Changes in Caffeine States Enhance Return of Fear in Spider Phobia
Jayson L. Mystkowski, Susan Mineka, Laura L. Vernon, and Richard E. Zinbarg
Treatment of phobias is sometimes followed by a return of fear. Animal and human research has shown
that changes in external and internal contexts between the time of treatment and follow-up tests often
enhance return of fear. The present study examined whether shifts in caffeine (C) state would enhance
return of fear. Participants who were highly afraid of spiders (n ⫽ 43) were treated in 1-session
exposure-based therapy and tested for follow-up 1 week later. Participants were randomly assigned to 1
of 4 groups and received either placebo (P) or C at treatment and follow-up sessions: CC, PP, CP, and
PC. Results demonstrated state-dependent learning. Participants experiencing incongruent drug states
during treatment and follow-up (CP and PC) exhibited greater return of fear than those experiencing
congruent drug states (CC and PP).
Exposure-based treatment approaches have an excellent record
of efficacy in the treatment of specific phobias (e.g., Barlow, 1988;
Craske, 1999). This approach involves controlled in vivo exposure
to anxiety-provoking stimuli (e.g., Craske & Rowe, 1997; O
1997). In addition to mere exposure, therapists often model non-
fearful behavior when in the presence of a phobic stimulus (e.g., a
tarantula) and encourage participant imitation for each treatment
task. Many (but not all) studies have shown support for the
superiority of treatments that include the following beneficial
elements: in vivo versus imaginal exposure, graded as opposed to
ungraded exposure, and participant modeling (e.g., Bandura, Blan-
chard, & Ritter, 1969; Foa & Kozak, 1986; Marks, 1987).
In spite of the marked efficacy of exposure therapy in reducing
fears and phobias, some fear may return with the passage of time
since treatment. Specifically, return of fear, which is analogous to
Pavlov’s (1927) spontaneous recovery phenomenon, is the “reap-
pearance of fear that has undergone partial or complete extinction”
(Rachman, 1989, p. 147). The partial fear recovery may emerge
either between exposure sessions or after a long follow-up period.
The underlying mechanisms of return of fear have proven dif-
ficult to isolate. To date, researchers have examined treatment
variables (e.g., intensity or duration of exposure trials), posttreat-
ment variables (e.g., stressful life events that occur after treatment
and before retesting at follow-up), and individual difference vari-
ables during treatment (e.g., level of state arousal) that might
predispose certain individuals to greater return of fear as assessed
either through heart-rate or self-reported levels of fear. However,
as reviewed by Craske (1999), none of the variables examined in
these three broad categories have been shown to affect return of
fear in a reliable fashion. Thus, although there are a number of
different factors, both during and after treatment, that may influ-
ence the return of fear, these are yet to be determined and verified
empirically in a consistent manner.
Within the last 2 decades, new research in animals has examined
the effects of changes in context on return of fear and provided
new insights into possible determinants of return of fear in hu-
mans. This work on return of fear in animals has been inspired by
significant work in animals and humans on contextual control of
learning and memory more generally. Specifically, Bouton and
colleagues (e.g., Bouton, 1988, 1991, 1994; Bouton & Bolles,
1979; Bouton & Nelson, 1998) have provided strong evidence that
partial fear recovery
occurs in rats when a conditioned stimulus
(CS) that was extinguished or counterconditioned in one context
(other than the one where acquisition occurred) is later presented
in a different recovery context. Having similarity across extinction
and recovery contexts, on the other hand, prevents or minimizes
fear recovery (Bouton & King, 1983). According to Bouton
(1993), when a discrepancy between extinction and recovery con-
texts occurs, contextual cues favor the retrieval of memories as-
sociated with fear acquisition rather than the memories of extinc-
tion. Bouton’s work with rats suggests the importance of studying
contextual cues in human fear extinction because it has potentially
important implications for treatment of phobias and quite possibly
other anxiety disorders as well.
Recently Rodriguez, Craske, Mineka, and Hladek (1999) at-
tempted to determine whether similar fear recovery effects occur
in humans by examining the effects of changes in external context
on return of fear in spider-fearful participants following their
successful treatment 2 weeks earlier of spider fear using a live
tarantula as the fear-provoking stimulus. The results provided only
Fear recovery is the term used most often in animal research, whereas
return of fear is the term used most often in human research. The two terms
Jayson L. Mystkowski, Susan Mineka, Laura L. Vernon, and Richard E.
Zinbarg, Department of Psychology, Northwestern University.
Jayson L. Mystkowski is now at the Department of Psychology, Uni-
versity of California, Los Angeles. Laura L. Vernon is now at the De-
partment of Psychology, Auburn University.
We thank the undergraduate research assistants that helped with all
phases of this study and Beverly Rodriguez for her very careful training
and supervision of these individuals in the exposure therapy procedures.
We also thank William Revelle and Kris Anderson for the advice concern-
ing the caffeine manipulation and in interpreting the EA and TA results of
the effects of caffeine in our study.
Correspondence concerning this article should be addressed to Susan
Mineka, Department of Psychology, Northwestern University, Evanston,
Illinois 60208. E-mail: email@example.com
Journal of Consulting and Clinical Psychology Copyright 2003 by the American Psychological Association, Inc.
2003, Vol. 71, No. 2, 243–250 0022-006X/03/$12.00 DOI: 10.1037/0022-006X.71.2.243
weak support for the hypothesized effects of change of context on
return of fear, with heart-rate data in only one of the two relevant
conditions indicating greater return of fear in the group experienc-
ing a context shift than in the group experiencing no context shift.
Self-report results were not significantly affected by the same
manipulation; such desynchrony among different measures of fear
in studies of fear extinction is quite common (see Craske, 1999, for
a review). There were, however, two potentially important meth-
odological confounds that may have contributed to the relative
lack of significant results in this study (see Mineka, Mystkowski,
Hladek, & Rodriguez, 1999; Rodriguez et al., 1999, for discussion
of these limitations).
Therefore Mineka et al. (1999) modified the Rodriguez et al.
