Pain 72 (1997) 107–113

Classical conditioning and the placebo effect

Guy H. Montgomery, Irving Kirsch*

University of Connecticut, Department of Psychology, U-20, 406 Babbidge Road, Storrs, CT 06269-1020, USA

Received 25 October 1996; revised version received 17 March 1997; accepted 25 March 1997

Abstract

Stimulus substitution models posit that placebo responses are due to pairings of conditional and unconditional stimuli. Expectancy theory
maintains that conditioning trials produce placebo response expectancies, rather than placebo responses, and that the expectancies elicit the
responses. We tested these opposing models by providing some participants with information intended to impede the formation of placebo
expectancies during conditioning trials and by assessing placebo expectancies. Although conditioning trials significantly enhanced placebo
responding, this effect was eliminated by adding expectancies to the regression equation, indicating that the effect of pairing trials on
placebo response was mediated completely by expectancy. Verbal information reversed the effect of conditioning trials on both placebo
expectancies and placebo responses, and the magnitude of the placebo effect increased significantly over 10 extinction trials. These data
disconfirm a stimulus substitution explanation and provide strong support for an expectancy interpretation of the conditioned placebo
enhancement produced by these methods.  1997 International Association for the Study of Pain. Published by Elsevier Science B.V.

Keywords: Stimulus substitution; Placebo responses; Expectancies; Conditioning

1. Classical conditioning and the placebo effect
Although effects of placebos are well-established
(Kirsch, in press), the mechanisms by which those effects
are produced have only recently been seriously investigated.
Hypothesized mechanisms include enhanced endorphin
release (Levine et al., 1978), reduced anxiety (Sternbach,
1968), classical conditioning (Wickramasekera, 1980), and
response expectancy (Kirsch, 1985, 1990). Montgomery
and Kirsch (1996) reported data that are difficult to recon-
cile with the hypothesis that placebo responses are mediated
by such global mechanisms as anxiety reduction or the
release of endogenous opioids. They obtained a placebo
effect by administering the placebo in the guise of a local
anesthetic and applying a pain stimulus to treated and
untreated parts of the body. Because the pain stimulus
was applied simultaneously to both the treated (by placebo)
and untreated locations, the differences in reported pain
could not have been due to any global changes in sensitivity,
perception, or affect. Conversely, this placebo effect can be
* Corresponding author. e-mail: irvingk@uconnvm.uconn.edu
explained both by classical conditioning and response
expectancy hypotheses.
According to stimulus substitution models of placebo
effects (Herrnstein, 1962; Ader, 1988; Wickramasekera,
1980; Turkkan, 1989), active treatments are unconditional
stimuli (USs) and the vehicles in which they are delivered
(i.e. pills, capsules, syringes, etc.) are conditional stimuli
(CSs). The medical treatments that people experience dur-
ing their lives constitute conditioning trials, during which
the vehicles are paired with their active ingredients. These
pairings endow the pills, capsules, and injections with the
capacity to evoke therapeutic effects as conditional
responses (CRs).
Note that these models are based on a Pavlovian, stimulus
substitution conception of conditioning. Research in the
broader field of classical conditioning has led to the repla-
cement of stimulus substitution models by a conception of
conditioning phenomena that is consistent with expectancy
theory (Rescorla, 1988). An example of the kind of data
leading to the rejection of stimulus substitution models is
the finding that under certain circumstances, the pairing of a
CS with a US does not lead to associative learning. The
common feature of these circumstances is the fact that the
information value of the CS as a predictor of the US is

