Clinical Psychology Review 25 (2005) 67 – 95
Possible mechanisms for why desensitization
and exposure therapy work
Warren W. Tryon*
Department of Psychology, Fordham University, Bronx, NY 10458-5198, United States
Received 20 May 2004; received in revised form 8 July 2004; accepted 25 August 2004
Abstract
Rosen and Davison [Rosen, G.M. and Davison, G.C. (2003). Psychology should list empirically
supported principles of change (ESPs) and not credential trademarked therapies or other treatment packages.
Behavior Modification, 27, 300–312] recommended that empirically supported principles be listed instead of
empirically supported treatments because the latter approach enables the creation of putatively new therapies
by adding functionally inert components to already listed effective treatments. This article attempts to
facilitate inquiry into empirically supported principles by reviewing possible mechanisms responsible for the
effectiveness of systematic desensitization and exposure therapy. These interventions were selected because
they were among the first empirically supported treatments for which some attempt was made at
explanation. Reciprocal inhibition, counterconditioning, habituation, extinction, two-factor model, cognitive
changes including expectation, self-efficacy, cognitive restructuring, and informal network-based emotional
processing explanations are considered. Logical problems and/or available empirical evidence attenuate or
undercut these explanations. A connectionist learning-memory mechanism supported by findings from
behavioral and neuroscience research is provided. It demonstrates the utility of preferring empirically
supported principles over treatments. Problems and limitations of connectionist explanations are presented.
This explanation warrants further consideration and should stimulate discussion concerning empirically
supported principles.
D 2004 Elsevier Ltd. All rights reserved.
Keywords: Systematic desensitization; Exposure therapy; Learning-memory mechanism
* Tel./fax: +1 914 941 0632.
E-mail address: [email protected].
0272-7358/$ - see front matter D 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.cpr.2004.08.005
68
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
1. Possible mechanisms for why desensitization and exposure therapy work
Psychologists seem to agree, and our professional ethics require, that only reliable and valid
psychological tests should be used. The new ethics code states that bPsychologists use assessment
instruments whose validity and reliability have been established for use with members of the
population testedQ (APA, 2002, p. 1071). Less agreement exists regarding psychotherapy. The
extent to which interventions are to be based on and informed by scientific research is one of the
primary professional issues facing psychologists today. The effort to establish empirically supported
treatments (ESTs) represents an important attempt to identify interventions that meet minimal
scientific standards (cf. Sanderson, 2003) but has met with serious opposition (Hebert, 2003). One
reason for opposing ESTs is that they do not provide an explanatory context in which to place and
understand a client’s presenting problem. Theories of psychotherapy and personality have
proliferated to address this explanatory need. Absence of empirical support about these theories,
and many interventions based on them, does not diminish the need for an explanatory context.
Psychologists can opt not to administer tests if no reliable and valid test is available for a specific
purpose, but they cannot avoid the need to understand their client’s presenting problem and
consequently they turn to the theoretical tradition in which they were trained even if that position
has limited or no empirical evidence to support its validity or the effectiveness of interventions
based on it. The need to explain seems to take precedence over the desire for empirical support.
The substantial and generally recognized gap between the science and practice of clinical
psychology demonstrates that empirical evidence of outcome alone is insufficient to persuade
clinicians to use ESTs let alone limit their practice to them. Effectiveness of efforts to persuade
clinicians to adopt ESTs may depend substantially on the extent to which science can explain why
ESTs work and thereby provide clinicians with an empirically supported explanatory context in
addition to effective interventions. This is reason enough to engage the explanatory discussion
initiated here.
Rosen and Davison (2003) objected to the current practice of listing ESTs because adding one or
more functionally inert components to an existing intervention based on sound psychological science
can both meet EST requirements and be perceived as a new therapy. This creates two problems. First, yet
another bnewQ therapy appears to have been developed that in fact succeeds because its active
ingredients are those of an already established therapy. Second, causal attribution is frequently made to
the new elements. Rosen and Davison illustrated these problems with a hypothetical intervention called
bPurple Hat TherapyQ (PHT), where the therapist asks the client to wear a large purple hat while
receiving exposure therapy for a phobia. PHT will be more effective than a control treatment because it
entails exposure therapy but PHT proponents will attribute curative powers to wearing the hat and then
establish seminars and proprietary rights to training therapists in this new therapy. Rosen and Davison
cited Eye Movement Desensitization and Reprocessing (EMDR) as a PHT because: (a) EMDR is
superior to control conditions thereby establishing it as an EST, (b) causal attribution has been claimed
for the eye movement component despite a lack of evidence showing it to incrementally add to clinical
outcome, and (c) a burgeoning proprietary training system has also been developed to promote EMDR.
The potential for psychotherapies to proliferate without limit is clear. It is possible that the panoply
of present-day psychotherapies may exist for these reasons. Rosen and Davison proposed that
listing Empirically Supported Principles (ESPs) rather than ESTs may solve this problem and
return our attention to matters of explanation as well as prediction. Their recommendation is
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
69
consistent with Shapiro’s (1995) view that a major focus of psychotherapy research should
concern how effective therapies work. Rosen and Davison recognized that considerable
controversy might arise when attempting to identify and categorize psychological principles
but welcomed such debate. They concluded that such argument is useful because, bDebates that
concern principles of change, rather than specific trademarked therapies, will return our attention
to what psychology is aboutQ (p. 308).
Proctor and Capaldi (2001) discussed the complementary yet competing roles of explanation and
prediction in science. Philosophy and religion explain much but predict little that can be
empirically verified and possibly falsified. Hypothesis testing distinguishes science because
it makes empirically testable predictions. However, at some point the resulting facts need
to be placed into an explanatory context. Mature sciences provide explanatory theories in
addition to making testable predictions.1 Proctor and Capaldi noted that British geologists accepted
plate tectonic theory well before American geologists did because they valued explanation as well as
prediction whereas American geologists almost exclusively limited themselves to prediction and
hypothesis testing (cf. Oreskes, 1999). Plate tectonics placed existing geological data into a coherent
explanatory context and provided a plausible mechanism for how certain geological facts came to be.
Proctor and Capaldi implied that American psychologists might learn from the mistakes their geology
colleagues and temper their interest in hypothesis testing with explanatory efforts. This article evaluates
the extent to which various explanations of systematic desensitization and exposure therapy fit with
known facts.2 Novel prediction is not a primary or essential part of this process just as it was not for the
evaluation of plate tectonic theory.
Darwin’s theory of evolution provides further evidence of the importance science places on
explanation. Darwin’s (1859) evolutionary theory was limited to functional statements about the
causal role of variation and selection in the origin and extinction of species and was largely
rejected by the scientific community upon its publication because no plausible proximal causal
mechanism was proposed to explain how variation was instantiated and how selection could
work as described (Bowler, 1983). Darwin’s evolutionary theory remained on the margins of
biology for approximately 75 years until population genetics provided the missing causal
explanatory mechanisms (Mayr, 1982). Tryon (1993b, 2002a) identified parallels with behavioral
psychology. The cognitive revolution in psychology occurred partly because functional
statements made by behaviorists lacked causal mediating explanatory mechanisms. The search
for causal mediators and moderators is evidence that behavioral scientists value explanation in
addition to novel prediction. Systematic desensitization and exposure therapy appear to work but
there is little agreement as to why they work. This article considers several explanatory
possibilities and seeks empirically supported principles for why systematic desensitization and
exposure therapy work. These interventions were selected because they were among the first
empirically supported treatments and because attempts have been made to explain why they
work.
1
The dichotomy between hypothesis-testing and explanation is complicated because hypotheses are often derived from explanatory
theories.
2
The factual status of generalizations based mainly on analog studies can be questioned. The use of nonclinical populations is a major
threat to external validity. However, analogue studies do systematically vary one or more independent variables while controlling for some
confounding variables to reduce threats to internal validity.
70
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
2. Explanatory bases and possible principles
We now understand that systematic desensitization depends upon exposure but the two therapies
are treated separately here because that is how the explanatory literature is organized. Most
literature citations regarding the explanatory basis of systematic desensitization are old because
exposure therapy all but replaced it subsequent to Marks (1975) literature review and because
research has focused almost exclusively on outcome research, hypothesis testing, and novel
prediction since then. The relevant literature is enormous and space limitations require reliance on
empirical literature reviews by Borkovec (1972), Brown (1973), Davison and Wilson (1972, 1973),
Emmelkamp (1982, 1994, 2001), Goldfried (1971), Kazdin and Wilcoxon (1976), Kazdin and
Wilson (1978), McGlynn, Mealiea, and Landau (1981), Morgan (1973), Rachman (1965), Spiegler
and Guevremont (1993, pp. 211–214), Wilkins (1971, 1972, 1973a, 1973b), and Yates (1975)
regarding why systematic desensitization is effective. Evidence from single studies is discussed
were pertinent.
