13 or "viewer discretion advised" warnings that are common across many different forms of
5
media. Trigger warnings are distinct in that they originated as a measure of protection
6
specifically for survivors of trauma. For those with posttraumatic stress disorder (PTSD),
7
viewing reminders of trauma can spark painful reexperiencing symptoms (e.g., flashbacks;
8
American Psychiatric Association [APA], 2013). Trigger warnings originated in online
9
discussion groups for survivors of sexual trauma, where individuals would warn readers before
10
discussing their experiences. Since their inception, trigger warnings have expanded far beyond
11
the boundaries of specialized online communities. Trigger warnings are now used in educational
12
settings, social media, entertainment, and other venues. In addition to their expansion in setting,
13
they have also expanded in scope beyond sexual violence (Wilson, 2015).
14
Trigger warnings have sparked considerable debate in higher education. Proponents of
15
trigger warnings emphasize their importance in creating an inclusive atmosphere for
16
disadvantaged groups on campus (e.g., Karasek, 2016). They argue that trigger warnings provide
17
agency to engage or not to engage and that they allow trauma survivors to adequately prepare to
18
engage with difficult material. Critics suggest that trigger warnings imperil free speech,
19
academic freedom, and effective teaching, preventing students from engaging with challenging
20
material (e.g., Ellison, 2016). Other critics have suggested that trigger warnings foster
21
unreasonable expectations about the world, hampering natural resilience among young people
22
(e.g., Lukianoff & Haidt, 2015). Further, trigger warnings could also be problematic for trauma
23
survivors in particular (McNally, 2016). Those who view trauma as a core part of their identity
24
PREPRINT VERSION
4
have worse symptoms (Berntsen & Rubin, 2006; Brown, Antonius, Kramer, Root, & Hirst, 2010;
25
Robinaugh & McNally, 2011). Therefore, trigger warnings might iatrogenically reinforce the
26
importance of past traumatic events for the very people they were originally designed to help.
27
The arguments surrounding trigger warnings are often complex. Before diving into this
28
complexity, a much more basic question should be answered: do trigger warnings actually work?
29
From the vantage point of clinical science, trigger warnings are a type of community-based
30
clinical intervention intended to foster emotional well-being among trauma survivors. Yet due to
31
their grassroots origin in a non-clinical setting, trigger warnings have expanded for years without
32
the rigorous scientific evaluation that normally accompanies such interventions.
33
Bellet, Jones, and McNally (2018) were among the first to experimentally test the effect
34
of trigger warnings. In a crowd-sourced sample of individuals who had not experienced past
35
trauma, they found that trigger warnings given before literature passages had no significant effect
36
on anxiety. Further, they found that trigger warnings undermined participants' sense of their
37
resilience to potential future traumatic events, and that of others. They also reported a
38
moderation effect – among individuals who believed that words were emotionally harmful,
39
trigger warnings acutely increased anxiety reactions.
40
Since this original study, the scientific literature has quickly expanded. Bellet et al.
41
(2019) conducted a preregistered replication of the same protocol of Bellet et al. (2018) with
42
undergraduate college students. Their results suggest that trigger warnings created a trivially
43
small, yet genuine increase in anxiety. However, they found strong evidence that the previously
44
observed effects on projected vulnerability and the moderation effect from Bellet et al. (2018)
45
did not replicate among college students. In the most comprehensive set of studies to date,
46
Sanson, Strange, and Garry (2019) concluded that trigger warnings had trivially small effects
47
PREPRINT VERSION
5
overall. Across six studies of varying sample characteristics, they found that negative affect and
48
intrusive memories were similar regardless of whether individuals received trigger warnings.
49
Bridgland, Greene, Oulton, and Takarangi (2019) similarly found that trigger warnings
50
had trivially small effects on arousal levels when participants viewed photos. Importantly,
51
however, their results differentiated anticipatory anxiety from response anxiety. Anticipatory
52
anxiety refers to levels of anxiety after viewing the trigger warning but before viewing the
53
stimulus, whereas response anxiety refers to anxiety after viewing the stimulus. Although trigger
54
warnings appeared to have a trivial effect on response anxiety, they reliably increased
55
anticipatory anxiety. Relatedly, Bruce (2017) found that trigger warnings produced greater
56
physiological markers of anticipatory anxiety compared to PG-13 warnings or no warnings.
57
Gainsburg and Earl (2018) found that trigger warnings increased negative anticipatory
58
affect, but slightly decreased negative response affect. However, they found that participants
59
who were given trigger warnings were more likely to avoid both film clips and essays, which
60
may exacerbate anxiety in the long run (McNally, 2016). This particular finding is contrasted by
61
Kimble (2019), who found that individuals rarely avoided material due to trigger warnings.
62
Articles evaluating the effect of trigger warnings on anxiety or negative affect are summarized in
63
Table 1.
64
PREPRINT VERSION
6
65
PREPRINT VERSION
7
66
This encouraging growth of studies has begun to converge on the consensus that trigger
67
warnings are not typically helpful in reducing anxiety. This finding has been consistent across
68
various operationalizations of trigger warnings and types of stimuli. For instance, Sanson et al.