0304-3959/97/$17.00  1997 International Association for the Study of Pain. Published by Elsevier Science B.V.
PII S0304-3959 (97 )0 0016-X
108 G.H. Montgomery, I. Kirsch / Pain 72 (1997) 107–113
masked. For example, if the US has a high base rate of
occurrence in the absence of the CS, the effect of US-CS
pairings can be negligible. Data of this sort have led con-
temporary theorists to describe conditioning as the learning
of relations between events. Conditioning leads to the acqui-
sition of expectancies that certain events will follow other
events, and its occurrence depends of ‘the information that
the CS provides about the US’ (Rescorla, 1988, p. 153),
rather than on contiguity. The unconditional response
(UR) can be viewed as an anticipatory response that pre-
pares the organism for the occurrence of the anticipated US
(Siegel, 1983). The response may be compensatory (e.g.
when performance inhibition is anticipated) or preparatory
(e.g. when food is anticipated).
The informational view of classical conditioning can be
applied to placebo phenomena (see Kirsch, 1990, 1997).
Doing so begins with the premise that for a stimulus to
function as a US, it has to be perceived. But the active nature
of a drug, with which its vehicle is paired, is perceived only
by its effects. Therefore, in terms of information value, the
drug response is the US, rather than the UR. What is learned
during conditioning is that active drugs produce particular
effects. Thus, classical conditioning is one means by which
the response expectancies are acquired (Kirsch, 1990).
The question that is not answered by a cognitive inter-
pretation of classical conditioning is how expectancies pro-
duce the expected responses. Kirsch has hypothesized that
response expectancies, defined as the anticipation of non-
volitional responses, are capable of eliciting the expected
response in much the same way that intentions elicit vol-
untary behaviors (cf. Ajzen and Fishbein, 1980). Because
conscious thoughts and feelings are covert stimuli and
responses (Hull, 1930), response expectancies can be
thought of as unconditional stimuli for corresponding sub-
jective experiences (covert responses). The effect of place-
bos depends on the strength of the person’s expectancies,
not on how those expectations were formed. The neurobiol-
ogy of the effects of expectancies and intentions remains to
be established.
Conditioned enhancement of placebo pain responses has
been demonstrated in a series of studies by Voudouris et al.
(1985, 1989, 1990). In each of these studies, Voudouris et
al. reported enhanced placebo effects following a series of
conditioning trials, during which the intensity of a pain
stimulus was surreptitiously lowered when paired with pla-
cebo administration, and in one of the studies (Voudouris et
al., 1990), they reported that this conditioning effect was
greater than that of a verbal expectancy manipulation. From
a cognitive perspective, conditioning trials are themselves
ex-pectancy manipulations. So what was demonstrated by
Voudouris et al. (1990) can be interpreted as indicating that
conditioning trials are more effective than at least some
verbal manipulations in altering expectancies, a finding
that had previously been reported by expectancy theorists
(Wickless and Kirsch, 1989).
The purpose of the present study was to provide a test of
conflicting predictions derived from stimulus substitution
and expectancy models of conditioned placebo effects.
We did this in two ways. Firstly, in addition to replicating
the conditioning procedure used by Voudouris et al., we
added an informed pairing condition, in which participants
were told that the stimulus intensity was being lowered
during conditioning trails. If the conditioning effect demon-
strated by Voudouris et al. (1985, 1989, 1990) is a direct
consequence of CS-US pairings, it should not be hindered
by informing participants of the manipulation. Conversely,
if the conditioning effect is mediated by expectancy, non-
deceptive pairings should fail to produce conditioning. We
also included an extinction condition. According to the sti-
mulus substitution model of placebo effects, repeated
administration of a placebo should weaken its effects
(Wickramasekera, 1980). In contrast, from an informational
perspective, the effect of a treatment is the US that rein-
forces expectancies. Thus, placebo effects should confirm
the expectancies that generate them, thereby preventing
extinction. Unlike Voudouris et al. (1985, 1989, 1990), we
also assessed pain response expectancies after the experi-
mental manipulations, so as to evaluate the extent to which
changes in reported pain were preceded by changes in
expected pain.
2. Method
2.1. Participants
Participants were 24 female and 24 male undergraduate
students, ranging in age from 18 to 26 years old, who volun-
teered to participate in exchange for partial course credit
and who reported an absence of medical conditions and
medication use that might interfere with pain sensitivity
or pose an increased risk of resultant tissue damage.
These conditions and medications were: high blood pres-
sure, circulatory problems (e.g. Reynaud’s disease or family
history of the same), diabetes, heart disease or any other
heart problems, asthma, seizures, frostbite, past trauma to
hands, lupus erythematosus, arthritis, other large or small
joint disease or injury, or use of psychoactive drugs, analge-
sics, antihistamines, and anti-inflammatory medications.
Participants were randomly assigned to groups with the
restriction that there would be the same proportions of
males and females in each condition.
2.2. Setting and materials
Pain was generated by an iontophoretic pain stimulator
modeled closely after the apparatus used by Voudouris et al.
(1985, 1990). Iontophoresis is the driving of ions into a
participant’s skin with electric current. The intensity of
the pain stimulus is dependent on the amount and duration
of electric current utilized. Iontophoresis of potassium ions
causes a prickling sensation at lower intensities and a
 G.H. Montgomery, I. Kirsch / Pain 72 (1997) 107–113 109