(1999) study design, attempting to make the necessary method-
ological changes to avoid these confounds. One important change
was increasing the distinctiveness of the different contexts relative
to those used by Rodriguez et al. Another important change in-
volved conducting follow-up testing in a completely novel context
that had not previously been encountered by the participants. (In
the Rodriguez et al. study, participants had previously been ex-
posed to the follow-up context with the tarantula present, prior to
treatment). Both of these changes enhanced external validity be-
cause individuals who encounter a phobic stimulus after therapy
are probably at least as likely to do so in novel and very distinctive
contexts as in contexts where they have previously encountered
their phobic stimulus (as in the Rodriguez et al. study). Highly
spider-fearful participants participated in a two-session study
(treatment and follow-up 1 week later). Results confirmed that
treatment was highly effective in decreasing fear. Moreover, par-
ticipants tested in a novel context at follow-up showed a greater
return of fear than participants tested in the same context as where
treatment had occurred. Nevertheless, it should be noted that the
observed effects were not as large as those seen in the animal-
conditioning literature. Mystkowski, Craske, and Echiverri (2002)
also found similar context shift effects using a within-groups
design in which all participants were tested in both the same and
a different context (in this case indoors and outdoors).
Beyond these significant yet preliminary results pertaining to
the effects of changing external contexts on return of fear in
humans, there is some experimental evidence from both animals
and humans suggesting that contexts are not limited to physical
settings. For an example particularly relevant to the current topic,
Bouton (1988, 1993) found that internal states/cues may also
enhance retrieval of memories associated with conditioning or
extinction, thus affecting a rat’s response to a previously extin-
guished stimulus. In particular, Bouton, Kenney, and Rosengard
(1990) found evidence in rats for context specificity of fear ex-
tinction using internal contexts (drug state) induced by diazepam
versus placebo (P). Following extinction of a fear CS in one drug
state, rats that were later tested for fear of the CS in a different drug
state (P vs. diazepam) showed greater return of fear than those
tested in the same drug state as during extinction. On the basis of
these findings, Bouton suggested that extinction of fear may be
specific to the drug context in which it occurs and that mismatch
of internal states experienced during treatment and follow-up can
lead to significant return of fear (Bouton & Swartzentruber, 1991).
Related results were reported by Cunningham (1979) using alcohol
rather than diazepam as an internal context.
Such findings using drug states as internal contexts in studies of
context specificity of fear extinction are one example of a more
general phenomenon known as state-dependent learning, which
can be defined as a decrement in learned performance after a shift
in physiological state. State-dependent learning has been investi-
gated primarily in learning of nonemotional responses (e.g., Over-
ton, 1978; Spear, 1973; see Craske, 1999, for a review) but also
occurs with regard to the retention of learning with emotional
content, particularly with regard to the retention of extinction, in
both humans and nonhuman animals (e.g., Bouton et al., 1990;
Gray, 1987; Marks, Viswanathan, Lipsedge, & Gardner, 1972).
One example particularly relevant to the current study was that
of Marks et al. (1972), who exposed situational phobics to fear-
provoking situations under one of two different drug conditions:
diazepam 4 hr before treatment withP1hrbefore treatment or P 4
hr before treatment with diazepam 1 hr before treatment. Marks et
al. found that individuals in the 4-hr diazepam condition (where
the drug’s anxiolytic effects were likely to be decreasing during
exposure therapy) had greater transfer of treatment gains when
later tested in a nondrug state than those receiving diazepam 1 hr
before treatment. Although Marks et al.’s study design did not use
the 2 ⫻ 2 design needed to demonstrate the full state-dependent
learning effect (which would require a group treated with the 4-hr
diazepam condition and tested with diazepam and a group treated
with the 1-hr diazepam condition and tested under diazepam), the
results of the study were nonetheless consistent with the state-
dependent learning phenomenon.
Rather than focusing on state-dependent learning effects with
anxiolytic medications, the current study focused on whether state-
dependent learning effects in extinction also can be demonstrated
using caffeine (C)—a very widely used drug that is reliably anx-
iogenic, at least in patients with generalized anxiety disorder,
social phobia, and panic disorder (e.g., Boer, 2000; Bruce, Scott,
Shine & Lader, 1992; Simmons, 1996). This is also an important
practical issue from a clinical standpoint given that the majority of
people treated for anxiety disorders undoubtedly ingest C from
time to time. This may sometimes occur before a treatment session
and is, in fact, likely to occur sometime during long-term follow-
up. Although there are no published accounts that we are aware of
on state-dependent learning in anxious individuals using C, Lowe
(1987, 1988) conducted a 2-day study of unselected participants
that examined the interaction of common social drugs on state-
dependent learning on a cognitive task. Results showed a signifi-
cant decrement in recall from the cognitive task in those partici-
pants who did not ingest C before the start of both the learning and
recall test sessions; no significant declines in recall emerged when
C was administered across both sessions.