The term exposure therapy derives from Marks’s (1975) review of the systematic desensitization
literature where he concluded that mere exposure to aversive cues was as effective as systematic
desensitization. Taylor (2002) identified four categorizes of exposure therapy based on two
dimensions: (1) real vs. imagined stimuli and (2) gradual vs. intense exposure (cf. Table 1).
McGlynn et al. (1981) noted that b. . . exposure theory is not an explanation of therapeutic
desensitization effects. Rather, it is simply a hypothesis concerning the necessary and sufficient
procedural ingredients within the technique. The therapeutic effects of the exposure remain to be
explained (e.g., as extinction, as counterconditioning, as habituation)Q (p. 154). Van Egeren (1971)
organized four of the most common explanations of systematic desensitization into a 22 table (cf.
Yates, 1975, p. 165) presented here in modified form as Table 2 and we begin our analysis with
them.
2.1. Reciprocal inhibition
Wolpe (1958, 1995) explained the effectiveness of systematic desensitization using Sherrington’s
(1906/1961) physiological concept of reciprocal inhibition. Reciprocal inhibition has been used
both as a psychological and biological mechanism of action. The psychological mechanism is
based on the premise that two incompatible psychological states cannot occur simultaneously. The
definition of what it means to be relaxed excludes what it means to be anxious. One must be
careful to avoid a tautology here as it is possible to restrict instances of reciprocal inhibition to
only those situations where a strong negative correlation occurs on the basis that other conditions
do not meet the definition. The biological mechanism derives from the study of reflexes. A tap
on the patellar tendon elicits inhibition (relaxation) of the leg’s flexor muscles along with
Table 1
Four types of exposure therapy
Imagined stimuli
Real stimuli
Gradual
Intense
Systematic desensitization
Graded in vivo exposure
Implosion
Flooding
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
71
Table 2
Four theoretical explanations of systematic desensitization
Antagonistic inhibition present
Antagonistic inhibition absent
Short-term effects
Long-term effects
Reciprocal inhibition
Habituation
Counterconditioning
Extinction
excitation of extensor muscles; a coordination of active inhibition and excitation in different
muscles. Parasympathetic nervous activation is known to inhibit sympathetic nervous activation.
Jacobson’s (1938) deep muscle relaxation technique was the major method Wolpe used to
increase parasympathetic activity, which was hypothesized to inhibit sympathetic activity and
therefore anxiety. However, even temporary paralysis by curare of all muscles did not preclude or
prevent anxiety (Davison, 1966). While a few studies have provided empirical support for the
view that relaxation is a necessary component of systematic desensitization (Davison, 1968; Kass
& Gilner, 1974) others have not (Miller & Nawas, 1970; Nawas, Welsch, & Fishman, 1970).
Agras et al. (1971), Cooke (1968), Craighead (1973), Crowder and Thornton (1970), Freeling and
Shemberg (1970), and Waters, McDonald, and Koresko (1972) reported that phobic anxiety is
reduced whether or not relaxation training is used. It does not appear necessary to pair relaxation
with imagery during desensitization (Aponte & Aponte, 1971). Allowing patients to relax in
between hierarchy items does not appear to alter treatment outcome (Benjamin, Marks, &
Hudson, 1972). Pairing of relaxation with graded imagination seems not to alter treatment
outcome (McGlynn, 1973). Van Egeren (1971) and Van Egeren, Feather, and Hein (1971) found
little empirical support for the reciprocal inhibitory explanation of systematic desensitization. If
reciprocal inhibition was causally necessary then flooding (Miller, 2002) and implosive therapy
(Levis, 2002) should not work because they do not involve any reciprocally inhibitory agent, yet
the parenthetical references cited here indicate that they do. Yates (1975) concluded his literature
as follows: (a) bSystematic desensitization is effective in reducing phobic anxiety, whether
relaxation training is part of the program or notQ (p. 156), and (b) b. . . neither individualized
hierarchies nor any special way of presenting the hierarchies are critical to the success of
desensitizationQ (p. 158). Kazdin and Wilson (1978) concluded that none of the therapeutic
ingredients postulated by Wolpe were in fact necessary (p. 37). In sum, the preponderance of
empirical evidence does not support the principle of reciprocal inhibition by deep muscle
relaxation.
Sexual arousal and aggression are other states that putatively inhibit anxiety but therapeutic
procedures based on them have not been widely developed; probably for ethical and legal reasons.
Assertiveness training and coping skills may also inhibit anxiety to some degree but they do not
appear to totally inhibit anxiety in the same way that leg flexion inhibits extension. A further
difficulty is that Table 2 classifies reciprocal inhibition as having short-term effects whereas it is
supposed to have long-term benefits. An anonymous reviewer objected to this well-established
classification on the basis that Wolpe’s theorizing was Hullian at its core; that reciprocal inhibition
was a theoretical substitute for Hull’s reactive inhibition that built up conditioned inhibition, and
that reactive inhibition was a long-term rather than short-term process. The reviewer argued that
the permanence of conditioned inhibition makes relearning impossible and that creates a serious
explanatory problem.
72
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
2.2. Counterconditioning
Wolpe (1958) used Guthrie’s (1952) concept of counterconditioning to explain the long-term
therapeutic effects of systematic desensitization (cf. Davison, 1968). The central concept here is
replacement of an old response with a new one, such as when relaxation putatively replaces
anxiety. The counterconditioning theory of systematic desensitization effects presumes that
decreasing classically conditioned emotionality sets the occasion for reduced instrumental
avoidance of phobic stimuli (Bandura, 1969, pp. 424–425). Table 2 shows that counterconditioning is understood to exert long-term effects. Davison (1968) stated that counterconditioning is the behavioral equivalent of the neurological process of reciprocal inhibition.
Marks (1975) reviewed the empirical literature and concluded that systematic desensitization
with relaxation is no more effective than graded exposure. The term bexposure therapyQ stems
from this seminal article. Wilkins (1971) had earlier concluded, on the basis of his review of the
empirical literature, that imagination of fearful scenes was the necessary and sufficient ingredient
for successful desensitization. The empirical effectiveness of flooding (Miller, 2002), and
implosive therapy (Levis, 2002) also contradict the counterconditioning explanation of systematic
desensitization as previously mentioned because these forms of treatment are not thought to
replace one emotional state with another. Van Egeren (1971) and Van Egeren et al. found little
empirical support for the counterconditioning explanation of systematic desensitization. These
empirical results along with empirical reviews cited above undercut both the reciprocal inhibition
and counterconditioning explanations of systematic desensitization.
Moreover, McGlynn (2002) noted that Wolpe based his explanation on Hull’s concept of
habits and thereby injected the still unresolved theoretical debates among classical learning
theorists (Clark L. Hull, Edwin R. Guthrie, and Edward C. Tolman) into explanations of the
effectiveness of systematic desensitization (McGlynn et al., 1981). Bandura (1969, p. 431) noted
that the drive-reduction theory of classical conditioning and a fatigue theory of extinction favored
by Wolpe have so far been contradicted by empirical evidence. In sum, the balance of empirical
evidence does not support a counterconditioning explanation of systematic desensitization and the
unresolved theoretical debates among classical learning theorists further diminish the adequacy of
this explanation.
2.3. Habituation
Harris (1943) based his comprehensive review of the early habituation literature on the
following operational definition: bresponse decrement due to repeated stimulationQ (p. 385). Lader
and Mathews (1968) proposed a habituation explanation of systematic desensitization based on
this entirely functional definition of habituation. Emmelkamp and Felten (1985) reported
supportive evidence in that they found that subjective anxiety and physiological arousal
decreased for patients with specific phobias during in vivo exposure. However, Van Egeren
(1971) and Van Egeren et al. found little empirical support for the habituation explanation of
systematic desensitization.
Watts (1979) based his review of the empirical evidence about the habituation explanation of
systematic desensitization on the same functional definition of habituation provided above and
concluded that sufficient supportive evidence existed to make the habituation model a viable
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
73
alternative to the reciprocal inhibition model. However, Thompson and Spencer (1966) identified
nine habituation phenomena that are so characteristic of habituation that they need to be present
in order to conclude that habituation is present. This has not been done regarding systematic
desensitization or exposure therapy. A response decrement is required (cf. Harris, 1943). The
habituated response must recover if the stimulus is withheld. Presentation of another, usually
strong, stimulus must cause dishabituation resulting in an increase of response strength.