cramping sensation at higher intensities (Voudouris et al.,
1985, 1989, 1990). The effect has been shown to be inde-
pendent of skin resistance (Benjamin and Helvey, 1963).
Our iontophoretic pain stimulator was powered by a ser-
ies of 12 volt direct current (DC) batteries. The console was
separate from the battery pack to provide increased maneu-
verability of the apparatus. The console contained controls
for intensity and duration of electric current and a battery
monitor to provide an easy check on battery levels. A cylin-
der was attached to participants’ arms with Velcro strips,
such that the skin on the ventral side of the forearm formed
the bottom of a cup. The interior of the cylinder (cup) con-
tained a platinum anode. When filled with 3% potassium
chloride solution, the solution formed the contact between
skin and anode. On the dorsal side of the middle forearm, a
metal cathode was covered by a 0.9% sodium chloride satu-
rated gauze to complete the circuit. Wires attached the
anode and cathode to the console. The apparatus could deli-
ver a range of 0–50 mA of current.
The placebo was a mix of iodine, oil of thyme, and water
which produced a brownish, medicinal smelling effect when
applied topically. The placebo was placed in a medicinal
looking bottle and labeled, ‘Trivaricane: approved for
research purposes only’. Placebo was applied to areas on
the ventral part of the forearm surrounding the cylinder. A
constant current source was incorporated in the apparatus to
control for changes in skin conductance due to use of pla-
cebo and ensure that differences in skin conductance would
not be confounded with placebo treatment (Voudouris et al.,
1985). An office in the University Health Center was used as
the setting for the experiment to alleviate skepticism on the
part of participants that might have been caused by prior
associations with the psychology department.
2.3. Measures
and introduced to the experimenter, who wore a white lab
coat and was described as a ‘Behavioral Medicine Re-
searcher’. Participants were told that a new, topical, local
anesthetic was being tested for its pain reducing effects.
They were told the drug’s name was, ‘Trivaricane,’ and
that it had been proven effective in reducing pain in preli-
minary studies at other universities. The number of pain
trials was described to each participant as part of the
informed consent procedure. Participants were told they
could discontinue participation at any time, without nega-
tive consequences. Those agreeing to participate then com-
pleted the medical screening form. Acceptable participants
were randomly assigned to one of four groups: extinction,
no treatment, informed pairing, or uninformed pairing.
The design of the study is depicted in Table 1. Partici-
pants in the informed pairing, uninformed pairing and
extinction groups experienced four blocks of pain trials
(calibration trials, pretest trials, manipulation trials, and
posttest trials). Participants in the no treatment group did
not participate in the block of manipulation trials. Calibra-
tion, pretest, and posttest trials were identical for all groups.
Order of administration (placebo trials first or second) was
counterbalanced during pretest, manipulation, and posttest
trials.
2.4.1. Calibration trials
To control for individual differences in pain perception, a
calibration procedure adapted from Voudouris et al., 1985
was utilized. In order to lessen the overall aversiveness of
study participation, we decreased the number and intensity
of calibration trials. Voudouris et al., 1985 increased
ascending calibration trials until a subjective ‘10’ on a 0–
10 scale was reported by participants, and descending trials
were followed by 15 trials of randomly chosen stimulus
intensity values sampled from each of their participants’ 0