69
(2019) find similar effects regardless of whether trigger warnings mention potential emotional
70
reactions (e.g., "You might find this content disturbing") or whether they only mention content
71
(e.g., "The following story contains violence and death"). Similar effects are found with literature
72
passages, stories, photos, and film clips. The literature also suggests several different types of
73
harm potentially caused by trigger warnings (e.g., anticipatory anxiety, avoidance, perception of
74
vulnerability), but with occasionally mixed or contradictory results.
75
There remain several important limitations to this area of research. First, none of the
76
studies has exclusively focused on the primary intended target of trigger warnings – survivors of
77
trauma. Although some of the studies have included trauma survivor subgroups in their sample
78
(e.g., Sanson et al., 2019), this has not been the main focus of any study. If trigger warnings are
79
considered as a clinical intervention to promote the well-being of trauma survivors, this is an
80
important limitation. Second, trigger warnings present unique concerns for trauma survivors,
81
especially those who are experiencing symptoms of PTSD. Trigger warnings might reinforce
82
survivors' belief that their trauma is central to their identity, and the severity of PTSD symptoms
83
might also moderate trigger warnings’ effects. Third, many studies used different
84
operationalizations and stimuli. Although the consistency of results across diverse studies
85
inspires confidence, direct replications are also essential.
86
In the current study, we tested the effect of trigger warnings in a large sample of trauma
87
survivors recruited from Amazon Mechanical Turk, a crowd-sourcing platform. This
88
preregistered study includes a direct replication of the experiment in Bellet et al. (2018) and
89
PREPRINT VERSION
8
extends the paradigm to address questions specific to trauma survivors. Further exploratory
90
analyses examined vulnerable subpopulations (e.g., those who had received a diagnosis of
91
PTSD).
92
Method
93
Participants
94
The preregistration for the study design and analysis plan are available on the Open
95
Science Framework (https://osf.io/gdxtr/). Any deviations or exploratory analyses that were not
96
specified in the preregistration are marked as such within this manuscript. Participants were
97
recruited online from Amazon Mechanical Turk. We prespecified a sequential data collection
98
procedure with a stopping rule based on Bayes Factors. However, the evidential criteria for our
99
stopping rule were not met at any intermediate step, so we recruited participants until the
100
specified ceiling of 600 participants had completed the study. Participants were excluded from
101
the study if they incorrectly answered an attention check or if they failed an English fluency
102
verifier
1
. This left a final sample of 451 participants.
103
Procedure
104
This study was a randomized controlled experiment with one control group (no trigger
105
warnings) and one experimental group (trigger warnings for distressing passages). After
106
providing informed consent, participants were asked to complete a CAPTCHA and answer three
107
questions to verify US residency (e.g., "What is the most common emergency number in the
108
United States? [0-0-0 / 9-1-1 / 0-0-0-9-5 / 9-9-9]"). Participants failing these items were
109
immediately barred from completing the study. Remaining participants then completed a single-
110
1
"In the space provided below, please describe your activities last weekend in exactly 3 sentences.
(This question is very important and will be used to determine payment. Please use complete English
sentences)."
PREPRINT VERSION
9
item question that screened for the presence of trauma according to PTSD diagnostic Criterion A
111
in the Diagnostic and Statistical Manual of Mental Disorders (DSM-5; American Psychiatric
112
Association, 2013). The DSM-5 defines trauma as "exposure to actual or threatened death,
113
serious injury, or sexual violence." Participants who had not experienced such a trauma were
114
excluded from the study.
115
Participants then read literature passages typical of a high school or college English class.
116
Passages were standardized by length, and participants were shown the passages for a minimum
117
of 20 seconds before they were allowed to proceed to the next screen. The passages were
118
previously rated on the degree to which they provoked anxiety in a pilot study (Bellet et al.,
119
2018). Depending on their content, passages are hereafter classified as either neutral (no
120
disturbing content; e.g., a character description from Melville's Moby-Dick), mildly distressing
121
(general themes of violence or harm with no graphic details; e.g., a description of a battle from
122
Bradley's Flags of our Fathers), or markedly distressing (graphic scenes of violence, injury, or
123
death; e.g., the murder scene from Dostoevsky's Crime and Punishment). After each passage,
124
participants rated their emotional state by using slider bars ranging from 0 (not at all) to 100
125
(very much) on seven emotions: sad, happy, afraid, anxious, angry, content, and disgusted. The
126
responses to the anxious rating were used as the primary outcome measure.
127
Participants in both conditions first read three mildly distressing passages in random
128
order to establish a baseline. Next, participants read a series of five neutral passages and five
129
markedly distressing passages intermixed in random order. In the experimental condition,
130
markedly distressing passages were preceded by a trigger warning (TRIGGER WARNING: The
131
passage you are about to read contains disturbing content and may trigger an anxiety response,
132
especially in those who have a history of trauma). In the control condition, passages were
133
PREPRINT VERSION
10
preceded by a screen that indicated they were about to view the next passage, which was
134
acknowledged by clicking a radio button. After these 10 passages, participants read three more
135
mildly distressing passages appearing in random order that served to test for any sensitization
136
effects.