Hergenhahn and Olson (1993, pp. 4, 12) and Kalat (2003, pp. 383384) noted that habituation
generally entails a short-term response reduction given repeated stimulation. Van Egeren (1971)
and Yates (1975, p. 165) both recognized the short-term effects of habituation. However, massed
habituation trials can result in habituation effects that last approximately 3 weeks (Kandel, 1991).
But sensitization, another short-term process, can restore defensive responses and reverse
habituation (Hergenhan & Olson, 1993, p. 4; Kandel, 1991). In sum, the temporary and
reversible nature of habituation makes it unable to explain durable long-term changes in response
strength.
2.4. Extinction
Extinction entails the lack of onset or offset of stimuli with positive or negative reinforcing
properties contingent upon either the emission or omission of a response. Response decrement is
explained in terms of the lack of reinforcement. Waters et al. (1972) explained the reduction of
phobic behavior subsequent to systematic desensitization in terms of extinction. Marks’s (1975)
concluded on the basis of a review of the empirical literature that exposure to the fearful stimuli
is the only necessary and sufficient condition for anxiety reduction. Emmelkamp (1994) also
concluded, based on a review of the empirical literature that exposure to phobic stimuli without
avoidance is the essential ingredient in effective behavioral treatment for anxiety disorders.
Exposure or exposure plus nonavoidance are partially consistent with an extinction explanation
because a complete extinction explanation needs to: (1) define the target behavior, (2) define the
reinforcer, and (3) show that no onset or offset of the reinforcer occur contingent upon either the
emission or omission of the target behavior. The empirical literature does not strongly support the
third criteria and only partially supports the other two.
More importantly, extinction refers to a functional relationship between response decrement and
absence of reinforcement; it does not explain why this relationship holds, and therefore cannot be
used to explain fear reduction. Behavioral psychologists have intentionally avoided investigation
into processes underlying extinction because formal behavioral explanations are restricted to
functional statements that exclude mediating processes (cf. Plaud & Eifert, 1998; Plaud &
Vogelantz, 1997). An extinction explanation of systematic desensitization must show that the
same mechanism underlies both extinction and systematic desensitization. No such explanation
can presently be provided and consequently systematic desensitization cannot presently be
explained in terms of extinction. The lack of a causal mechanism deprives extinction of
explanatory force.
Bandura (1969) observed that b. . . conventional extinction procedures often involve a form of
unguided counterconditioning b(pp. 429–430) because other stimuli co-occur naturally during
extinction trials. Concurrent cognitions are inevitably associated with, and therefore inextricably
confounded with, extinction procedures in humans. It may therefore not be possible to conduct a
74
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
study that avoids this problem. This would preclude establishing extinction as an explanation of
systematic desensitization and exposure therapy in humans. Mahoney (1974) and Meichenbaum
(1974) exploited the fact that people are cognitively active during desensitization by having
participants make corrective self-statements and visualize themselves engaged in adaptive coping
behaviors during desensitization to enhance its effectiveness. The argument that extinction is an
active learning process and therefore consistent with cognitive processing converts it into a
cognitive-behavioral explanation because strict behavioral explanations exclude mediating cognitive
processes from formal explanations of behavior (cf. Plaud & Eifert, 1998; Plaud & Vogelantz,
1997). The current cognitive-behavioral debate (cf. Tryon 1995b, 1996) continues to center on
whether or not to permit mediational processes such as cognition as legitimate scientific
explanations of behavior. Presence and effectiveness of concurrent cognitive processes undercuts
the extinction explanation in humans. Interpreting extinction as counterconditioning is undercut by
the same facts that vitiated the counterconditioning hypothesis reviewed above.
Wolpe (1995) criticized the extinction explanation on two other grounds. First, repeatedly
evoking a fear response in the absence of reinforcement was, in his experience, insufficient to
permanently reduce anxiety. He claimed that the natural histories of phobic patients often include
frequent exposure to fear stimuli with no anxiety reduction. It is possible that these natural
histories also entail avoidance while still anxious followed by subsequent anxiety reduction
thereby reinforcing avoidance and preventing desensitization. Second, Wolpe criticized exposure
explanations for lacking a causal fear reduction mechanism such as reciprocal inhibition.
Evidence presented above constitutes serious empirical and theoretical weaknesses regarding the
extinction explanation of systematic desensitization. The explanatory limitations noted above
suggest that obtaining methodologically rigorous empirical support for the extinction hypothesis in
humans is unlikely.
2.5. Two-factor model
Mowrer’s (1960) two-factor theory is frequently cited as an explanatory basis for exposure
therapies (McAllister & McAllister, 1995; McGlynn, 2002). Mowrer proposed that fears are
acquired according to classical conditioning and are maintained by fear reduction that comes
from escape and avoidance of the phobic object. Menzies and Clarke (1995) critically reviewed
the evidence for and against the traumatic associative conditioning hypothesis regarding the
etiology of phobias and found that the preponderance of the evidence did not support it. The
crucial empirical fact is that few phobic persons can recall a traumatic onset of their phobia. A
patient’s inability to recall relevant trauma does not prove that traumatic events never occurred
but neither does it provide empirical support for the traumatic conditioning etiology of phobias.
The burden of proof remains with the proponents of hypotheses; not with the opponents.
Kheriaty, Kleinknecht, and Hyman (1999) surveyed undergraduate students with blood/injection or
dog phobias regarding traumatic memories using either the Phobia Origins Questionnaire (POQ:
Öst & Hugdahl, 1981) or Phobia Origins Structured Interview (POSI: Kleinknecht, 1994). While
over a quarter (28.6%) of their respondents had no memory for traumatic events the following
evidence of method dependency was reported; 46.0% of respondents had no recall when
examined by structured interview but only 7.3% had no recall when examined by questionnaire.
While 71.8% of respondents to the POQ reported conditioning-like events, 71.8% also reported
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
75
vicarious events (e.g., observing trauma in others) and 84.6% reported information events
pertinent to the origin of their phobia. The impact of each type of experience on the intensity of
current phobia remains unknown. For example, Kheriaty et al. (1999) did not demonstrate that
phobic severity was proportional to the severity of conditioning-like events. Kheriaty et al.’s
participants comprised an analog rather than a clinical sample. Perhaps, the same method
dependency exists in clinical samples. Or, recall could be inversely related to severity of
impairment and therefore better in an analog than clinical sample.
A theory may incorrectly explain etiology but correctly explain intervention effects. Anxiety
may develop for reasons other than classical conditioning but once established, anxiety reduction
produced by escape/avoidance conditioning may strengthen fear-related behaviors and emotions.
The empirical evidence first summarized by Marks (1975) indicates that exposure to fear stimuli
and nonavoidance are crucial to fear reduction. One might refer to this as the Exposure and
Nonavoidance empirically supported principle. However, McGlynn et al. (1981) noted that
b. . . exposure theory is not an explanation of therapeutic desensitization effects. Rather, it is
simply a hypothesis concerning the necessary and sufficient procedural ingredients within the
technique. The therapeutic effects of the exposure remain to be explained (e.g., as extinction, as
counterconditioning, as habituation)Q (p. 154). We have already seen that there is little supportive
evidence for these explanations.
2.6. Cognitive changes
Patients are conscious during systematic desensitization and are therefore likely to actively
construe this experience. The following cognitive-behavioral explanations of systematic
desensitization constitute the more popular cognitive-behavioral explanations of systematic
desensitization and exposure therapy. Modeling is not discussed because the more common
therapies do not call for the therapist to model desired behaviors.
2.6.1. Expectation
The psychological principle that persuasion is part of healing has long been accepted (Frank,
1961). However, bOnce it has been established that a placebo intervention is better than no
treatment for a particular problem, then a principle of change has to perform better than placebo
to receive separate recognitionQ (Rosen & Davison, 2003, p. 307). This criterion appears to
distinguish placebo from psychological principles though one might argue that placebo is a
primary psychological principle. Baskin, Tierney, Minami, and Wampold (2003) reported that
structurally equivalent placebos are nearly as effective as treatments. The structural factors
considered were: (a) the number of sessions, (b) length of sessions, (c) format, group vs.
individual, (d) therapist training, (e) whether or not treatment was individualized to the client,
and (f) whether therapists could discuss topics pertinent to the presenting problem or if they
were restricted to neutral topics. Structurally equivalent placebo groups differed from experimental
groups by only 0.149 pooled standard deviations on average; the 95% confidence interval ranged
from 0.055 to 0.292. Placebo interventions that were not structurally equivalent differed from the
experimental groups by 0.465 pooled standard deviations on average; the 95% confidence interval
ranged from 0.309 to 0.621. Structurally equivalent placebos appear to be more credible,
engender greater expectation for change, and to produce larger therapeutic effects.