2.3.1. Pain intensity scale

Table 1
Participants were asked to rate the intensity of their pain
Stimulation intensity as a function of experimental condition and trial
on a visually presented, 11-point, analog scale, anchored
block
with the captions ‘no pain’ and ‘intolerable pain,’ for 0
and 10 respectively. Participants were shown the scale and Group Trial block
responded by verbally stating the number on the scale cor- Pretest Manipulation Posttest
responding to their experienced intensity of the pain.
No treatment
Placebo trials 6 – 6
2.3.2. Pain expectancy measure No-Placebo trials 6 – 6
Participants were asked, ‘What do you expect the pain Extinction
intensity to be with the Trivaricane,’ and ‘What do you Placebo trials 6 6 6
expect the pain intensity to be without the Trivaricane?’ No-placebo trials 6 6 6
Informed pairing
Pain expectancy ratings were on an 11-point intensity Placebo trials 6 3 6
scale with the same anchors as those on the pain intensity No-placebo trials 6 6 6
scale. Uninformed pairing
Placebo trials 6 3 6
2.4. Procedure No-placebo trials 6 6 6

Note: During the manipulation block, the intensity reduction on placebo

Participants were greeted in the lobby of the University trials was made explicit to the informed pairing group but not to the
Health Center, escorted to an office by a research assistant, uninformed pairing group.
110 G.H. Montgomery, I. Kirsch / Pain 72 (1997) 107–113

(no pain) to 10 (intolerable pain) range. In the present study,
all participants were given an ascending series of pain sti-
mulation trials, beginning at 1 mA and then increasing by 3
mA until the participant rated the pain intensity as 7 or
higher. The highest mA level was then repeated and fol-
lowed by a descending series of trials, each 3 mA less
until reaching 1 mA. Pain trials were separated by approxi-
mately 5 s, and participants verbally rated pain intensity
immediately following each trial. At the conclusion of cali-
bration trials, a separate regression equation was calculated
for each participant by imputing the verbal intensity ratings
and their corresponding stimulus intensity levels in milli-
amperes. This was used to calculate the stimulus intensity
level corresponding to each individual’s rating of 6 and 3, as
predicted by the regression equation. For brevity, these
numbers will be used to indicate the corresponding stimulus
intensity level, although this level differs between indivi-
dual participants according to their personal perceptions of
pain.
informed pairing participants were told that the intensity
level during medication trials would be raised to its original
setting.
2.4.4. Posttest
Posttesting was begun 3 min after the conclusion of the
manipulation trials. Procedurally, the posttest was identical
to the pretest. For all participants, the intensity level
remained at 6 for placebo and no-placebo trials. Participants
were debriefed following completion of data collection.
3. Results
Following the procedures used by Voudouris et al. (1985,
1989, 1990), pain intensity ratings on trials with placebo
were subtracted from pain intensity ratings on trials without
Table 2