On the basis of the results reflecting C’s effect on state-
dependent learning and on anxiety in individuals with anxiety
disorders, the clinically relevant question is simply whether
changes in C state affect return of fear (or potentially the short-
term effectiveness of treatment as well). Thus, the effects of
internal context on return of fear were investigated by manipulat-
ing drug state through C versus P ingestion in a one-session
exposure-based therapy and then again at follow-up, approxi-
mately 1 week later. We hypothesized that the results would
demonstrate state-dependent learning, with participants experienc-
ing incongruent drug states (C/P and P/C) exhibiting a signifi-
cantly greater return of self-reported fear (measured during a
MYSTKOWSKI, MINEKA, VERNON, AND ZINBARG
behavioral approach task) from posttreatment to follow-up than
those participants experiencing congruent drug states (C/C and
Forty-three fearful participants (33 women, 10 men) were selected from
introductory psychology classes at Northwestern University using the
31-item Spider Questionnaire (Klorman, Hastings, Weerts, Melamed, &
administered during a 1st-week mass testing session. Initially
participants were chosen from those in the top 20% of the distribution for
this questionnaire; scores ranged from moderate to extreme fear (range:
15–27, M ⫽ 18.28, SD ⫽ 3.63). In addition, participants reported that they
exhibited moderate avoidance of spiders (on a 0–8 point scale, M ⫽ 4.70,
SD ⫽ 2.03), as well as moderate impairment of functioning (on a 0–8 point
scale, M ⫽ 3.37, SD ⫽ 2.21). The participants were primarily college
freshmen of Caucasian background. Prior to the treatment session, partic-
ipants were informed of the nature of the study and of possible C con-
sumption that day. Five potential participants were excluded for the fol-
lowing reasons: refusing treatment, failing to complete treatment in the
time allotted, or displaying insufficiently strong fear during an initial
behavioral approach task (BAT). The primary criterion for insufficiently
strong fear was a participant reporting a fear level below 70 during the
pretreatment BAT, as measured by the Subjective Units of Distress Scale
(SUDS; Wolpe, 1973) on a 1–100 range. However, those participants who
had an initial fear level between 60 and 70 but refused to touch the spider
when prompted to do so (n ⫽ 5) were included because their behavioral
avoidance was significant in spite of slightly lower levels of reported fear.
The SUDS included the following anchors 0 ⫽ no fear,25⫽ mild fear,
50 ⫽ moderate fear,75⫽ strong fear, and 100 ⫽ extreme fear. Partici-
pants were given class credit for their participation in both sessions of the
study; Session 1 (treatment) lasted up to 3 hr and Session 2 (follow-up)
lasted 1 hr.
On arrival, participants were assigned to one of four groups according to
the experiment’s2⫻ 2 between-groups factorial design: CC (n ⫽ 11), PP
(n ⫽ 11), CP (n ⫽ 10), and PC (n ⫽ 11). Although assignment to groups
was initially random, about halfway through the experiment, we noticed
that pretreatment levels of fear during the first BAT were not equivalent
across groups. To remedy this problem thereafter, we used a matching
procedure whereby a qualified participant’s group assignment was based
on how many other participants in that group were in different parts of the
fear range during the initial BAT.
During the BATs and during treatment, the primary measure of self-
reported fear was the widely used SUDS described previously. Before and
after treatment and at follow-up, participants also completed a question-
naire that measured spider beliefs—the Spider Phobia Beliefs Question-
naire (SBQ; Arntz, Lavy, van den Berg, & van Rijsoort, 1993).
subscales of this questionnaire, given before each BAT, assessed a partic-
ipant’s thoughts about spiders and how they believed they would react in
the presence of a spider.
To monitor changes in arousal, specifically energetic and tense arousal,
we administered 40 items from Thayer’s Activation–Deactivation Adjec-
tive Checklist (AD-ACL; 1978, 1986) during each session before the first
exposure to the tarantula and again just before the postdrink BAT, which
occurred 0.5 hr after C or P had been ingested. Thayer’s AD-ACL is highly
reliable and has two core dimensions: energetic arousal (EA; lively, alert)
and tense arousal (TA; anxious, fearful). C, relative to P, reliably increases
both EA and TA as a linear function of increasing the dose of C from 0 to 4
mg/kg body weight (Anderson, 1994).
Experimental settings for the study included an instruction room and a
therapy room. Both were well lit and contained tables, chairs, and wall
decorations. One large, nonpoisonous Chilean Rose-Haired tarantula
(Phrixotrichus spatulata; legspan approximately 6 in. [15.2 cm]) served as
the phobic stimulus. In the therapy room, several exposure task items,
including garden and latex gloves, paintbrushes, and a small open container
for the tarantula were used. The tarantula was housed in a plastic cage with
an easily removable top.
To manipulate the internal state of the participants by a double-blind
procedure, we prepared two solutions (C citrate and a similar tasting P
solution of flat quinine water) and randomly designated them as either A or
B for the experimental therapist, who did not know whether A or B was C.
In reality, A was C and B was quinine. Participants were told that they
might or might not receive C. The amount of A or B solution administered
to a participant was set at 4 mg per kg of body weight or the equivalent
amount of C contained in three cups of coffee. Body weight was deter-
mined by self-report. A juice mixture, orange-flavored Tang (Kraft Foods),
was then used to mask the taste of both solutions.
There were four (one male, three female) highly trained undergraduate
research assistants who were used as experimental therapists. The research
assistants received in-depth training with a treatment manual and were
supervised by a graduate student with several years of experience in
treating spider phobia. Before completing training, each research assistant
had to practice with 2–3 practice participants under the close scrutiny of the
supervising graduate student.
The study consisted of two sessions separated by approximately 1 week:
Session 1 (treatment) and Session 2 (follow-up). On arrival, participants
were escorted to an instruction room where they were instructed about the
standardized nature of the experimental and treatment procedures and then
asked to give formal written consent to participate in the study. They also
completed an information sheet to indicate their degree of avoidance and
impairment caused by their spider fear (using two 0–8 point scales). Then,
participants were trained to use the SUDS for rating their level of fear/
anxiety throughout the course of the study. Next, participants completed
the SBQ (SBQ 1) and the AD-ACL (AD-ACL 1), followed by the pre-
treatment BAT (BAT 1).
For BAT 1, the tarantula was in its home cage, a small open container
on a table situated in one corner of the therapy room. Participants were
asked to quickly approach and touch the tarantula if they were able to do
so, or, if not, to touch the bottom of the container, or if not able to do either,
to approach as closely as possible, reporting the peak amount of fear they
were experiencing using the 0–100 point SUDS. Both the approach dis-
tance (measured in feet from the spider) and fear level were recorded by the
experimenter. The furthest approach distance was 13 ft (3.96 m), whereas
the closest distance was touching the spider with a fingertip. Following this
task, participants were given a 5-min rest period in the instruction room
(for details, see Rodriguez et al., 1999).