76
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
Wilkins (1971) argued that systematic desensitization works because: (1) therapists foster an
expectation of success, (2) progress up the hierarchy provides confirmatory feedback that success is
occurring, and (3) the patient learns how to control the onset and offset of fearful imagery.
Borkovec (1973) identified nine studies supporting the view that expectancy influenced outcome
and ten studies that did not. Yates (1975, p. 170) also reviewed the literature and found similarly
mixed support. Wilkins (1973b) reviewed the empirical literature regarding expectancy of
therapeutic gain regarding systematic desensitization and concluded that the expectancy explanation
of why systematic desensitization works is unsupported by empirical evidence. On the other hand,
Kazdin and Wilcoxon’s (1976) review of the empirical literature revealed that the large majority of
studies failed to use control groups that were as credible and generated as much expectation of
change as did the systematic desensitization conditions. Systematic desensitization and exposure
therapy may work to some extent because of persuasion but the degree to which this is true is
unknown at this time. Nor has a mechanism of action been offered for how expectation leads to
long-lasting behavioral change. Absence of such an explanatory mechanism deprives expectation of
explanatory force.
A review of the literature reveals that patients with panic disorder and phobias are prone to
over estimate how fearful they will be when exposed to a fear stimulus (Rachman & Bichard,
1988); they expect to be more fearful than they actually are when exposed to the phobic
stimulus. Taylor and Rachman (1994) explained this phenomenon as due to: (a) the over
prediction of danger elements, and (b) the under prediction of safety resources. Exposure therapy
is hypothesized to work because it presumably provides corrective evidence in that participants
see that reality is not as bad as their expectations. This explanation is known as stimulus
estimation theory or match–mismatch theory. Two subsequent studies did not support this
explanation (Arntz, Hildebrand, & van den Hout, 1994; Telch, Valentiner, & Bolte, 1994) but
were criticized by Taylor (1995) on the methodological grounds that they did not obtain both
predictions and reports of danger and safety and they relied on correlational rather than
experimental manipulation. An experimental test of this hypothesis by Wright, Holborn, and
Rezutek (2002) that obtained predictions and reports of danger and safety in snake-fearful
university students provided empirical support for this hypothesis. Effectiveness of systematic
desensitization and exposure therapy can therefore be explained by a consonance seeking process
where fear expectations decrease so as to be more consistent with experience. Mechanism
information about how this learning takes place has not been provided. No explanation has been
given for how exposure brings about lowered fear. The authors imply that some form of
comparison between imagined and experienced states is involved and that this discrepancy
somehow reduces fear. The absence of mechanism information deprives the stimulus estimation
explanation of explanatory force. Another explanatory limitation of the stimulus estimation
explanation is that systematic desensitization and exposure therapy should not reduce fearful
expectations below the fear level experienced in the presence of the fear stimulus yet exposure
therapy frequently appears to reduce fear to lower levels.
2.6.2. Self-efficacy
The coping procedures initiated by Meichenbaum (1974), Mahoney (1974), and others were
designed to promote the kind of cognitive changes that subsequently have been termed selfefficacy; a positive view of one’s ability to cope. Fear reduction can be explained as the result
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
77
of increases in self-efficacy (Bandura, 1977, 1978, 1982, 1998). Systematic desensitization is said
to alter both the response-outcome and efficacy expectation components of self-efficacy. It is
understandable how hierarchy construction from easier to more difficult scenes could improve
one’s response-outcome expectation in favor of expecting that they should eventually be able to
remain calm in the face of fearful cues by the end of treatment. It is also understandable that
successful progression up the hierarchy provides evidence that one is systematically approaching
this clinical goal. What is missing is mechanism information to explain how fear reduction takes
place. How does expectation of fear reduction actually reduce fear? How is it that the individual
can progress up through the hierarchy and thereby validate the new response-outcome
expectation? If the person did not actually become less fearful, then imaginal or actual exposure
to the fearful cues would reconfirm their phobic reaction and reaffirm the original responseoutcome expectation, that they will respond anxiously, and reinforce their original efficacy
expectation that they cannot approach what makes them anxious. Absence of causal mechanism
information regarding self-efficacy changes deprives it of explanatory force.
2.6.3. Cognitive restructuring
The underlying psychological principle for cognitive restructuring appears to be that one
behaves and feels in ways that are consistent with what one thinks and that by changing
cognitions one necessarily changes behaviors and emotions. Rational-emotive behavior therapy
(Ellis & Blau, 1998; Ellis & Whiteley, 1979) and cognitive therapy (Beck, 1976, 1995) aim to
reduce anxiety by restructuring cognitive appraisals. An effort is made to help the person
understand that the phobic object or situation is really not dangerous. Such altered cognitions
may reduce anxiety or persons who become less anxious as a result of systematic desensitization
may think more rationally about their fears. Persons who do not improve as a result of
systematic desensitization may not think more rationally about their fears and therefore may not
be less anxious. Data as to the temporal change sequence and whether cognitions change before
anxiety levels drop or vice versa are needed to decide this matter.
DeRubeis et al. (1990) used Baron and Kenny’s (1986) criteria for establishing mediation as the
methodological basis for determining if changed cognitions mediated the effects of cognitive
therapy. They were unable to satisfy all of Baron and Kenny’s criteria for mediation and
were therefore unable to clearly establish cognitive mediation. These authors did not provide
any causal mechanism to explain how cognitive factors changed emotion and behavior
despite their title bHow does cognitive therapy work?Q thereby failing to make their
explanatory case.
2.6.4. Emotional processing models
Lang (1977, 1979, 1985) and Drobes and Lang (1995) explained the cognitive and emotional
changes produced by systematic desensitization and exposure therapy using an informal cognitiveemotional-behavioral network theory described as a bioinformational model. These authors
attempted to move beyond the cognitive information processing metaphor by hypothesizing that
fear is mediated by a memory-based network containing information about stimulus characteristics, verbal and nonverbal response tendencies, feelings, and propositions about the meaning of
these events in different situations. They hypothesized that each information source constitutes a
network node. Excitatory and inhibitory connections were hypothesized to exist among these
78
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
nodes and to control behavior and the psychophysiological responses fearful people made while
reading scripts of their traumatic experiences.
Foa and Kozak (1986) proposed that fears entail memory-based networks of associations called
fear structures that integrate perceptual, cognitive, and behavioral tendencies. Activation of a fear
network by a perception is said to motivate avoidance and escape behavior. Foa and Kozak
(1986) and Rachman (1980, 1990) discussed the therapeutic concept of bemotional processingQ
defined as the modification of memory-based fear structures associated with thoughts, feelings, and
actions. A gradual reduction in emotional responding over time is expected given (a) repeated
activation of the fear network and (b) incorporation of corrective fear-incongruent information into
the network thereby revising traumatic memories. This is a more active process than extinction
(Forgas, 1999). The author’s nonstandard use of the term habituation incorrectly implies
cumulative stable long-term fear reduction. The nine criteria for habituation established by
Thompson and Spencer (1966) were not met thereby failing to establish a habituation explanation.
Moreover, the reversible nature of habituation documented above precludes explanation of stable
long-term outcomes.
Creamer, Burgess, and Pattison (1992) postulated a similar form of bnetwork resolution
processingQ. Fear networks were presumed to vary in size, structure (interconnectedness), and
accessibility (cf. Foa, Steketee, & Rothbaum, 1989). Chemtob, Roiblat, Hamada, Carlson, and
Twentyman (1988) discussed a four-level cognitive schema network wherein one level influenced
all others through bspreading activationQ thereby interrelating thoughts, feelings, and actions.
All of these informal network theories advanced our understanding because they provided a network
mechanism with causal properties sufficient account for observed functional relationships. However, the
informal presentation of these network theories limits their usefulness. It is not possible to predict
network behavior, and therefore empirically test these theories without further details regarding network
structure, how each node influences all others, and how connections among network nodes change as a
result of learning. Absence of such specifics precludes the possibility of altering these network models
based on empirical research. Formal network theory, discussed below, has made progress regarding all of
these issues and provides a basis for understanding how systematic desensitization and exposure therapy
work.