137
After reading all passages, participants completed the questionnaires detailed below.
138
Participants also answered questions about demographic information and psychiatric history,
139
completed an English fluency verifier, and answered validity-related questions that did not
140
impact payment (e.g., "What do you think was the purpose of this study?", "Is there any reason
141
you think that your data should not be used (this will not impact payment)?"). At the end of the
142
study, they received a debriefing form explaining the purpose of the study in detail.
143
Measures
144
Centrality of Event Scale (CES). The CES is a 7-item questionnaire that measures the
145
extent to which participants view the memory of their worst event as a reference point for
146
personal identity and the attribution of meaning to other experiences in their life (Berntsen &
147
Rubin, 2006). Items (e.g., "I feel that this event has become a central part of my life story") are
148
rated on a 5-point Likert scale (1 = totally disagree, 5 = totally agree). The CES displayed
149
excellent internal consistency in the current study (α = 0.94).
150
Perceived Posttraumatic Vulnerability Scale – Self (PPVS-S). This 19-item
151
questionnaire measures participants projections of their own emotional impairment and
152
posttraumatic symptoms if they were to hypothetically experience a trauma in the future (Bellet
153
et al., 2018). Participants are asked to imagine being exposed to an attempt on their life, and then
154
indicate their agreement with the effects of that experience (e.g., "I would not be able to work a
155
PREPRINT VERSION
11
job, or take care of myself") on a 100-point scale (1 = disagree, 100 = agree). The PPVS-S
156
displayed excellent internal consistency in the current study (α = 0.95).
157
Perceived Posttraumatic Vulnerability Scale – Other (PPVS-O). This 19-item
158
questionnaire measures participants' belief that if an "average" person were to experience a
159
trauma, they would experience persistent and debilitating emotional harm (Bellet et al., 2018).
160
Participants are asked to imagine an average person being exposed to an attempt on his or her
161
life, and then indicate their agreement with the effects of that experience (e.g., "he/she would
162
have nightmares of the event") on a 100-point scale (1 = disagree, 100 = agree). The PPVS-O
163
displayed excellent internal consistency in the current study (α = 0.95).
164
Life Events Checklist for DSM-5 (LEC-5). The LEC-5 is a self-report instrument that
165
identifies specific traumatic events that have occurred in one's lifetime (Weathers, Blake, et al.,
166
2013). The LEC-5 contains 16 events known to potentially result in PTSD or distress (e.g., "life
167
threatening illness or injury") and an additional option for "any other very stressful event or
168
experience". In our study, participants were initially screened by a question assessing the
169
presence of a Criterion A trauma. Later in the study, they were provided with the LEC-5 and
170
asked to choose the event description that best matched their most stressful or traumatic event.
171
PTSD Checklist for DSM-5 (PCL-5). The PCL-5 is a 20-item questionnaire that assesses
172
the presence and severity of PTSD symptoms in the past month (Weathers, Litz, et al., 2013).
173
When answering the PCL-5, participants in our study were instructed to answer the questions
174
keeping in mind their worst event as selected on the LEC-5. Items on the PCL-5 correspond
175
closely to DSM-5 criteria for PTSD (e.g., "In the past month, how much were you bothered by
176
repeated, disturbing, and unwanted memories of the stressful experience"). The PCL-5 is often
177
used to monitor symptoms over time, to screen for PTSD, or assist in making a provisional
178
PREPRINT VERSION
12
diagnosis of PTSD. For exploratory analyses involving the PCL-5, we used the cutoff score of 33
179
recommended by the United States Department of Veterans Affairs (Weathers, Litz, et al., 2013),
180
and based on research (Bovin et al., 2017; Wortmann et al., 2017). The PCL-5 displayed
181
excellent internal consistency in the current study (α = 0.96).
182
Words Can Harm Scale (WCHS). The WCHS is a 10-item scale that measures the
183
degree to which participants believe that words can cause serious and lasting emotional harm
184
(Bellet et al., 2018). Participants rated their agreement with each statement (e.g., "Even a simple
185
phrase can be emotionally traumatizing for someone vulnerable") on a 100-point scale (1 =
186
disagree, 100 = agree). The WCHS displayed excellent internal consistency in the current study
187
(α = 0.92).
188
Trigger Warnings Attitudes Assessment (TWAA). We administered three items to assess
189
participants' prior exposure to and attitudes about trigger warnings. First, we provided
190
participants with a definition of trigger warnings (i.e., "A trigger warning is a statement given
191
prior to presented material that allows the viewer to prepare for or avoid distress that it may
192
cause, particularly if the viewer has clinical mental health issues"). Participants were then asked
193
to give a binary rating of whether they believe that trigger warnings should be given prior to
194
potentially distressing material. If the participants selected "yes", they were shown a checklist
195
asking why they think trigger warnings should be used (e.g., "Trigger warnings help to protect
196
vulnerable populations…") including an "Other" option with the ability to write in a response.