Means and standard deviations for pain ratings, expected pain ratings,
2.4.2. Pretest placebo effect, and expected placebo effect as a function of placebo ad-
The pretest consisted of two trials with placebo and two ministration and experimental condition
trials without placebo. The mA setting during pretest was
Variable Group
always equivalent to each participant’s 6 for all trials. The
medication was allowed at least 1 min to take effect and at Extinction No Informed Unin-
least 1 min to wear off following cleaning of the skin with treatment pairing formed
pairing
70% rubbing alcohol. Placebo was applied liberally with a
cotton swab to areas on the forearm surrounding the cup. Pain ratings
Pretest
Placebo
2.4.3. Manipulation trials
Mean 4.83 5.50 5.71 5.75
Three minutes after the conclusion of the pretest, manip- SD 1.34 1.48 1.42 1.73
ulation trials were commenced. All participants in the No-placebo
extinction, informed pairing, and uninformed pairing Mean 5.13 5.79 6.17 5.88
groups received ten pain trials with placebo and ten without SD 1.17 1.16 1.54 1.13
Placebo effect
placebo. No treatment participants waited for 5 min, follow-
Mean 0.29 0.29 0.46 0.13
ing which, the posttest trials were administered. Stimulus SD 0.84 0.99 1.08 1.26
intensity levels depended on group assignment. The unin- Posttest
formed pairing group replicated the conditioning groups of Placebo
Voudouris et al. (1985, 1990). Participants received stimu- Mean 5.42 5.38 5.96 4.71
SD 1.33 1.15 1.05 1.27
lation corresponding to their level 3 ratings on placebo trials
No-placebo
and to their level 6 ratings on no-placebo trials. They were Mean 5.63 5.96 6.33 6.75
not informed that stimulus intensity levels were being SD 1.45 1.16 1.23 1.48
altered. The informed pairing group was designed to test Placebo effect
the importance of verbal information and expectancies in Mean 0.21 0.58 0.38 2.04
SD 1.12 1.02 1.03 1.54
establishing placebo analgesia. Stimulus intensity levels
were identical to those in the uninformed pairing group. Expected pain
However, unlike the uninformed pairing group, the ratings
informed pairing group was advised of the intensity reduc- Placebo
tion on placebo trials prior to the administration of the Mean 5.50 5.33 4.92 2.75
SD 1.51 1.07 0.67 1.22
manipulation trials. The experimenter stated that the inten-
No-placebo
sity would be reduced on medication trials to examine the Mean 6.42 6.58 6.83 6.92
effectiveness of Trivaricane at lower intensities. For the SD 1.31 1.08 1.40 1.16
extinction group, both placebo and no-placebo trials had Expected placebo
an intensity level corresponding to their rating of 6. Follow- effect
Mean 0.92 1.25 1.92 4.17
ing completion of the manipulation trials, the expectancy
SD 0.79 1.22 1.16 1.47
measure was administered. Prior to its administration,
 G.H. Montgomery, I. Kirsch / Pain 72 (1997) 107–113
placebo. The resulting placebo response score is the change
in pain intensity rating due to placebo administration.
Expected placebo response was calculated similarly
(expected no-placebo pain ratings minus expected placebo
pain ratings). Means and standard deviations of placebo
responses and expected placebo responses as a function of
group membership are presented in Table 2.
A t-test on pretest placebo response scores (mean = 0.29,
SD = 1.02) indicated that these scores were significantly
different than 0, t(47) = 1.97, P , .05. This indicates that
the creme produced a small but significant effect even
prior to conditioning. A one-way analysis of variance on
pretest placebo response scores failed to reveal any signifi-
cant differences as a function of group assignment, F(3,44) =
0.20, P , 0.90.
One-way analysis of covariance, with pretest placebo
response as the covariate, revealed significant group differ-
ences for placebo response, F(3,43) = 5.62, P , 0.01, and
expected placebo response, F(3,43) = 17.95, P , 0.01.
Mean scores, adjusted for pretest placebo responses, are
presented in Table 3. Tukey’s HSD tests indicated that
posttest placebo responses and expected placebo responses
were significantly greater in the uninformed pairing group
than in each of the other three groups, P , 0.05. No other
comparisons between groups were significant.
Differences in posttest adjusted means could be due to
improvement in some conditions or deterioration in others.
Therefore, it was important to analyze within group changes
as well as between group differences. Paired comparison t-