Following BAT 1, participants ingested the drink associated with their
assigned experimental condition (A or B) in the instruction room and
completed filler questionnaires and read magazines for 30 min. Participants
then completed the AD-ACL again (AD-ACL 2) and participated in a
second BAT (BAT 2) in the therapy room, identical to the first BAT, to
assess possible differential changes in fear caused by C or P ingestion. A
5-min rest period followed the avoidance task and preceded exposure
Klorman et al. (1974) reported the Spider Questionnaire to have high
internal consistency (
s ⫽ .83 to .90).
The two subscales (Spider Beliefs and Responses to Spiders) have
shown high internal consistency (
⫽ .94) and significant test–retest
reliability, r(25) ⫽ 0.68 to 0.84; Arntz et al., 1993.
CAFFEINE AND RETURN OF FEAR IN SPIDER PHOBIA
Exposure-based treatment was the longest part of Session 1. Participants
were escorted to the therapy room, given the treatment rationale, and asked
to rate the perceived credibility of the proposed 14-step treatment, which
consisted of a standard fear hierarchy (for details, see Rodriguez et al.,
1999). The experimenter explained and modeled each task for the partic-
ipant. Participants then performed the task and were periodically prompted
to rate their fear levels. The task was terminated when the participant’s fear
was reduced to mild levels, corresponding to 25 or less on the 100-point
scale. Treatment was considered complete if the participants reached the
14th step in the hierarchy (holding the tarantula in their bare hand). The
total amount of time allotted to treatment was 1.5 hr. On completing
treatment, participants were brought back to the instruction room where
they completed the SBQ 2.
Participants then engaged in a posttreatment BAT (BAT 3) in the
treatment room. During this BAT, participants were first asked to return to
the BAT 1 distance (the closest they had been able to approach on the first
BAT) and to rate their fear level. Next, participants were asked to proceed
further and touch the tarantula (2 were unable to do so and they rated their
fear at the closest point that they could reach). At each approach distance,
participants were asked to rate their fear level on the 100-point scale. The
approximate time to complete Session 1 was 3 hr.
One week later, participants participated in the hour-long follow-up
session. Participants were given a brief description of the procedures for
the final part of the study, as well as a review of the SUDS, and they also
completed SBQ 3 and AD-ACL 3. (Because the SBQ 3 was administered
before the C/P manipulation it was not expected to be sensitive to state-
dependency effects.) Meanwhile, the experimental therapist, who remained
blind, prepared the C/P solution according to each participant’s group
assignment. Participants were then given the juice drink and instructed to
read magazines for 30 min. Following this 30-min period, participants
completed AD-ACL 4 and engaged in another BAT (BAT 4) in the therapy
room. During the BAT 4, participants first reported their fear level at their
BAT 1 distance, at their BAT 3 distance, and, if they had not touched the
tarantula at BAT 3, they were prompted to do so and to report their fear
level at that closer approach distance for BAT 4. These were the only
dependent measures used for testing the state dependency hypothesis. As
before, approach distances and maximum fear levels were recorded by the
experimenter. This was followed by a 5-min rest period.
Finally, participants completed an exit interview assessing their thoughts
about the experiment and were then debriefed. Experimenters made sure
that participants were not anxious or highly aroused when leaving the
Data Analytic Strategy
To account for potential pretreatment differences between those partic-
ipants who ingested C and those who did not, the amount of change in
self-reported fear from the pretreatment BAT (BAT 1) to postdrink BAT
(BAT 2) was calculated and used as a covariate in the following analyses
that examined self-reported fear during BATs.
To examine treatment effectiveness and return of fear on our measures
of self-reported fear and spider phobic cognitions, we conducted analyses
of trend on the repeated measures factor focusing on the linear trend as an
index of the overall effectiveness in reducing fear from pretreatment
through follow-up and then examined the quadratic trend as an index of
return of fear.
Initially four mixed-design analyses of covariance (ANCOVAs)
were conducted, focusing on the primary dependent measure of
self-reported fear during the BATs, to determine whether (a) the
two groups tested after treatment and follow-up in the same drug
state showed comparable patterns of fear reduction at both the
initial and the final approach distances and (b) the two groups
tested after treatment and follow-up in different drug states showed
comparable patterns of fear reduction at both the initial and final
approach distances. These four mixed design ANCOVAs with 2
groups (AA vs. BB or AB vs. BA) ⫻ 4 (pretreatment, postdrink,
posttreatment, and follow-up repeated measures)
treatment was highly effective in decreasing levels of self-reported
fear when participants were asked to approach a tarantula during a
In particular, a 2 groups (AA vs. BB) ⫻ 4 repeated measures
ANCOVA with an analysis of trend on the repeated measures
factor at the initial approach distance (closest distance attained at
the pretreatment BAT, such as standing near the spider’s con-
tainer) revealed a significant downward linear trend for SUDS
scores, from BATs 1–4, for those participants under the same drug
state (i.e., AA and BB) at treatment and follow-up, F(1, 17) ⫽
884.77, p ⬍ .0001. There was no significant difference between
these groups in their rates of fear reduction, expressed by the lack
of a Group ⫻ Linear Trend interaction, at the initial approach
distance, F ⬍ 1. A parallel ANCOVA for the final approach
distance (closest distance attained at the posttreatment BAT, usu-
ally touching the spider) revealed the same significant downward
linear trend, F(1, 18) ⫽ 312.89, p ⬍ .0001, without a Group ⫻
Linear Trend interaction, F ⬍ 1. Similarly, the 2 groups (AB vs.
BA) ⫻ 4 repeated measures ANCOVAs with analyses of trend on
the repeated measures factor also revealed a significant downward
linear trend for those participants under different drug states at
treatment and follow-up (i.e., AB and BA), F(1, 17) ⫽ 104.93, p ⬍
.0001, for the initial approach distance, F(1, 17) ⫽ 31.06, p ⬍
.0001, for the final approach distance. There were no significant
Group ⫻ Linear Trend interactions, indicating no differences
between these two groups in their rates of fear reduction at either
the initial approach distance, F(1, 17) ⫽ 2.56, p ⬍ .128, or the
final approach distance, F ⬍ 1. Finally, all but 1 of the participants
(n ⫽ 42) were able to hold the tarantula in their bare hand by the
end of treatment, with the 1 participant only being able to touch the
tarantula with a fingertip by the end of the treatment session. This
participant’s data were included since touching the tarantula with
a fingertip was the closest approach distance possible during the
posttreatment and follow-up BATs.