197
Participants were then asked to rate their agreement with the statement "I have personally seen
198
many trigger warnings used before" on a 7-point Likert scale (1 = strongly disagree, 7 =
199
strongly agree). Only the last item was used for analysis in the present study, as specified in the
200
PREPRINT VERSION
13
preregistration. The other items are to be used in future studies addressing attitudes about trigger
201
warnings.
202
Trauma-Matching Passages. We asked participants if any of the literary passages they
203
read during the study reminded them of their worst event. If participants answered yes, we
204
provided them with a checklist of passages, and asked them to identify which ones reminded
205
them of their worst event. These passages were marked as "trauma-matching" passages.
206
Demographics Questionnaire. We asked participants to report their gender, race,
207
ethnicity, religiosity, political orientation, and age. Religiosity and political orientation were
208
assessed with a 5-point Likert scale (1 = not religious, 5 = extremely religious; 1 = very liberal, 5
209
= very conservative). We also asked participants to report whether they are currently a full-time
210
undergraduate student.
211
Psychiatric History. At the beginning of the study, all participants were given a screener
212
assessing for the presence of a Criterion A traumatic event. Participants were only included in
213
the study if they indicated the presence of a Criterion A event. Near the end of the study, we
214
asked participants whether they had “ever been diagnosed with a psychiatric or psychological
215
problem.” If participants answer yes, we asked them to choose all diagnoses that apply from a
216
list including PTSD and “Other” (to allow for a free response of any disorders not listed).
217
Analyses
218
All analyses were conducted in R (R Core Team, 2019). Analyses used linear regressions
219
with trigger warning condition (trigger warnings versus no trigger warning) as the primary
220
independent variable. For all analyses, we used Bayes Factors (BFs) as our inference criteria.
221
BFs give relative evidence between two competing hypotheses. For all tests, we used a
222
preregistered minimum BF value of 3 (or 1/3) as a criterion for "substantial evidence" relative to
223
PREPRINT VERSION
14
the null or alternative hypothesis. As indicated in the preregistration, we first examined whether
224
demographic or psychiatric history differed by condition. If this were the case, we added those
225
variables as covariates in regression analyses. Code for all analyses is available in the
226
supplemental materials.
227
Preregistered Replication Tests. We preregistered five replication tests, each related to
228
a previous effect observed in Bellet et al. (2018). For replication tests, our two competing
229
hypotheses were that (1) the observed effect was equal to zero (tobs = 0) or that (2) the observed
230
effect was equal to the effect in the previous study by Bellet et al. (tobs = torig). Replication BFs
231
were computed following the t-value comparison procedure described by Verhagen and
232
Wagenmakers (2014). We tested the effect of trigger warnings on (1) participants' perceptions of
233
their own posttraumatic vulnerability via the PPVS-S, (2) participants' perceptions on others'
234
posttraumatic vulnerability via the PPVS-O, (3) immediate anxiety response following markedly
235
distressing passages, (4) subsequent anxiety response to mildly distressing passages presented
236
without a trigger warning, and (5) an interaction effect between trigger warning condition and
237
the WCHS on immediate anxiety response (including a simple slopes analysis if the interaction
238
was significant).
239
Trauma-Specific Preregistered Tests. We preregistered several additional tests to
240
answer specific questions about trauma survivors. For these tests, our two competing hypotheses
241
were (1) the observed effect was equal to zero (tobs = 0) or that (2) the observed effect was not
242
equal to zero (tobs ≠ 0). Specifically, this is done by comparing a linear model which includes the
243
parameter of interest (e.g., condition) against a linear model without that parameter (e.g.,
244
intercept only model) using the lmBF function in the BayesFactor package (Morey & Rouder,
245
2018). First, we tested whether trigger warnings affected participants’ ratings of trauma
246
PREPRINT VERSION
15
centrality on the CES. Second, we tested whether PTSD severity scores on the PCL moderated
247
any of the previous tests (e.g., effect on PPVS-S, PPVS-O, etc.). Third, we tested whether
248
participants' self-reported prior exposure to trigger warnings (see TWAA) moderated any of the
249
previous tests.
250
Exploratory Tests. Based on critiques we received of our past work, we were interested
251
in whether the effect of trigger warnings differed in specific subgroups of our sample. It may be
252
that trigger warnings are not helpful for trauma survivors broadly but are indeed helpful for those
253
who have severe PTSD symptoms or have been diagnosed with PTSD. Accordingly, we tested
254
the effect of trigger warnings among the subgroup of individuals who (1) met a clinical cutoff for
255
a probable PTSD diagnosis or (2) reported a past diagnosis of PTSD.
256
Another possibility is that trigger warnings are only helpful when the content of the
257
passage matches the traumatic experience of the survivor (i.e., the passage actually triggers
258
remembrance of the trauma). We asked participants to identify "trauma-matching" passages,
259
allowing for a direct test of this hypothesis. We selected only the individuals who specified
260
trauma-matching passages and selected only the responses in reference to those specific
261
passages. We then tested whether trigger warnings prior to these trauma-matching passages
262
affected anxiety. In addition to testing trauma matching, we also tested whether the effect of
263
trigger warnings on anxiety was moderated by the type of trauma reported by participants.