3. Empirically supported mechanisms
Parallel Distributed Processing Connectionist Neural Network (PDP-CNN) models are memory
mechanisms that learn. Both learning and memory are driven by experience which makes both
processes dependent upon sensation and perception. Hence, action mechanisms associated with the
processes of learning, memory, sensation, and perception are pertinent to our understanding of
how systematic desensitization and exposure therapy work. Some information regarding
empirically supported PDP-CNN mechanisms is needed prior to presenting a connectionist
explanation of how systematic desensitization and exposure therapy work. The first section below
establishes that learning and memory share much in common. The second section documents that
learning entails synaptic change; modeled by modifying connection weights. These changes alter
how the network functions and therefore how psychological and behavioral variables change. The
third section illustrates how activation can cascade across a network. The fourth section discusses
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
79
constraints on the learning process and their satisfaction. The fifth section discusses the role of
consonance in network processing. A short list of successful simulations is provided and then a
connectionist explanation of how systematic desensitization and exposure therapy is given.
3.1. Learning-memory
Learning is perhaps the most generic psychological principle and has received an enormous amount of
empirical support from highly controlled laboratory conditions in a wide variety of species regarding a
broad range of behaviors. Our educational system is based upon the principle that people learn. Tryon
(2000) observed that all psychological theories of psychopathology maintain that learning plays an
etiological role in psychological and behavioral disorders and that all persons who benefit from
psychological interventions learn something. That psychologists differ on what is learned and how best
to teach should not obscure our common view that learning is a fundamental empirically supported
principle. All specific conditions under which learning has been demonstrated to occur constitute a list of
necessary and sufficient conditions that psychologists can use to promote therapeutic change.
Cumulative learning presumes retention and that implies some form of memory. Conversely,
memories are learned in that they are derived from experience. Many of the biological structures
necessary for learning are shared with those necessary to form memory. The interdependence of
learning and memory supports reference to a learning-memory principle. The next section
documents that neuroscience has made considerable progress in understanding the mechanisms
that enable learning and memory formation. Cognitive scientists who sought to understand the
bmicrostructure of cognitionQ (e.g., Rumelhart & McClelland, 1986; McClelland, & Rumelhart,
1987) explored parallel distributed processing in artificial neural networks and fostered the
development of formal network learning theory. These and related developments form the basis
for the explanation of systematic desensitization and exposure therapy provided below.
3.2. Learning entails synaptic change
Hebb (1949) hypothesized, and neuroscience subsequently confirmed (Bottjer & Arnold, 1997;
Gluck, Meeter, & Myers, 2003; Kalat, 2001; Kandel, 1991; Kolb & Whishaw, 1998; Krasne,
2002; Martin, Grimwood, & Morris, 2000; Packard & Knowlton, 2002), that learning entails
synaptic change; i.e., brain plasticity. Donahoe and Palmer (1994; pp. 66–67, Note 70; Spitzer,
1993, pp. 42–51), and Rolls and Treves (1998, pp. 322–325) briefly summarize the main cellular
mechanisms responsible for the long-lasting synaptic changes associated with learning. Krasne
(2002), Lynch (2000), Martin, Bartsch, Bailey, and Kandel (2000), and Matzel (2002) provide
more detailed accounts. PDP-CNN models use learning equations to simulate synaptic change
across learning trials. For example, the Hebbian learning equation produces an updated
connection weight equal to the old weight plus the product of the activations of the two
neurons connected by the synapse being altered times a learning rate parameter. This method is
also used to create associative memories (cf. Rolls & Treves, 1998, pp. 42–53). On this view,
learning and memory are two facets of a learning-memory mechanism that appears to be
common to most, if not all, learning including learning due to reinforcement, habituation, and
extinction. That the functional properties of PDP-CNN models depend upon synaptic, connection
weight, changes and resulting activation levels of network nodes has been empirically well
80
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
Fig. 1. A hypothetical three-layered feed forward network. The top bSQ layer receives stimulus input. The bhiddenQ middle bOQ
layer forms concepts. The bottom bRQ layer represents behavior.
established. Details are provided in introductory texts, such as Bechtel and Abrahamsen (2002),
Martindale (1991), McLeod, Plunkett, and Rolls (1998), O’Reilly and Munakata (2000), and
Spitzer (1999).
3.3. Network architecture and cascade
Synaptic change implies neural network architecture. A PDP-CNN explanatory advance over the
emotional processing models reviewed above is that they provide a detailed description of the network
architecture associated with each model rather than rely on a generic network metaphor. Fig. 1 depicts a
simple feed forward network3 comprised of three layers of nodes, illustrated as circles, and two layers of
connections (synapses), denoted as solid lines. Note that each node in one layer is connected to all nodes
in the next layer and that no connections exist between the top and bottom layers. These connections can
be excitatory, modeled with positive weights, or inhibitory, modeled with negative weights. This
diagram is admittedly a gross over simplification of any real neural structure. However, there is
precedent in science for beginning with simple preparations as initial models of complex phenomena.
Computer simulations based on such simple network structures appear to implement enough by way of
fundamental neuroscience to enable informative simulations of many behavioral and psychological
phenomena. A brief description of how processing cascades across, and is transformed by, each layer of
synapses, connection weights, is prerequisite to understanding the explanation of how systematic
desensitization and exposure therapy work provided below.
Fig. 2 illustrates how activation cascades across the network. Perception, not diagramed here, has
excited the first, third and fifth neurons in the top layer of input nodes as illustrated by the filled circles;
the unfilled circles remain inactive. All connections from inactive neurons have been omitted for clarity
because they do not contribute to cascade processing. This illustration pertains to a single processing
cycle. Therefore, the designated connection, synaptic, weights of 0.1, 0.2, and 0.3 for the first neuron,
0.2, 0.5, and 0.3 for the middle neuron, and 0.4, 0.2, and 0.1 for the last neuron are presumed to have
3
The layers of a network can refer either to layers of neurons (nodes) or synapses (connections). The terms neuron and synapse imply a
biological model whereas the terms nodes and connection weights imply a psychological model. These terms are used interchangeably in this
article. Fig. 1 has three layers of neurons and two layers of synapses and can therefore be described as a three- or two-layered network,
respectively. Fig. 1 is described as a three-layered network to emphasize the different functions performed by each of the three layers of nodes.
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
81
Fig. 2. A numerical example of cascade processing.
resulted from prior learning. The activation level of the left neuron in the second, sometimes called
hidden, layer equals its current activation level, not shown, plus the sum of the connection weights4
coming from the active neurons leading into it. The net input now becomes (0.1)+( 0.2)+(0.4)=0.3.
Ignoring the prior level of activation for simplicity and assuming a threshold function such that
cumulative activation greater than zero excites the neuron, this node fires as designated by the filled
circle. Likewise, the sum of the connection weights leading into the middle neuron in the middle layer of
( 2)+( 5)+(2)= 0.5 is less than our zero threshold and therefore this node does not fire as designated
by the unfilled circle. The sum of the connection weights leading into the right neuron in the middle
layer of (0.3)+(0.3)+(0.1)=0.7 exceeds zero and excites this node as designated by the filled circle.
The cascade from the middle to lower layer continues as follows. The connection weights associated
with the active neurons in the middle layer feeding into the left neuron on the bottom level are
(0.5)+( 0.4)=0.1 which exceeds zero and therefore causing it to fire. Notice that the connection weight of
0.2 associated with the second neuron in the middle layer is not involved because that neuron was not
active. The connection weights associated with the active neurons in the middle layer feeding into the right
neuron on the bottom level are (0.3)+( 0.6)= 0.3 which is less than our zero threshold causing this node
to stay off. Hence, the binary input pattern of 1, 0, 1, 0, 1 is transformed into an output pattern of 1, 0.
Transformation is central to cognitive theory. The cascade process is one way that PDP-CNN
networks transform their inputs. The learning-driven connection weight changes noted above alter the
transformation performed as the sign and magnitude of the connection weights change across the
network. Geometric and mathematical parallels with factor analysis provide a way to understand how
and why transformation occurs and why it is dependent upon the sign and magnitude of connection
weights. Consider Fig. 1 from the perspective of factor analysis. Let each circle in the bottom level
represent a single test item and visualize as many circles as the test has items. Let each circle in the
middle level represents an orthogonal factor5 from a factor analysis of those items. Visualize as many
circles as the test has factors. Let the lines connecting the factors to the items constitute factor loadings.
Just as the meaning of each factor depends on the size and sign of each factor loading, the meaning
4
Each active neuron, one that fires, is coded 1 and each inactive neuron is coded 0. The activation coming into the left neuron on the
middle level from all three active neurons on the top level is technically (1)(0.1)+(0)(?)+(1)( 0.2)+(0)(?)+(1)(0.3)=0.3. The question marks
refer to the undefined and not shown connections from inactive neurons. These connection weights are unimportant because zero times any
number is zero.