Because none of these four ANCOVAs (two for AA vs. BB and
two for AB vs. BA) revealed any significant Group ⫻ Downward
Linear Trend interactions at either approach distance from the
pretreatment to the follow-up BATs, participants in groups AA and
BB were pooled into the Same group and participants in groups
AB and BA were pooled into the Different group in order to
maximize the power of subsequent analyses.
As expected after merging the four groups into the two afore-
mentioned groups, a 2 groups (Same vs. Different) ⫻ 4 (pretreat-
ment, postdrink, posttreatment, and follow-up repeated measures)
mixed design ANCOVA with an analysis of trend on the repeated
measures factor revealed significant declines in levels of self-
reported fear at the initial and final approach distances from the
For these analyses, initial and final approach distances were the same
for BATs 1 and 2.
MYSTKOWSKI, MINEKA, VERNON, AND ZINBARG
pretreatment to the follow-up BATs. Specifically, at the initial
approach distance, there was a significant downward linear trend
for groups, F(1, 37) ⫽ 454.75, p ⬍ .0001, that did not interact with
group assignment, F ⬍ 1. Moreover, at the final approach distance,
there were also significant decreases in self-reported fear from
pretreatment to follow-up, as indicated by a significant downward
linear trend, F(1, 38) ⫽ 171.27, p ⬍ .0001, that did not interact
with group assignment, F(1, 38) ⫽ 1.77, p ⬍ .191. (See Table 1
and Figures 1 and 2.) Thus the overall effectiveness of treatment
from pretreatment to follow-up did not differ in the Same and
With regard to the two subscales of spider phobic cognitions
(SBQ), 2 groups (Same vs. Different) ⫻ 3 (pretreatment, posttreat-
ment, and follow-up repeated measures) mixed-design ANOVAs
with analyses of trend on the repeated measures factor indicated
that exposure therapy reduced phobic thoughts about spiders as
indicated by a significant downward linear trend, F(1, 39) ⫽
149.55, p ⬍ .0001, as well as thoughts about how one might react
when in the presence of a spider as indicated again by a significant
downward linear trend, F(1, 39) ⫽ 58.08, p ⬍ .0001. However,
there were no significant differences by group or Group ⫻ Down-
ward Linear Trend interactions observed for either type of phobic
cognitions: F(1, 39) ⫽ 2.85, p ⬍ .100, for phobic thoughts, F ⬍ 1,
for reactions to spiders (see Table 2).
Return of Fear
To examine possible differential return of fear from BAT 3 to
BAT 4 1 week later, we examined the results for the quadratic
trends from the two groups by four repeated measures ANCOVAs
described previously. There was a significant quadratic trend in
self-reported fear from pretreatment to follow-up while at the final
approach distance, F(1, 38) ⫽ 8.13, p ⬍ .007. Moreover, there was
a significant interaction between groups (Same vs. Different) and
the quadratic trend at the final approach distance, F(1, 38) ⫽ 5.19,
p ⬍ .028. Examination of the interaction indicated that there was
more self-reported return of fear in those participants under dif-
ferent drug states at treatment and follow-up than there was in
those with consistent treatment and follow-up drug states. This
same quadratic trend was not observed at the initial approach
distance, F ⬍ 1; nor was there a significant interaction between
group and the quadratic trend at the initial approach distance,
F ⬍ 1 (see Table 1 and Figures 1 and 2).
Mixed-design ANCOVAs also indicated reliable quadratic
trends in phobic thoughts about spiders and in thoughts of how one
might react when exposed to a spider, Fs(1, 39) ⫽ 82.88
and 58.08, respectively, both ps ⬍ .0001. There were no signifi-
cant interactions of group with these quadratic trends, Fs ⬍ 1.
However, as noted above, this is not surprising given that the
SBQ 3 was administered before the C/P ingestion manipulation.
Predictors of Return of Fear
A simultaneous multiple regression analysis, collapsing across
all four groups, of various pretreatment, treatment, and posttreat-
ment variables (e.g., initial fear level and phobic cognitions about
the stimulus, treatment credibility and total treatment time, and
posttreatment fear level and phobic cognitions, respectively) was
conducted to determine predictors of return of fear (other than
context) at both the initial and final approach distances. Although
overall there was very little return of fear at the initial approach
Self-Reported Fear During Behavioral Approach Tasks (BATs):
Group SUDS1 SUDS2 SUDS3A SUDS3B SUDS4A SUDS4B
M 74.35 67.20 12.90 19.57 15.15 20.81
SD 10.87 17.87 11.28 15.87 13.51 16.24
M 80.30 68.50 16.75 17.85 18.00 31.75
SD 14.89 18.57 24.88 21.16 20.55 33.10
Note. Because of missing data, the Subjective Units of Distress Scale
(SUDS) values of 20 Group Same and 20 Group Different participants
were used to determine the adjusted means at the initial approach distance,
and the SUDS values of 21 Group Same and 20 Group Different partici-
pants were used to determine the adjusted means at the final approach
distance. SUDS1 ⫽ SUDS recorded at the pretreatment BAT; SUDS2 ⫽
SUDS recorded during postdrink BAT; SUDS3A ⫽ SUDS recorded during
the posttreatment BAT at the initial approach distance; SUDS3B ⫽ SUDS
recorded at the posttreatment BAT at the final approach distance;
SUDS4A ⫽ SUDS recorded during the posttreatment BAT at the initial
approach distance; SUDS4B ⫽ SUDS recorded at the posttreatment BAT
at the final approach distance.
Figure 1. Self-reported fear levels (SUDS) at initial approach distance
(measured during BAT 1) at pretreatment (Pre-Tx), postdrink, posttreat-
ment (Post-Tx), and follow-up.