264
Results
265
Sample Characteristics
266
Our sample contained a majority of self-identified females (n = 239, 53%) with a
267
significant minority of males (n = 208, 46%) and a small number of participants who specified
268
another gender (n = 4, 1%). Participants had a mean age of 37 (SD = 11.2 years), and identified
269
PREPRINT VERSION
16
their race as Caucasian (n = 336, 75%), Black/African American (n = 39, 9%), Asian/Pacific
270
Islander (n = 23, 5%), Hispanic (n = 23, 5%), Native American/Alaska Native (n = 5, 1%), or
271
multi-racial/selected multiple categories (n = 25, 6%). A substantial minority of participants
272
identified their ethnicity as Hispanic (n = 41, 9%). Participants identified as not religious (n =
273
201, 45%), somewhat religious (n = 72, 16%), moderately religious (n = 79, 18%), very
274
religious (n = 67, 15%), or extremely religious (n = 32, 7%). A minority of participants
275
identified themselves as full-time undergraduate students (n = 44, 10%). Participants were
276
skewed slightly toward liberal political orientation (mean = 2.64; 1 = very liberal to 5 = very
277
conservative). Participants reported a wide diversity of traumatic experiences on the LEC-5. All
278
16 categories were represented, with the largest categories being natural disaster (n = 95, 21%),
279
transportation accident (n = 79, 18%), sexual assault (n = 78, 17%), and physical assault (n = 47,
280
10%).
281
Preregistered Replication Tests
282
The results of the replication tests appear in Figure 1. Overall, replication tests either
283
favored the null hypothesis or gave ambiguous evidence. In the original study by Bellet et al.
284
(2018), a significant effect was found by trigger warning condition on perceived vulnerability to
285
self (PPVS-S) and perceived vulnerability of others (PPVS-O). Neither of these significant
286
effects replicated in our sample, with substantial evidence favoring the null hypothesis for an
287
effect on perceived vulnerability of others (PPVS-O). A significant interaction effect was also
288
found in the original experiment, such that participants' belief that words can harm (WCHS)
289
moderated the effect of trigger warnings on immediate increases in anxiety. This interaction
290
effect did not replicate in our sample, with substantial evidence favoring the null hypothesis. For
291
immediate increases in anxiety or sensitization to anxiety (which were nonsignificant in the
292
PREPRINT VERSION
17
original study), we found ambiguous evidence and substantial evidence favoring the null
293
hypothesis, respectively.
294
295
296
PREPRINT VERSION
18
297
298
Figure 1. Replication Tests
299
Bayes Factors (log) are presented representing relative evidence for either the null hypothesis
300
(tobs = 0) or relative evidence for a hypothesis of equivalence with the effect from the original
301
study (tobs = torig). Overall, evidence was either ambiguous or favored the null hypothesis.
302
303
PREPRINT VERSION
19
Trauma-Specific Preregistered Tests
304
First, we tested whether trigger warnings affected participants' ratings of trauma
305
centrality on the CES. We found substantial evidence that trigger warnings increased the degree
306
to which participants viewed their worst event as central to their life narrative (BF = 3.26).
307
Second, we tested whether PTSD severity scores on the PCL moderated any of the
308
previous tests (e.g., effect on PPVS-S, PPVS-O, etc.). We found substantial evidence favoring
309
the null hypothesis for a moderation effect on trauma centrality (BF = 0.11) and on perceived
310
vulnerability (self or other; BFs = 0.10, 0.13). We found ambiguous evidence for a moderation
311
effect of PTSD severity on anxiety sensitization (BF = 0.86). We found substantial evidence that
312
PTSD severity moderates immediate anxiety reactions (BF = 3.14). That is, individuals who
313
scored higher on the PCL had increased anxiety when they were given trigger warnings.
314
Third, we tested whether participants' self-reported prior exposure to trigger warnings
315
(see TWAA) moderated any of the previous tests. We found substantial evidence favoring the
316
null hypothesis for a moderation effect on trauma centrality (BF = 0.27), perceived vulnerability
317
(self or other; BFs = 0.19, 0.22), and anxiety sensitization (BF = 0.21). We found ambiguous
318
evidence for a moderation effect on immediate anxiety reaction (BF = 2.17).
319
Exploratory Tests
320
Critics of recent trigger warning research have suggested the plausible hypothesis that
321
whereas trigger warnings may not be helpful for college students generally (e.g., Bellet et al.,
322
2019) or even for trauma survivors generally, they may be helpful for more specific
323
subpopulations. For instance, it is possible that trigger warnings are only helpful for (1)
324
individuals with clinical-level PTSD symptoms or (2) individuals who have received a diagnosis
325
of PTSD. Furthermore, it may be that trigger warnings are only helpful when (3) the content of
326
PREPRINT VERSION
20
the literature passage directly matches the content of their trauma (i.e., it triggers a remembrance
327
of the trauma). We tested each of these hypotheses in exploratory tests. The results of these tests
328
are presented in Figure 2.