tests revealed that the uninformed pairing group had greater
placebo responses at posttest than at pretest, t(11) = 2.81,
P , 0.02. Placebo responses did not change significantly
in the extinction group, t(11) = −0.18, P , 0.86, informed
pairing group t(11) = − 0.23, P , 0.83, or no treatment
group t(11) = 0.77, P , 0.46.
To test the hypothesis that changes in placebo response
were mediated by expectancy, we added expectancy scores
to the regression equation in the analysis of covariance of
posttest placebo response scores. This analysis indicated
that expectancy was significantly related to changes in pla-
cebo response, F(1,42) = 20.35, P , 0.001, and with expec-
tancy controlled, between group differences in posttest pla-
cebo response were not significant, F(3,42) = 1.93, P , 0.17.
Table 3
Mean posttest placebo response and expected placebo response adjusted
for pretest placebo response
Variable Group
Extinction No Informed Uninformed
treatment pairing pairing
Placebo response 0.21a 0.58a 0.40a 2.01b
Expected placebo 0.92a 1.25a 1.91a 4.18b
response
Note: Adjusted means sharing a common superscript do not differ signifi-
cantly from each other.
111
Table 4
Mean posttest placebo response adjusted for pretest scores and expectancy
Group
Extinction No treatment Informed Uninformed
pairing pairing
0.86 1.05 0.49 0.81
Mean posttest placebo response scores, adjusted for pretest
scores and expectancy, are reported in Table 4. The correla-
tion between posttest placebo response and expectancy was
highly significant, r = 0.70, P , 0.001.
The pain response scores of participants in the extinction
condition during the ten extinction trials are depicted in Fig.
1. A 10 × 2 (trial by placebo) repeated measures analysis of
variance revealed significant effects for trial, F(9,99) = 4.20,
P , 0.001, placebo, F(1,11) = 11.09, P , 0.01, and the trial
by placebo interaction, F(9,99) = 2.07, P , 0.04. Trend
analysis revealed significant linear effects for trial,
F(1,11) = 8.43, P , 0.02, and the linear trial by placebo
interaction, F(1,11) = 5.29, P , 0.04. Analyses of simple
effects revealed a significant linear effect across no-placebo
trials, F(1,11) = 22.71, P , 0.001, but not across placebo
trails, F(1,11) = 1.24, P , 0.29. Interpreted in conjunction
with Fig. 1, these data indicate a significant increase in
pain when the placebo was not administered, which was
inhibited by the administration of placebo, thus resulting
in an increase in the placebo effect across the ten extinction
trials1.
4. Discussion
According to Baron and Kenny, 1986: ‘To establish med-
iation, the following conditions must hold: First, the inde-
pendent variable must affect the mediator...second, the
independent variable must be shown to affect the dependent
variable...and third, the mediator must affect the dependent
variable...Perfect mediation holds if the independent vari-
able has no effect when the mediator is controlled’ (p.
1177).
By these criteria, we have established that the effect of the
Voudouris et al. (1985, 1989, 1990) conditioning procedure
on responses to placebo is completely mediated by expec-
tancy. Conditioning trials in the uninformed pairing group
altered participants’ placebo response expectancies. They
also altered their pain reports. Expectancy was highly pre-
dictive of pain reports, and conditioning had no effect when
expectancy was controlled.
The mediating role of expectancy was further supported
by the ability of expectancy-altering verbal information to
1
The trial by placebo interaction is mathematically equivalent to a main
effect on placebo response scores (no placebo pain ratings minus placebo
pain ratings for each pair of trials).
112 G.H. Montgomery, I. Kirsch / Pain 72 (1997) 107–113
Fig. 1. Increases in the magnitude of the placebo effect during extinction
trials.
obstruct the effect of conditioning trials. Our verbal manip-
ulation was designed to alter participants’ interpretations of
their experience during conditioning trials. Participants in
the uninformed pairing condition were led to attribute pain
reduction on placebo trials to the effect of the ointment.
Those in the informed pairing condition were led to attribute
pain reduction to changes in stimulus intensity. Because
significant enhancement of the placebo effect was limited
to participants in the uninformed pairing group, we con-
clude that it is the interpretation of conditioning trials, rather
than the trials themselves, that affects placebo responding.