Figure 2. Self-reported fear levels (SUDS) at final approach distance
(measured during BAT 3) at pretreatment (Pre-Tx), postdrink, posttreat-
ment (Post-Tx), and follow-up.
CAFFEINE AND RETURN OF FEAR IN SPIDER PHOBIA
distance, the regression analysis indicated that high levels of
self-reported fear at the posttreatment BAT (BAT 3) predicted
return of fear (change from BAT 3 to BAT 4) at the follow-up
BAT (BAT 4) at the initial approach distance,
⫽ .662, R
F(1, 37) ⫽ 22.39, p ⬍ .0001, but not at the final approach distance.
No other predictors were significant at either approach distance.
Effects of C
The expected differential effects of C versus P on changes in
self-reported arousal received only very limited support in this
investigation. A great deal of prior research using the same dose of
C on unselected participants has shown that those given C show
EA and TA scores significantly higher than those given P 30 min
following ingestion of the drink (n ⫽ 2,730; d ⫽ .46 for EA, and
d ⫽ .37 for TA; W. Revelle, unpublished data, personal commu-
nication, June 20, 2001). By contrast, the two P groups and the two
C groups in the present study did not differ at postdrink, EA:
t(39) ⫽ 0.95, p ⬍ .35, d ⫽ .30; TA: t(39) ⫽⫺0.09, p ⬍ .93, d ⫽
.03. Moreover, in our study both C and P groups demonstrated
only a very slight increase in TA and a very slight decrease in EA
from the pretreatment AD-ACL (AD-ACL 1) to postdrink AD-
ACL (AD-ACL 2). Participants in both C groups (AA and AB)
during Session 1 exhibited a very slightly larger increase in TA
(pretreatment: M ⫽ 14.05, SD ⫽ 3.72; postdrink: M ⫽ 14.95,
SD ⫽ 6.07; ⌬⫽0.90), than those participants in the P groups (BB
and BA) during Session 1 (pretreatment: M ⫽ 14.35, SD ⫽ 6.62;
postdrink: M ⫽ 15.11, SD ⫽ 5.33; ⌬⫽0.76). With regard to EA
during Session 1, there was a slightly smaller decrease in mean EA
in the C groups (pretreatment: M ⫽ 14.25, SD ⫽ 6.87; postdrink:
M ⫽ 14.10, SD ⫽ 7.15; ⌬⫽⫺0.15) than in the P groups
(pretreatment: M ⫽ 12.90, SD ⫽ 6.50; postdrink: M ⫽ 11.93,
SD ⫽ 7.40; ⌬⫽⫺0.97). Mixed design ANOVAs failed to reach
significance for any of the observed mean differences, Fs ⬍ 1.
This failure to find expected significant increases in both EA
and TA for C groups in the present study seems primarily attrib-
utable to the fact that all our participants started with significantly
higher baseline EA and TA before consumption of either drink
than observed in prior studies with unselected participants about to
be studied on simple cognitive tasks. Specifically, in Revelle and
Anderson’s prior research (e.g., Anderson, 1994; Anderson &
Revelle, 1994), levels of EA and TA in their P participants were
significantly lower than in our study, t(1856) ⫽ 2.20, p ⬍ .025 for
TA, d ⫽ .35, and t(1900) ⫽ 2.42, p ⬍ .010 for EA, d ⫽ .38. Such
baseline elevations in our study are undoubtedly attributable to the
increased anticipatory anxiety that spider fearful participants
would be expected to experience when anticipating being treated
for their spider fear shortly.
In agreement with previous literature on the effectiveness of
exposure therapy for specific phobias, this investigation demon-
strated that graded exposure-based therapy with participant mod-
eling is highly effective in significantly reducing behavioral avoid-
ance, self-reported fear levels, and phobic beliefs pertaining to
spiders. Moreover, it also demonstrated that C ingestion per se
does not appear to reduce the efficacy of exposure therapy for at
least some specific phobias. Of more importance, however, the
results from this investigation confirmed our state-dependency
hypothesis regarding return of fear. Participants tested under con-
sistent drug states from treatment to follow-up exhibited signifi-
cantly less return of fear (at the final approach distance) than
participants tested under inconsistent drug states. This study also
extended prior work in humans (e.g., Marks et al., 1972) and
animals (e.g., Bouton et al., 1990) by showing that similar effects
to those obtained with anxiolytic drugs can be obtained with C—a
somewhat anxiogenic drug. In addition, these data replicate and
extend the previous work by Rodriguez et al. (1999), Mineka et al.
(1999), and Mystkowski et al. (2002), as well as extant findings in
the animal conditioning literature (e.g., Bouton, 1988, 1991, 1993,
1994; Bouton & Bolles, 1979; Bouton & Nelson, 1998), whereby
changes in external context from treatment to follow-up retention
tests lead to increased return of fear. Together, these results lend
further support to the notion that changes in context (including
internal drug state) from treatment to follow-up serve to enhance
return of fear among individuals receiving treatment for specific
fears and phobias.
One possible limitation of the present study is that the state-
dependency hypothesis was only supported at the final approach
distance. In several other studies on this topic using same or
different external contexts, results have been significant at the
initial approach distance. However, it seems quite possible that the
initial approach distance would not be as sensitive a measure in the
present experiment, where the physical context used during treat-
ment and follow-up for the Different groups was the same, as it is
when the physical context changes from treatment to follow-up for
Different groups as in Mineka et al. (1999) and Mystkowski et al.
(2002). That is, the effects of change in drug state may be manifest
most fully when one is about to encounter the feared stimulus itself
rather than at some relatively arbitrary physical place where one
had stopped during the very first behavioral approach task (and
that physical place is identical during follow-up).