329
Full Sample. When comparing the null hypothesis (tobs = 0) to an open alternative
330
hypothesis (tobs ≠ 0) , the full sample showed substantial evidence favoring the null hypothesis
331
(BF = 0.14, n = 451). In other words, trigger warnings did not appear to affect immediate anxiety
332
reactions in our full sample.
333
Clinical Cutoff. When examining only individuals who met the cutoff of 33 on the PCL
334
for a probable diagnosis of PTSD recommended by the United States Department of Veterans
335
Affairs (Weathers et al., 2013), we found substantial evidence favoring the alternative hypothesis
336
(BF = 3.86, n = 150). Among these individuals, trigger warnings increased immediate anxiety
337
reactions. This is consistent with our preregistered test suggesting that PTSD severity scores
338
moderated the effect of trigger warnings on anxiety reactions.
339
Self-Reported Diagnosis of PTSD. For individuals who self-reported receiving a past
340
diagnosis of PTSD, we found substantial evidence favoring the null hypothesis (BF = 0.32, n =
341
53). That is, trigger warnings did not affect anxiety reactions for individuals who reported a
342
diagnosis of PTSD.
343
Matching Trauma Passages. We asked individuals whether the passages triggered
344
memories of their worst event and asked them to identify which passages were triggering.
345
Examining only the individuals who reported triggering passages, and examining only the
346
relevant passages, we found ambiguous evidence (BF = 0.88, n = 133) for an effect of trigger
347
warnings on anxiety. The effect was in the direction of increasing anxiety. That is, individuals
348
PREPRINT VERSION
21
who saw trigger warnings had trivially increased anxiety, suggesting that trigger warnings did
349
not reduce anxiety reactions when passages matched past traumatic experiences.
350
Trauma Type. We used the LEC-5 to assess the type of trauma that best characterized
351
each individual's worst event. Using the 16 categories from the LEC-5, we tested whether the
352
type of trauma impacted the effect of trigger warnings. We found substantial evidence favoring
353
the null hypothesis (BF < 0.001). However, some of the 16 categories had very few observations,
354
limiting the statistical validity of the test. Therefore, we tested for the influence of trauma type
355
by condensing the LEC-5 categories into 5 broad groups: sexual violence, other interpersonal
356
violence, accidental injury or illness, natural or other disaster, and other. Using these categories,
357
we again found substantial evidence favoring the null hypothesis (BF = 0.003). That is, the type
358
of trauma did not moderate the effect of trigger warnings.
359
360
PREPRINT VERSION
22
361
362
Figure 2. Trigger Warnings' Effect on Anxiety in Vulnerable Groups
363
Mean difference in anxiety change between the trigger warnings condition and the control
364
condition across subgroups. Shapes correspond to a Bayesian comparison of the null hypothesis
365
(tobs = 0) and an alternative hypothesis (tobs ≠ 0). The shaded region corresponds to the
366
boundaries of frequentist critical regions (p < 0.05, two-sided).
367
368
PREPRINT VERSION
23
Discussion
369
Past research has indicated that trigger warnings are unhelpful in reducing anxiety. The
370
results of this study are consistent with that conclusion. This study was the first to focus on a
371
sample of people who had survived Criterion A trauma as defined by the DSM-5 (APA, 2013).
372
Trigger warnings did not reduce anxiety for this sample broadly. Trigger warnings also did not
373
reduce anxiety among those who met a clinical cutoff for PTSD symptoms, reported a diagnosis
374
of PTSD, or those who reported that the stimuli matched the content of their past trauma. Trigger
375
warnings appeared to have trivially small effects on response anxiety overall. When effects did
376
emerge, they tended towards small increases in anxiety rather than decreases.
377
Bellet et al. (2018) previously found that trigger warnings increased individuals'
378
projections of their own vulnerability to future trauma, as well as the vulnerability of others. Our
379
results suggested substantial evidence that these effects did not replicate. Bellet et al. (2018) also
380
reported that individuals who endorsed the belief that words are emotionally harmful showed
381
greater anxiety in response to trigger warnings compared to individuals who did not endorse that
382
belief. Again, we found substantial evidence that this effect did not replicate. One possibility is
383
that these effects were unique to the trigger-warning naïve (trauma-naïve), crowd-sourced, older
384
sample used by Bellet et al. (2018). However, given that these effects originally had a small
385
effect size and did not replicate in larger samples of college students (Bellet et al., 2019) or
386
trauma survivors (present study), the original results may have been a false positive.
387
We found substantial evidence that giving trigger warnings to trauma survivors caused
388
them to view trauma as more central to their life narrative. This effect is a reason for worry.