Contemporary interpretations of conditioning phenomena
stress cognitive functions (Rescorla, 1988). Conditioning is
a means by which representations of the environment are
acquired. With respect to placebo phenomena, they are one
way in which expectancies about the effect of treatments are
acquired. However, they do not explain how these expec-
tancies elicit the expected responses. This stands in marked
contrast to stimulus substitution conceptions of classical
conditioning, upon which conditioning models of placebo
effects have been based (e.g. Herrnstein, 1962; Wickrama-
sekera, 1980; Ader, 1988; Turkkan, 1989). According to
stimulus substitution formulations, expectancies are epiphe-
nomenal, at best, and conditioning trials are directly respon-
sible for the production of placebo responses. This model of
conditioned placebo effects is not supported by the data of
the study reported here.
A final aspect of the data worth noting is the failure of the
initial pain reduction response to extinguish. Although this
is inconsistent with the report of Fedele et al. (1989) of a
gradual decrease in the effectiveness of placebo treatment of
dysmenorrhea over 3 consecutive menstrual cycles, it is
consistent with most other clinical data. For example, 8
weeks of daily placebo administration failed to reduce the
effectiveness of placebo treatment for panic disorder (Cor-
yell and Noyes, 1988); placebos retained their effectiveness
in the treatment of angina over a 6-month period (Boissel et
al., 1986); and continued placebo treatment of rheumatoid
arthritis was shown to be effective for as long as 30 months
(Traut and Passarelli, 1957). These data are inconsistent
with a stimulus substitution model (Wickramasekera,
1980), but not with an informational model of conditioned
placebo responses (Kirsch, in press). According to the sti-
mulus substitution formulation, the active ingredients of
medications are USs. From an informational perspective,
it is the effect of treatment, rather than the active ingredients
of treatment, that constitute the US. Extinction fails to occur
because by mimicking the effects of drugs, placebos pro-
duce the US that confirms expectancies of the drug
responses.
We not only failed to obtain evidence of extinction
over ten extinction trials, we found a significant increase
in the placebo effect over these trials. Pain stimulation trials
without placebo produced an increase in reported pain.
Increases in reported pain across repeated administration
of nociceptive stimulation has been reported previously
(Baker and Kirsch, 1993). Surprisingly, in the present
study, placebo administration inhibited this increase in
pain reports over trials, resulting in an increase in the pla-
cebo effect, which returned to baseline levels, however,
after the 3-min rest period. The reason for this temporary
enhancement of the placebo effect remains to be deter-
mined. One possibility is that the magnitude of the placebo
effect is proportional to the magnitude of experienced pain
without the placebo. In any case, the finding of a significant
increase in the placebo effect during extinction trials is
inconsistent with the stimulus substitution hypothesis (see
Wickramasekera, 1980).
Some caution in interpreting these extinction data is war-
ranted. The apparent preconditioning effect of the creme
was very small and may not have been a placebo effect.
Conversely, an expectancy interpretation is supported by
the finding that the posttreatment placebo effect in the
extinction group was highly correlated with the expected
placebo response (r = 0.79, P , 0.002). The extinction
hypothesis should be tested in future studies by conducting
extinction trials following conditioning trials.
Besides their theoretical significance, the questions of
whether and how conditioning affects placebo analgesia
have important implications for clinical practice. Placebo
effects and drug effects are often additive, and drug expec-
tancies can potentiate drug effects (Kirsch, 1990; Kirsch and
Rosadino, 1993). For this reason, placebo factors can be
used to enhance the effectiveness of active treatments.
Our data is consistent with those of previous studies in
indicating that direct experience is a potent means of enhan-
cing the effects of a pain reduction intervention. For exam-
ple, large initial doses of a pain medication should increase
the efficacy of subsequent smaller doses. However, our data
also suggest that this effect might be blocked if patients