Another possible limitation of the present study is that changes
in levels of TA and EA did not differ significantly between the C
and P participants on the AD-ACL. Specifically, participants in the
C group did not demonstrate a significant increase in either EA or
TA following C ingestion as anticipated; nor did their arousal
levels following C ingestion differ significantly from those in the
P groups (although all group differences were in the expected
Spider Phobic Beliefs
Group SBQ1A SBQ2A SBQ3A SBQ1B SBQ2A SBQ2B
M 50.15 17.07 13.26 25.49 7.48 4.85
SD 17.58 14.45 8.76 20.32 10.58 6.39
M 45.92 15.43 15.91 29.91 8.60 9.22
SD 21.62 10.74 15.23 21.45 9.74 13.64
Note. Spider Phobia Beliefs Questionnaire (SBQ) values of 22 Group
Same and 21 Group Different participants were used to determine the
means for SBQ1A and, due to missing data, the SBQ values of 21 Group
Same and 21 Group Different participants were used to determine the
remaining means. A ⫽ thoughts about the spider; B ⫽ what do you think
your response to the spider will be? 1 ⫽ recorded before treatment; 2 ⫽
recorded after treatment; 3 ⫽ recorded at follow-up.
MYSTKOWSKI, MINEKA, VERNON, AND ZINBARG
directions). This is in spite of the fact that the same C dose and
ingestion procedure was used here that has previously generated
significant increases in TA and EA, as well as significant differ-
ences from P groups. Although it is possible that we did not have
an adequate sample size to detect reliable differences, we do not
think this is the most likely explanation of our failure to observe
reliable changes in EA and TA. Instead, we think the more likely
explanation of our failure to observe hypothesized increases in
arousal stems from the already increased state of arousal that our
fearful participants were experiencing at the outset of the experi-
ment. Specifically, the prior studies demonstrating these C effects
used unselected participants about to participate in an information
processing study, who presumably were not in a state of anticipa-
tory anxiety (Anderson, 1994; Anderson & Revelle, 1994). Our
participants, by contrast, were all highly fearful of spiders and
knew that they had arrived at an experiment in which they were to
be treated for their fear. Consistent with this, participants’ baseline
levels of TA and EA were significantly higher than in the studies
conducted by Anderson (1994) and Anderson and Revelle (1994;
Revelle, personal communication, June 20, 2001). Moreover, the
postingestion measures of EA and TA also occurred 30 min
following the first BAT, which means that residual arousal from
the BAT itself could have obscured possible arousal differences
due to C.
Because changes in C state are the only reasonable explanation
for the state-dependent results exhibited in this investigation, it
seems likely that the AD-ACL is simply not sensitive enough to
detect additive effects of arousal stemming from C ingestion and
arousal stemming from the frightening nature of the anticipated
experimental tasks. Thus the AD-ACL does not seem to be an
adequate measure of state arousal induced by C when studying
participants selected as being highly fearful of an object or situa-
tion they are about to encounter. Whether there are other measures
of arousal that are more capable of reliably differentiating between
various sources of arousal among fearful individuals about to
encounter their feared objects remains to be determined.
Results of this study support the idea that state-dependent ef-
fects of C can influence return of fear in human participants treated
for spider phobia. As such, they complement and extend other
studies examining the effects of anxiolytic drugs on the outcome of
behavioral or cognitive–behavioral treatment for other anxiety
disorders. In addition to the study of Marks et al. (1972) on
state-dependent return of fear using diazepam in situational pho-
bics discussed previously, Marks et al. (1993) found related results
in an important study of behavioral and anxiolytic drug treatment
of agoraphobics. Using a 2 ⫻ 2 design, four groups of agorapho-
bics were treated for 8 weeks with (a) alprazolam and exposure
therapy, (b) P and exposure therapy, (c) alprazolam and relaxation
training (conceived as a less active therapy condition), or (d) P and
relaxation training. Medications were then tapered over an 8-week
period and assessments were continued for approximately 27 more
weeks. Results indicated that the group treated with alprazolam
and exposure showed a significantly higher relapse rate than the
group treated with P and exposure, supporting the state-dependent
learning phenomenon with anxiolytic drugs and exposure therapy
in a sophisticated large scale study of these issues in the treatment
The issues raised by the two Marks et al. (1972, 1993) studies
and those presented here are important for two reasons. First,
significant numbers of people undergoing behavioral treatment for
anxiety disorders receive a combination of anxiolytic medication
and behavioral treatment, and withdrawal from the medications
can clearly enhance return of fear. Second, fluctuations in C state
are also very common over the course of treatment, and follow-up
and such fluctuations may also enhance return of fear. Moreover,
one might also expect that among those who combine anxiolytic
medication and behavioral treatment with regular or irregular C
use, fluctuations in fear levels following exposure therapy might
be especially likely. However, further research will be necessary to
corroborate this hypothesis, and it would be particularly important
to test this hypothesis in a treatment-seeking clinical sample to
determine the generalizability of our results.
In conclusion, the most specific implication of our study is that
for clients who consume caffeinated beverages it may be wise to
conduct some but not all exposure sessions after drinking a caf-
feinated beverage. More generally, it may be beneficial to conduct
exposure therapy in a variety of contexts—both internal and
external—to minimize the likelihood of return of fear.
Anderson, K. J. (1994). Impulsivity, caffeine, and task difficulty: A within-
subjects test of the Yerkes–Dodson Law. Personality and Individual
Differences, 16, 813–829.
Anderson, K. J., & Revelle, W. (1994). Impulsivity and time of day: Is rate
of change in arousal a function of impulsivity? Journal of Personality
and Social Psychology, 67, 334–344.
Arntz, A., Lavy, E., van den Berg, G., & van Rijsoort, S. (1993). Negative
beliefs of spider phobics: A psychometric evaluation of the Spider
Phobia Beliefs Questionnaire. Advances in Behaviour Research & Ther-
apy, 15, 257–277.
Bandura, A., Blanchard, E., & Ritter, B. (1969). Desensitization and
modeling to induce behavioral, affective, and attitudinal change. Journal
of Personality and Social Psychology, 13, 173–179.
Barlow, D. H. (1988). Anxiety and its disorders: The nature and treatment
of anxiety and panic. New York: Guilford Press.
Bouton, M. E. (1988). Context and ambiguity in the extinction of emo-
tional learning: Implications for exposure therapy. Behaviour Research
& Therapy, 26, 137–149.