389
Some trigger warnings explicitly suggest that trauma survivors are uniquely vulnerable (e.g., "
390
…especially in those with a history of trauma"). Even when trigger warnings only mention
391
PREPRINT VERSION
24
content, the implicit message that trauma survivors are vulnerable remains (why else provide a
392
warning?). These messages may reinforce the notion that trauma is invariably a watershed event
393
that causes permanent psychological change. In reality, a majority of trauma survivors are
394
resilient, experiencing little if any lasting psychological changes due to their experience
395
(Bonanno, 2004; Bonanno & Mancini, 2008). Aggregated across various types of trauma, only
396
4% of potentially traumatic events result in PTSD (Liu et al., 2017). However, trauma survivors
397
who view their traumatic experience as central to their life have elevated PTSD symptoms
398
(Berntsen & Rubin, 2006; Brown et., 2010; Robinaugh & McNally, 2011). Trauma centrality
399
prospectively predicts elevated PTSD symptoms, whereas the reverse is not true (Boals &
400
Ruggero, 2016). Decreases in trauma centrality mediated therapy outcomes (Boals & Murrell,
401
2016). This suggests that increasing trauma centrality is directly countertherapeutic. In other
402
words, trigger warnings may harm survivors by increasing trauma centrality.
403
We tested whether the severity of PTSD symptoms in our sample moderated any of our
404
tested hypotheses. In most cases, we found either evidence for no moderation or ambiguous
405
evidence. However, we did find substantial evidence that PTSD symptoms moderated the effect
406
of trigger warnings on response anxiety. For individuals who had more severe PTSD, trigger
407
warnings increased anxiety. This effect is ironic in the sense that trigger warnings may be most
408
harmful for the individuals they were designed to protect. We found no evidence that individuals'
409
prior exposure to trigger warnings moderated any of the previous effects.
410
A limitation of past research was that trigger warnings were primarily tested among
411
individuals who were trauma-naïve or in mixed samples. That is, the possibility remained that
412
despite being unhelpful for most who view them, trigger warnings may have been helpful for
413
trauma survivors or individuals with PTSD. This study largely puts these questions to rest.
414
PREPRINT VERSION
25
Trigger warnings were not helpful for trauma survivors. For individuals who met a clinical cutoff
415
for severity of PTSD symptoms, trigger warnings slightly increased anxiety. Trigger warnings
416
were not helpful for individuals who self-reported a diagnosis of PTSD. Perhaps most
417
convincingly, trigger warnings were not helpful even when they warned about content that
418
closely matched survivors' traumas. That is, when considering only the passages which
419
participants reported as reminding them of past trauma, trigger warnings were still unhelpful.
420
Public arguments regarding trigger warnings have been politically charged, complex, and
421
data-poor. Recent research on trigger warnings can importantly inform or perhaps even settle
422
some of these debates. Trigger warnings are unhelpful for trauma survivors, college students,
423
trauma-naïve individuals, and mixed groups of participants (Bellet et al., 2018; Bellet et al.,
424
2019, Bridgland et al., 2019; Sanson et al., 2019). Given this consistent conclusion, we find no
425
evidence-based reason for educators, administrators, or clinicians to use trigger warnings.
426
Whether trigger warnings are explicitly harmful is less clear. We found evidence that
427
trigger warnings increase the narrative centrality of trauma among survivors, which is
428
countertherapeutic (Boals & Murrell, 2016). We also found that trigger warnings increase
429
anxiety for those with more severe symptoms of PTSD. Although these effects were
430
preregistered and found in a large sample, the size of the effects were small and have not yet
431
been rigorously tested across multiple studies. However, such knowledge is unnecessary to
432
adjudicate whether to use trigger warnings – if there is no good reason to deploy them in the first
433
place, we need not require strong evidence of harm before abandoning them. Trigger warnings
434
should serve as an important caution to both clinical and nonclinical professionals who use
435
interventions aimed to improve well-being among trauma survivors. Such practices should be
436
PREPRINT VERSION
26
thoroughly vetted via appropriate scientific techniques before they are adopted. Using unvetted
437
interventions is irresponsible to victims of trauma.
438
PREPRINT VERSION
27
References
439
American Psychiatric Association. (2013). Diagnostic and statistical manual of mental disorders
440
(5th ed.). Washington, DC: Author.
441
Bellet, B. W., Jones, P. J., & McNally, R. J. (2018). Trigger warning: Empirical evidence ahead.
442
Journal of Behavior Therapy and Experimental Psychiatry, 61, 134-141.
443
https://doi.org/10.1016/j/jbtep.2018.07.002
444
Bellet, B. W., Jones, P. J., & McNally, R. J. (2019). Trigger warnings and resilience in college
445
students: A preregistered replication. Retrieved from osf.io/693gj
446
Berntsen, D., & Rubin, D. C. (2006). The centrality of event scale: A measure of integrating a
447
trauma into one's identity and its relation to post-traumatic stress disorder symptoms.
448
Behaviour Research and Therapy, 44, 219-231.
449
Boals, A., & Murrell, A. R. (2016). I am > trauma: Experimentally reducing event centrality and
450
PTSD symptoms in a clinical trial. Journal of Loss and Trauma, 21, 471-483.
451
Boals, A., & Ruggero, C. (2016). Event centrality prospectively predicts PTSD symptoms.
452
Anxiety, Stress, & Coping, 29, 533-541.
453
Bonanno, G. A. (2004). Loss, trauma, and human resilience: Have we underestimated the human
454
capacity to thrive after extremely aversive events?. American Psychologist, 59, 20.
455
Bonanno, G. A., & Mancini, A. D. (2008). The human capacity to thrive in the face of potential
456
trauma. Pediatrics, 121, 369.