5
Correlated, also known as oblique, factors can be represented if horizontal connections are allowed among the middle nodes. Strong
parallels exist between structural equation models and neural network architectures.
82
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
associated with each node depends on the size and sign of each connection weight.6 Likewise, just as
changing the size and sign of factor loadings changes the concept represented by the factor, so changing
the size and sign of connection weights changes the concept represented by each node in the middle
layer. Let each circle in the top level represents a second-order factor. Visualize fewer circles here than in
the middle layer. Each second-order factor loads on, is connected to, all of the primary factors and forms
a higher-order construct. Its meaning depends upon the size and sign of connections with nodes in the
middle layer.
Processing in Fig. 1 occurs from top down so we need to invert this example and let the stimulus
features correspond to test items, retain our understanding of the nodes in the middle layer as factors, and
consider the bottom layer of nodes as higher order factors. This parallel with factor analysis means that
each network layer transforms input and creates higher-order concepts like factor analysis and higherorder factor analysis do. Factor loadings provide differential emphasis and in this sense transform the
meaning of item content. Connection weights likewise provide differential emphasis of stimulus
characteristics and consequently determine what the network bthinksQ and bfeelsQ7 about the stimulus
events. Factor loadings are ordinarily computed once after all data are collected. Connection weights can
change after each processing cycle thereby sequentially altering meaning and giving rise to
psychological and behavioral development.
The ability to reduce network processing explanations to physical/chemical processes integrates
psychology into the mature sciences of biology, chemistry, and physics thereby advancing theoretical
consilience (Wilson, 1998). However, network explanations appear to be appropriate psychological
explanations and this presentation is limited to them.
3.4. Constraint satisfaction
Constraints, by definition, limit, restrict, or restrain. Consider a traveling salesman who must visit
several cities and wishes to do so by traveling the fewest miles. The requirement to minimize miles
traveled constrains what route will be taken. This is a single constraint problem. The decision to
purchase a car may entail multiple constraints including: (1) how much money one is willing to pay, (2)
how much performance is important, (3) how much gas mileage is desired, and (4) appearance. Each
facet and the importance placed upon it constrain the choice made. Linear regression positions a straight
line through a data set constrained by the requirement that it minimizes the sum of squared deviations,
vertical distances, between data points and the prediction line. Beliefs and attitudes constituted
constraints in Schultz and Lepper’s (1996) cognitive dissonance reduction network models. PDP-CNNs
attempt to reach the best possible solution by satisfying as many constraints as possible proportional to
their importance. McLeod et al. (1998) stated: bA system which works by constraint satisfaction has a
number of desirable characteristics for modeling human cognition. The main one is that it allows a
decision to be reached by a consensus of evidence, a reasonable fit between input and memory, rather
than requiring an exact matchQ (p. 46). Network nodes usually represent features and connection weights
represent their importance. The connection weights constitute constraints because a positive connection
tends to place the receiving node in the same state as the sending node whereas a negative connection
6
The mathematics used to compute a factor score is also used to compute what is called netinput to every node in the middle and bottom
layers.
7
Tryon (1999) discusses how emotions can be encoded into connectionist models.
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
83
tends to place the receiving node in the opposite state as the sending node. The actual state of the
receiving node depends upon whether or not the cumulative input of a receiving node exceeds a
threshold level as noted above. Learning entails changing size and sometimes the sign of connection
weights and that changes the constraints during the next processing cycle. Learning functions typically
change connection weights in a way that satisfies as many competing interests as possible and allows the
network to settle into a final activation state that constitutes the best possible compromise solution to the
problem at hand.
3.5. Consonance
Heider (1946, 1958) proposed that consonance is an organizing cognitive principle. Festinger (1957)
described consonance seeking as dissonance reduction. Abelson et al. (1968) edited an 84-chapter
volume entitled bTheories of Cognitive Consistency. Cognitive explanations presume consistency
between cognition and action. For example, cognitive restructuring is based on the view that one can
alter behavior by changing cognition. Changed behavior is explained in terms of changed cognition.
Shultz and Lepper (1996) speculated that bthe study of cognitive consistency seems to have fallen out
of favor, perhaps in part because of an inability to further penetrate its underlying reasoning
mechanismsQ (p. 219). Using connectionist network learning theory, they successfully simulated
important findings of the two major fields of dissonance research, insufficient justification and free
choice. Read and Miller (1998) reviewed connectionist models of social reasoning and social behavior.
Thagard (2000) summarized empirical evidence that coherence pertains to both thought and action.
The process of constraint satisfaction promotes consonance between external environmental
stimulation and current activation states of network nodes. Learning driven changes in connection
weights promotes consonance. More highly valued connections take longer and are more difficult to
change than are less valued connections. Each network iteration attempts to satisfy more constraints and
the process continues until the network stabilizes; i.e., further progress cannot be made. Network output,
behavior, is consequently the result of a consensus process that emphasizes goodness of fit between
current and cumulative prior experience. The resulting compromise provides network models with
clinical relevance.
3.6. Successful simulations
Formal PDP-CNN models have successfully simulated many behavioral (Commons, Grossberg, &
Staddon, 1991; Krasne, 2002) and psychological phenomena (Bechtel & Abrahamsen, 2002; McLeod
et al., 1998). Tryon (1995a) presented an elementary introduction to this field and discussed its
relevance to behavior therapists. Bechtel and Abrahamsen (2002), Martindale (1991), McLeod et al.
(1998), O’Reilly and Munakata (2000), and Spitzer (1999) present more extensive introductions to this
field and successful simulations.
3.7. Explanation of desensitization and exposure therapy
The schematic presentation of a simple three-layer PDP-CNN feed-forward network provided in Fig.
1 is sufficient for a general explanation of how systematic desensitization and exposure therapy can
work.
84
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
3.7.1. Pretreatment
People have a long developmental history by the time they seek treatment. Fig. 1 constitutes a
snap shot at a particular point in time just prior to intervention. The top layer of nodes represents
the results of sensation and perception; output from one or more additional networks not shown.
The middle layer of nodes represents one or more facets of cognitions and emotions7 present at the
time the snap shot was taken. The bottom layer of nodes represents one or more facets of the
pretreatment behavioral repertoire at the time the snap shot was taken. The cascade process described
above mediates, explains, how perception gives rise to fear-related, and other, cognitions, emotions, and
behaviors at pre-treatment. The network is in a consonant state because it has gone through many
learning cycles and has stabilized into its present configuration.
3.7.2. Treatment
The therapist creates dissonance by having the person behave in a therapeutic way, e.g.,
remaining relaxed or engaging in the desired behavior, while presenting a fear stimulus. The
stimulus presentation activates the pre-treatment cascade which now conflicts with the state of the
output nodes. Technically, one is said to bclampQ output node activation levels to represent the
desired response while applying the fear stimulus. How this dissonant network state is created is
unimportant from an explanatory perspective but may constitute crucial differences in clinical
technique depending upon the person being treated. Therapeutic change occurs because networks
seek consonance through an iterative gradient descent constraint satisfaction process wherein the
learning process modifies connection (synaptic) weights. Each exposure therapy trial repeats the
process of dissonance formation followed by consonance seeking wherein connection (synaptic)
weights become increasingly consistent with desired behavior as represented by the output layer.
The goal is to change connection weights across the network so that the stimulus cascades into the
desired therapeutic response rather than the pretreatment fear response. Successful cascade changes
causally modify mediating cognitions and emotions as described above.
An important novel prediction is that cognitions, emotions, and behaviors change simultaneously
during each processing cycle. This prediction of simultaneous change contrasts markedly with
sequential expectations by: (a) cognitive models which predict that cognitive changes precede and
mediate behavioral and affective changes, (b) behavioral models which predict that behavior
changes precede and mediate cognitive, and affective changes, and (c) affective models which
predict that emotional changes precede and mediate cognitive and behavioral changes. Simultaneous change integrates, and thereby unifies, all three standard models. Further implications are
discussed in Answering clinical questions below.
A treatment corollary of the simultaneity prediction is that interventions should simultaneously
foster cognitive, affective, and behavioral consistency with the desired outcome. Methods of getting
people to think, feel, and act in new ways that are consistent with the desired outcome should be
integrated, pursued simultaneously, in order to promote the physical processes of synaptic change
that alters the real neural network cascade that causally mediates changed perception, cognition,
affect, and behavior.