Bouton, M. E. (1991). A contextual analysis of fear extinction. In P. R.
Martin (Ed.), Handbook of behavior therapy and psychological science:
An integrative approach (pp. 435–453). Elmsford, NY: Pergamon Press.
Bouton, M. E. (1993). Context, time, and memory retrieval in the inter-
ference paradigms of Pavlovian learning. Psychological Bulletin, 114,
Bouton, M. E. (1994). Conditioning, remembering, and forgetting. Journal
of Experimental Psychology: Animal Behavior Processes, 20, 219–231.
Bouton, M. E., & Bolles, R. C. (1979). Contextual control of the extinction
of conditioned fear. Learning and Motivation, 10, 445–466.
Bouton, M. E., Kenney, F. A., & Rosengard, C. (1990). State-dependent
fear extinction with two benzodiazepine tranquilizers. Behavioral Neu-
roscience, 104, 44–45.
Bouton, M. E., & King, D. A. (1983). Contextual control of the extinction
of conditioned fear: Tests for the associative value of the context.
Journal of Experimental Psychology: Animal Behavior Processes, 9,
Bouton, M. E., & Nelson, J. B. (1998). The role of context in classical
conditioning: Some implications for cognitive behavior therapy. In
W. T. O’Donohue (Ed.), Learning theory and behavior therapy (pp.
59–83). Boston: Allyn & Bacon.
Bouton, M. E., & Swartzentruber, D. (1991). Sources of relapse after
CAFFEINE AND RETURN OF FEAR IN SPIDER PHOBIA
extinction in Pavlovian and instrumental learning. Clinical Psychology
Review, 11, 123–140.
Bruce, M., Scott, N., Shine, P., & Lader, M. (1992). Anxiogenic effects of
caffeine in patients with anxiety disorders. Archives of General Psychi-
atry, 49, 867–869.
Craske, M. G. (1999). Anxiety disorders: Psychological approaches to
theory and treatment. Boulder, CO: Westview Press.
Craske, M. G., & Rowe, M. (1997). A comparison of behavioral and
cognitive treatments of phobias. In G. Davey (Ed.), Phobias: A hand-
book of theory, research, and treatment (pp. 247–280). Chichester,
Cunningham, C. L. (1979). Alcohol as a cue for extinction: State depen-
dency produced by conditioned inhibition. Animal Learning & Behav-
ior, 7, 45–52.
den Boer, J. A. (2000). Social anxiety disorder/social phobia: Epidemiol-
ogy, diagnosis, neurobiology, and treatment. Comprehensive Psychia-
try, 41, 405–415.
Foa, E., & Kozak, M. (1986). Emotional processing of fear: Exposure to
corrective information. Psychological Bulletin, 99, 20–35.
Gray, J. A. (1987). The psychology of fear and stress. New York: Cam-
bridge University Press.
Klorman, R., Hastings, J. E., Weerts, T. C., Melamed, B. G., & Lang, P. J.
(1974). Psychometric description of some specific-fear questionnaires.
Behavior Therapy, 5, 401–409.
Lowe, G. (1987). Combined effects of alcohol and caffeine on human
state-dependent learning. Medical Science Research, 15, 25–26.
Lowe, G. (1988). State-dependent retrieval effects with social drugs. Brit-
ish Journal of Addiction, 83, 99–103.
Marks, I. M. (1987). Fears, phobias and ritual. Oxford, England: Oxford
Marks, I. M., Swinson, R. P., Basoglu, M., Kuch, K., Noshirvani, H., &
O’Sullivan, G. (1993). Alprazolam and exposure alone and combined in
panic disorder with agoraphobia: A controlled study in London and
Toronto. British Journal of Psychiatry, 162, 776–787.
Marks, I. M., Viswanathan, R., Lipsedge, M. S., & Gardner, R. (1972).
Enhanced relief of phobias following flooding during waning diazepam
effect. British Journal of Psychiatry, 121, 493–505.
Mineka, S., Mystkowski, J. L., Hladek, D., & Rodriguez, B. I. (1999). The
effects of changing contexts on return of fear following exposure treat-
ment for spider fear. Journal of Consulting and Clinical Psychology, 67,
Mystkowski, J. L., Craske, M. G., & Echiverri, A. M. (2002). Treatment
context and return of fear in spider phobia. Behavior Therapy, 33,
st, L. G. (1997). Rapid treatment of specific phobias. In G. Davey (Ed.),
Phobias: A handbook of theory, research, and treatment (pp. 227–246).
Chichester, England: Wiley.
Overton, D. A. (1978). Major theories of state dependent learning. In B. T.
Ho, D. W. Richards III, & D. L. Chute (Eds.), Drug discrimination and
state dependent learning (pp. 283–318). New York: Academic Press.
Pavlov, I. P. (1927). Conditioned reflexes. London: Oxford University
Rachman, S. (1989). The return of fear: Review and prospect. Clinical
Psychology Review, 9, 147–168.
Rodriguez, B. I., Craske, M. G., Mineka, S., & Hladek, D. (1999). Context-
specificity of relapse: Effects of therapist and environmental context on
return of fear. Behaviour Research & Therapy, 37, 845–862.
Simmons, D. H. (1996). Caffeine and its effect on persons with mental
disorders. Archives of Psychiatric Nursing, 10, 116–122.
Spear, N. E. (1973). Retrieval of memory in animals. Psychological Re-
view, 80, 163–194.
Thayer, R. E. (1978). Factor analytic and reliability studies on the
Activation-Deactivation Adjective Check List. Psychological Re-
ports, 42, 747–756.
Thayer, R. E. (1986). Activation-Deactivation Adjective Check List: Cur-
rent overview and structural analysis. Psychological Reports, 58, 607–
Wolpe, J. (1973). The practice of behavior therapy. New York: Pergamon
Received August 29, 2001
Revision received January 20, 2002
Accepted January 30, 2002 䡲
MYSTKOWSKI, MINEKA, VERNON, AND ZINBARG