were aware of the subsequent reduction in dosage. Based
on our data, it is reasonable to suspect that pain reduction
experiences have an impact on subsequent pain reduction
 G.H. Montgomery, I. Kirsch / Pain 72 (1997) 107–113
only to the extent that they are provided in such a way as to
impact patient’s expectations.
Acknowledgements
This article is based on a doctoral dissertation conducted
by Guy H. Montgomery under the supervision of Irving
Kirsch. We thank Michael Kurland and the staff of the
University of Connecticut Student Health Center for their
cooperation and for providing the space in which this
research was conducted. We thank Robin Bogner for her
assistance in the practical application of iontophoresis.
Guy Montgomery is now at Memorial Sloan-Kettering Can-
cer Center.
References
Ader, R., The placebo effect as a conditioned response. In: R. Ader, H.
Weiner and A. Baum (Eds.), Experimental foundations of behavioral
medicine: conditioning approaches, Lawrence Erlbaum, Hillside, NJ,
1988, 47–66.
Ajzen, I. and Fishbein, M., Understanding attitudes and predicting social
behavior, Prentice-Hall Englewood Cliffs, NJ, 1980.
Baker, S.L. and Kirsch, I., Hypnotic and placebo analgesia: order effects
and the placebo label, Contemp. Hypnosis, 10 (1993) 117–126.
Baron, R.M. and Kenny, D.A., The moderator-mediator variable distinc-
tion in social psychological research: conceptual, strategic and statistical
considerations, J. Pers. Soc. Psychol., 51 (1986) 1173–1182.
Benjamin, F.B. and Helvey, W.M., Iontophoresis of potassium for experi-
mental determination of pain endurance in man, Proc. Soc. Exp. Biol.
Med., 113 (1963) 566–568.
Boissel, J.P., Philippon, A.M., Gauthier, E., Schbath, J., Destors, J.M. and
the B.I.S. Research Group, Time course of long-term placebo therapy
effects in angina pectoris, Eur. Heart J., 7 (1986) 1030–1036.
Coryell, W. and Noyes, R., Placebo response in panic disorder, Am. J.
Psychiatr., 145 (1988) 1138–1140.
Fedele, L., Maurizio, M., Acaia, B., Garagiola, U. and Tiengo, M.,
Dynamics and significance of placebo response in primary dysmenor-
rhea, Pain, 36, (1989) 43–47.
113
Herrnstein, R., Placebo effect in the rat, Science, 138 (1962) 677–678.
Hull, C.L., Knowledge and purpose as habit mechanisms, Psychol. Rev.,
(1930) 511–525.
Kirsch, I., Response expectancy as a determinant of experience and beha-
vior, Am. Psychol., 40 (1985) 1189–1202.
Kirsch, I., Changing expectations: a key to effective psychotherapy,
Brooks/Cole, Pacific Grove, CA, 1990, 231 pp.
Kirsch, I., Specifying nonspecifics: Psychological Mechanisms of Placebo
Effects, In: A. Harrington (Ed.), The placebo effect, Harvard University
Press, Cambridge, MA, (1997) 166–186.
Kirsch, I. and Rosadino, M.J., Do double-blind studies with informed
consent yield externally valid results? An empirical test, Psychophar-
macology, 110 (1993) 437–442.
Levine, J.D., Gordon, N.C. and Fields, H.L., The mechanism of placebo
analgesia, Lancet, 2 (1978) 654–657.
Montgomery, G. and Kirsch, I., Mechanisms of placebo pain reduction: an
empirical investigation, Psychol. Sci., 7 (1996) 174–176.
Rescorla, R.A., Pavlovian conditioning: it’s not what you think it is, Am.
Psychol., 43 (1988) 151–160.
Siegel, S., Classical conditioning, drug tolerance and drug dependence. In:
Y. Israel, F.B. Glaser, H. Kalant, R.E. Popham, W. Schmidt and R.G.
Smart (Eds.), Research Advances in Alcohol and Drug Problems, (Vol.
7), Plenum Press, New York, 1983.
Sternbach, R.A., Pain: a Psychological Analysis, Academic Press, New
York, 1968.
Traut, E.F. and Passarelli, E.W., Placebos in the treatment of rheumatoid
arthritis and other rheumatic conditions, Ann. Rheum. Dis., 16 (1957)
18–22.
Turkkan, J.S., Classical conditioning: the new hegemony, Behav. Brain
Sci., 12 (1989) 121–179.
Voudouris, N.J., Peck, C.L. and Coleman, G., Conditioned placebo
responses, J. Pers. Soc. Psychol., 48 (1985) 47–53.
Voudouris, N.J., Peck, C.L. and Coleman, G., Conditioned response mod-
els of placebo phenomena: further support, Pain, 38 (1989) 109–116.
Voudouris, N.J., Peck, C.L. and Coleman, G., The role of conditioning and
verbal expectancy in the placebo response, Pain, 43 (1990) 121–128.
Wickless, C. and Kirsch, I., The effects of verbal and experiential expec-
tancy manipulations on hypnotic susceptibility, J. Pers. Soc. Psychol.,
57 (1989) 762–768.
Wickramasekera, I., A conditioned response model of the placebo effect:
predictions from the model, Biofeedback and Self-Regulation, 5, (1980)
5–18.