3.7.3. Principles vs. treatments
The first section of this article argued that certifying ESTs enables the proliferation of seemingly
different clinical interventions whereas ESPs should not. The above section showing that systematic
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
85
desensitization and exposure therapy both work because they alter the network cascade process
shows that they work for the same reason, are based upon the same principle, and therefore
constitute a single treatment, not two different ones. Multiple methods for creating dissonance may
constitute different clinical techniques but do not constitute different therapies because they
implement the same empirically supported causal mechanism.
4. Strengths of connectionist explanations
4.1. Empirical support from neuroscience
4.1.1. Synaptic change
The network model presented above is empirically supported by evidence from neuroscience
laboratories that has clearly demonstrated that learning and memory entail synaptic change; i.e.,
brain placticity. Connectionist models are memory systems that learn.
4.1.2. Neuroimaging studies
Kandel (1991) hypothesized that the biological changes produced by learning should be
detectable with modern neuroimaging equipment. Baxter et al. (1992) reported PET results showing
that nine OCD patients treated with CBT for 10 weeks demonstrated the same glucose metabolic
rate decreases as nine patients treated for 10 weeks with oral fluoxetine hydrochloride. Schwartz,
Stoessel, Baxter, Martin, and Phelps (1996) reported a significantly greater bilateral decrease in
caudate glucose metabolic rates for OCD responders than nonresponders to CBT. It is therefore
possible that systematic desensitization and exposure therapy also produce detectable brain changes.
4.2. Empirical support from psychology
The connectionist explanation of systematic desensitization and exposure therapy presented
above is consistent with and is empirically supported by the evidence supporting Mowrer’s (1960)
two-factor theory, Taylor and Rachman’s (1994) match–mismatch model, Lang’s (1977, 1979,
1985) and Drobes and Lang’s (1995) bioinformational model, Foa and Kozak’s (1968) and
Rachman’s (1980, 1990) emotional processing model, and Creamer et al.’s (1992) network
resolution processing model. This constitutes a considerable body of empirical support for the
proposed network theory.
4.2.1. Connectionist model of PTSD
Tryon’s (1998, 1999) bidirectional associative model (BAM) is a connectionist model that has
extended our understanding of Post-traumatic Stress Disorder (PTSD) in ways that satisfy all four of
Jones and Barlow (1990) and all five of Brewin, Dalgleish, and Joseph’s (1996) requirements for a
comprehensive understanding of this disorder.
4.2.2. Novel predictions
The prediction of simultaneous cognitive, affective, and behavioral change discussed above is
novel. Tryon’s (1999) BAM model of PTSD also makes several novel empirically testable
86
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
hypotheses. While it is desirable for new theories to make novel predictions, it is not essential that
they do so. The emotional processing informal network models reviewed above gained broad
acceptance even though they did not make novel predictions different from those made by
Mowrer’s two-factor theory. Acceptance of these models appears to have been based on the
additional explanatory power provided by their associated network theories.
4.2.3. Additional mechanism information
Formal connectionist models provide at least two forms of additional mechanism information;
valuable contributions in their own right. First, connectionist models engage the issue of network
architecture; how the network nodes are interconnected. Much evidence supports the view that
network architecture impacts network performance. Second, connectionist models engage the issue
of how best to change the connection weights among network nodes in response to experience.
Various learning equations are used but space does not permit considering the advantages and
disadvantages of currently available options.
4.2.4. Computer simulation and empirical confirmation
Formal network models are sufficiently specific that computer programs can be used to calculate
specific results that can be compared with empirical results. Close agreement between computed results
and empirical findings provides a demonstration proof that the model is capable of explaining the
phenomena under study. Disagreement between computed results and empirical findings sets the
occasion for empirically directed model changes. Commercially available software for connectionist
modeling is increasingly available to construct and test these models (cf. Tryon, 1995a). Some textbooks
provide free software (e.g., McLeod et al., 1998; O’Reilly & Munakata, 2000).
4.2.5. Answering clinical questions
The connectionist explanation of how systematic desensitization and exposure therapy work
presented above makes the novel prediction that cognitive, affective, and behavioral changes occur
simultaneously rather than sequentially as previously noted. This illustrates how knowledge of a
change mechanism can help resolve differential predictions made by behavioral, cognitive, and
affective models of therapeutic change.
DeRubeis et al. (1990) asked bHow does cognitive therapy work?Q but did not provide any causal
mechanism information. The PDP-CNN network cascade model presented above provides a
possible answer. Short-term cognitive changes occur primarily as a result of learning-based synaptic
change; altered connection weights. These modifications change how the middle nodes bloadQ on
stimulus features resulting in changed cognition just as changing the size and sign of factor
loadings changes factor constructs. DeRubeis et al. also reported that cognitive therapy was only
partially explained by cognitive theory in that all of Baron and Kenny’s (1986) criteria for
cognitive mediation were not satisfied. Their understanding of mediation is sequential as noted
above; cognitive changes precede behavioral changes. The Baron and Kenny analytic procedure
uses linear regression methods based on unidirectional linear causation to test for mediation. The
PDP-CNN cascade model presented above entails complex iterative interactions across multiple
network layers while the network settles into a new more congruent configuration. This model
predicts that cognitive, affective, and behavioral changes occur simultaneously during each network
iteration because connection weight changes occur at all network levels during each processing cycle. The
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
87
linear regression tests prescribed by Baron and Kenny (1986), and subsequently elaborated by Holmbeck
(1997), were not developed to detect such simultaneous change. The standard mediational model of
sequential change may be incorrect and may need to be replaced by an iterative network cascade model
such as illustrated in Fig. 2.
Changes in neural architecture may answer some etiological questions. For example, synaptic pruning
is a normal developmental process that continues from birth through late adolescence but appears to
continue in persons who become schizophrenic (Huttenlocher, 1979). Connectionist simulations have
shown that excessive synaptic pruning may cause the cognitive distortions found in schizophrenia
(Hoffman & Dobscha, 1989). Traumatic brain injury can alter neural architecture and result in
psychological and behavioral change.
5. Weaknesses and limitations of connectionist explanations
The emerging field of PDP connectionism is not without its limitations and problems. O’Reilly and
Munakata (2000) noted, b. . . the history of neural network modeling has been dominated by periods of
either extreme hype or extreme skepticismQ (p. 413). These authors summarized general and specific
challenges to computational models (pp. 413–421): (a) PDP-CNN models have been criticized as being
too simple thereby omitting potentially important details. Alternatively, simplification has been viewed
as strength on the basis that essential elements are extracted and need to be understood first before
additional complexities can be properly understood (Elman, 1993). (b) PDP-CNN models have been
criticized for being too complex. The interaction of even a few fundamental principles across multiple
network layers is too complicated to fully convey verbally and requires computer simulation to fully
articulate all details, track all developmental changes, and make specific predictions. A balance must be
struck where models are sufficiently complex to capture essential features but simple enough to be
properly studied. (c) PDP-CNN models have been criticized on the basis that they can do anything given
enough free parameters and therefore it is uninteresting to show that they explain so many psychological
and behavioral phenomena. However, PDP-CNN models are frequently based on principled
considerations of how learning occurs rather than ad-hoc parameter fitting as charged. Tests of
generalization to new situations not part of developmental training diminish this criticism. (d) Some
investigators restrict their models to known neuroscience mechanisms whereas other investigators
simulate neuroscience phenomena using methods that are not biologically plausible but that implement
known biological functions. For example, learning is known to modify synaptic functions and back
propagation methods are frequently used to simulate these changes even though back propagation is not
a biologically plausible process. This decision is justified on the basis that it is a way to mathematically
simulate synaptic change in the absence of a complete understanding of how all of these changes
actually take place. PDP-CNNs mainly model function versus exact process. Likewise, mathematical
models are not biologically plausible per se but can be used to simulate biological functions. Moreover,
PDP-CNN models can be purely psychological and need not be based entirely or even partially on
biological facts. (e) PDP-CNN models have been criticized as being reductionistic. This is the mind–
body problem and is not unique to PDP-CNN models. As noted above, while the ability to reduce
network explanations to physical/chemical processes integrates psychology into the mature sciences,
network explanations are appropriate psychological explanations. Some critics remain unconvinced that
any theory of brain can inform theories of mind and behavior and reject mechanistic explanations
88
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
generally because they seem to lack free will (cf. Rychlak 1976, 1981). (f) Ilardi (2002) noted that
cognitive psychology has traditionally grounded its theory in terms of symbol manipulation rather than
brain network function. Neuroscience takes the opposite perspective that only brain events are important.