457
Bovin, M. J., Marx, B. P., Weathers, F. W., Gallagher, M. W., Rodriguez, P., Schnurr, P. P., &
458
Keane, T. M. (2016). Psychometric properties of the PTSD checklist for diagnostic and
459
statistical manual of mental disorders–fifth edition (PCL-5) in veterans. Psychological
460
Assessment, 28, 1379.
461
PREPRINT VERSION
28
Boysen, G. A. (2017). Evidence-based answers to questions about trigger warnings for clinically-
462
based distress: A review for teachers. Scholarship of Teaching and Learning in
463
Psychology, 3, 163-177. http://dx.doi.org/10.1037/stl0000084
464
Bridgland, V. M., Green, D. M., Oulton, J. M., & Takarangi, M. K. (2019). Expecting the worst:
465
Investigating the effects of trigger warnings on reactions to ambiguously themed photos.
466
Journal of Experimental Psychology: Applied. Advance online publication.
467
http://dx.doi.org/10.1037/xap0000215
468
Brown, A. D., Antonius, D., Kramer, M., Root, J. C., & Hirst, W. (2010). Trauma centrality and
469
PTSD in veterans returning from Iraq and Afghanistan. Journal of Traumatic Stress,
470
23(4), 496-499.
471
Bruce, M. J. (2017). Does trauma centrality predict trigger warning use? Physiological
472
responses to using a trigger warning. Poster presented at the annual meeting of the
473
Midwest Psychological Association, Chicago, IL.
474
Ellison, J. (2016). Dear class of 2020 student. Retrieved from
475
https://news.uchicago.edu/sites/default/files/attachments/Dear_Class_of_2020_Students.p
476
df
477
Gainsburg, I., & Earl, A. (2018). Trigger warnings as an interpersonal emotion-regulation tool:
478
Avoidance, attention, and affect depend on beliefs. Journal of Experimental Social
479
Psychology, 79, 252-263.
480
Karasek, S. (2016). Trust me, trigger warnings are helpful. The New York Times. Retrieved from
481
https://www.nytimes.com/roomfordebate/2016/09/13/do-trigger-warnings-work/trust-me-
482
trigger-warnings-are-helpful
483
PREPRINT VERSION
29
Kimble, M. (2019). Why the warning? Student responses to triggering material based on trauma
484
history and symptom profiles. Symposium presented at the International Conference for
485
Psychological Science, Paris, France.
486
Liu, H., Petukhova, M. V., Sampson, N. A., Aguilar-Gaxiola, S., Alonso, J., Andrade, L. H., ... &
487
Kawakami, N. (2017). Association of DSM-IV posttraumatic stress disorder with
488
traumatic experience type and history in the World Health Organization World Mental
489
Health Surveys. JAMA Psychiatry, 74, 270-281.
490
Lukianoff, G., & Haidt, J. (2015). The coddling of the American mind. The Atlantic, 316(2), 42-
491
52.
492
McNally, R. J. (2016). If you need a trigger warning, you need PTSD treatment. The New York
493
Times. Retrieved from https://www.nytimes.com/roomfordebate/2016/09/13/do-trigger-
494
warnings-work/if-you-need-a-trigger-warning-you-need-ptsd-treatment
495
Morey, R. D., & Rouder, J. N. (2018). BayesFactor: Computation of Bayes factors for common
496
designs. R package version 0.9.12-4.2. https://CRAN.R-
497
project.org/package=BayesFactor
498
Robinaugh, D. J., & McNally, R. J. (2011). Trauma centrality and PTSD symptom severity in
499
adult survivors of childhood sexual abuse. Journal of Traumatic Stress, 24(4), 483-486.
500
Sanson, M., Strange, D., & Garry, M. (2019). Trigger warnings are trivially helpful at reducing
501
negative affect, intrusive thoughts, and avoidance. Clinical Psychological Science.
502
Advance online publication. https://doi.org/10.1177%2F2167702619827018
503
Weathers, F.W., Blake, D.D., Schnurr, P.P., Kaloupek, D.G., Marx, B.P., & Keane, T.M. (2013).
504
The Life Events Checklist for DSM-5 (LEC-5). Instrument available from the National
505
Center for PTSD at www.ptsd.va.gov
506
PREPRINT VERSION
30
Weathers, F.W., Litz, B.T., Keane, T.M., Palmieri, P.A., Marx, B.P., & Schnurr, P.P. (2013).
507
The PTSD Checklist for DSM-5 (PCL-5). Scale available from the National Center for
508
PTSD at www.ptsd.va.gov.
509
Wilson, R. (2015). Students' requests for trigger warnings grow more varied. The Chronicle of
510
Higher Education. Retrieved from https://www.chronicle.com/article/Students-Requests-
511
for/233043
512
Wortmann, J. H., Jordan, A. H., Weathers, F. W., Resick, P. A., Dondanville, K. A., Hall-Clark,
513
B., ... & Mintz, J. (2016). Psychometric analysis of the PTSD Checklist-5 (PCL-5) among
514
treatment-seeking military service members. Psychological Assessment, 28, 1392.
515
516