Cognitive neuroscience provides an interactionist synthesis of these opposite positions in its quest to
understand how mind emerges from brain (cf. Ilardi, 2002). PDP-CNN models constitute part of
cognitive neuroscience. They address the bhardQ neuroscience question of how brain networks instantiate
psychology and behavior; how mind and behavior emerge from brain function. Eliminative
connectionism maintains that rules and symbols will be explained in terms of network function. While
it may seem obvious that the brain actually consists of neural networks and therefore our ability to
abstract symbols and follow rules must be understandable in terms of network functions, critics charge
that our current conceptualization of connectionism at best does not yet fully realize this objective and at
worst will never do so (cf. Fodor, 1997; Fodor & McLaughlin, 1990; Fodor & Pylyshyn, 1988; Marcus,
1998a, 1998b, 1999; Quartz, 1993). The exponentially growing interdisciplinary literature reporting
positive findings and demonstration proofs regarding the simulation of diverse psychological and
behavioral phenomena provides positive evidence to the contrary. (g) Most neural network models
pertain to gradual cognitive change and most current models have difficulty with rapid change.
However, given that the brain implements rapid learning it seems that eventually network models should
also be able to do so. (h) PDP-CNN models have been criticized for a lack of cumulative research. This
criticism can be made of all new areas that have not had sufficient time to develop a substantial body of
cumulative findings. This criticism was once true for all currently well-established models at some early
point in their development.
Obstacles to acceptance remain even if all of the issues identified above could be resolved. Garson
(1998, pp. 16–17) identified additional obstacles to the spread of neural network analysis in the social
sciences. His first obstacle concerns explanation. Understanding how a connectionist neural network
model functions after it has gone through its developmental training can be difficult because the
functional properties of a large set of connection weights cannot be readily described. While various
elements of the network can easily be described, their complex interactions require computer simulation
to follow. Simple causal analyses are frequently unavailable in verbal terms. The complex iterative
network cascade causing simultaneous changes at every processing cycle is a good example.
Mathematical analyses are available but they are unfamiliar to most social scientists who are likely to
reject them as not part of their field although mathematical psychologists may take exception to this
position. New analytical methods need to be developed along with ways to teach them to psychologists.A
second obstacle to acceptance identified by Garson (1998) is that there are many implementations
of connectionist models. Conceptually augmenting the S–R model to the S–O–R model was a
simple extension. Explicating the mediating O step with a wide variety of mathematically stated
models not covered in most graduate psychology programs may deter many readers.
A third obstacle identified by Garson (1998) is the slow rate with which neural network models
have become incorporated into common statistical packages such as SPSS and SAS. Access to user
friendly software greatly facilitated the use of complex multivariate statistics such as factor analysis
and multiple regression that are now routinely used by social scientists. Additional software is
becoming available. Some textbooks come bundled with free software (e.g., McLeod et al., 1998;
O’Reilly & Munakata, 2000).
A fourth obstacle identified by Garson (1998) is the new vocabulary needed to discuss connectionist
models. Reluctance to learn new terms may cause some readers to avoid engaging the connectionist
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
89
literature thereby rendering its positive contributions unavailable to them. Scientific progress is
frequently left to young new investigators who are learning for the first time (Kuhn, 1996).
Behavioral psychologists may question the value of pursuing network explanations on the basis
that it is just another form of homunculus like all other cognitive explanations they reject. Donahoe
(1991, 1997), Donahoe and Palmer (1994), and Tryon (2002a) have argued against this thesis on the
basis that PDP-CNN models are selectionist systems that are completely consistent with Skinner’s
explanatory style. Donahoe (1997) goes further and argues for bthe necessity of neural networksQ in
the development of behavior theory.
Perhaps the most sweeping criticism is to suggest that the connectionist position presented above does
not advance our understanding of how systematic desensitization and exposure therapy work beyond the
simplistic and obvious assertion that if a psychological intervention has therapeutic effects, then it must
entail biological change. All connectionist explanations of every psychological and behavioral
phenomena can be dismissed in this way. This view: (a) assumes a reductionist perspective that is
not shared by all psychologists (cf. Rycklak, 1976, 1981, 1994), (b) assumes that this perspective is selfevident rather than a working hypothesis that requires evidential support, and (c) assumes that the details
of how psychology and behavior emerge from biology are unimportant and not worth knowing. Failure
to consider possible proximal causal mechanisms for how systematic desensitization and exposure
therapy work created the knowledge gap that set the occasion for this article. Future focus on empirically
supported principles is a positive step towards filling this knowledge gap and PDP-CNN models provide
a rich source of mechanism information.
6. Conclusions
Reciprocal inhibition, counterconditioning, habituation, extinction, two-factor model, cognitive
changes including expectation, self-efficacy, and cognitive restructuring, and emotional processing were
considered as possible explanatory mechanisms for the effectiveness of systematic desensitization and
exposure therapy. Various problems were identified that attenuate or undercut their explanatory force. A
connectionist network cascade mechanism was presented that provides information beyond that
specified by informal network theories that accords rather well with empirical evidence. It at least
provides a starting position for conducting additional research to critically appraise the merits and
limitations of such an explanation. It also provides a step towards theoretical unification within
psychology based on an empirically supported learning principle (Tryon, 1993a, 2002b). Perhaps this
will increase the number of clinicians who can conceptualize their client’s presenting problem from a
learning-memory perspective and therefore move our field closer to the vision of Eysenck (1964), Wolpe
and Lazarus (1966), and Wolpe (1969) who hoped that intervention would be based on modern learning
theory.
7. Uncited references
Davis & Palladino, 2000
Noll, 1995
Tierney, 1995
90
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
References
Abelson, R. P., Aronson, E., McGuire, W. J., Newcomb, T. M., Rosenberg, M. J., & Tannenbaum, P. H. (Eds.) (1968). Theories
of cognitive consistency: A sourcebook. Chicago7 Rand McNally.
Agras, W. S., Leitenberg, H., Barlow, D. H., Curtis, N. A., Edwards, J., & Wright, D. (1971). Relaxation in systematic
desensitization. Archives of General Psychiatry, 25, 511 – 514.
APA (2002). Ethical principles of psychologists and code of conduct. American Psychologist, 57, 1060 – 1073.
Aponte, J. F., & Aponte, C. E. (1971). Group preprogrammed systematic desensitization without the simultaneous presentation
of aversive scenes with relaxation training. Behaviour Research and Therapy, 9, 337 – 346.
Arntz, A., Hildebrand, M., & van den Hout, M. (1994). Overprediction of anxiety and disconfirmatory processes in anxiety
disorders. Behaviour Research and Therapy, 32, 709 – 722.
Bandura, A. (1969). Principles of behavior modification. New York7 Holt.
Bandura, A. (1977). Self-efficacy: Toward a unifying theory of behavior change. Psychological Review, 84, 191 – 215.
Bandura, A. (1978). The self system in reciprocal determinism. American Psychologist, 33, 344 – 358.
Bandura, A. (1982). Self-efficacy mechanism in human agency. American Psychologist, 37, 122 – 147.
Bandura, A. (1998). Self-Efficacy: The exercise of control. San Francisco, CA7 Freeman.
Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinctions in social psychological research:
Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173 – 1182.
Baskin, T. W., Tierney, S. C., Minami, T., & Wampold, B. E. (2003). Establishing specificity in psychotherapy: A meta-analysis
of structural equivalence of placebo controls. Journal of Counseling and Clinical Psychology, 71, 973 – 979.
Baxter, L. R., Jr., Schwartz, J. M., Bergman, K. S., Szuba, M. P., Guze, B. H., Mazziotta, J. C., Alazraki, A., Selin, C. E., Ferng,
H. K., Munford, P., & Phelps, M. E. (1992). Caudate glucose metabolic rate changes with both drug and behavior therapy
for obsessive–compulsive disorder. Archives of General Psychiatry, 49, 681 – 689.
Bechtel, W., & Abrahamsen, A. (2002). Connectionism and the mind: Parallel processing, dynamics, and evolution in networks
(2nd ed.). Cambridge, MA7 Blackwell Publishers.
Beck, A. T. (1976). Cognitive therapy and the emotional disorders. New York7 International Universities Press.
Beck, J. S. (1995). Cognitive therapy: Basics and beyond. New York7 The Guilford Press.
Benjamin, S., Marks, I. M., & Hudson, J. (1972). Active muscular relaxation in desensitization of phobic patients.
Psychological Medicine, 2, 381 – 390.
Borkovec, T. D. (1972). Effects of expectancy on the outcome of systematic desensitization and implosive treatments for
analogue anxiety. Behavior Therapy, 3, 29 – 40.
Borkovec, T. D. (1973). The role of expectancy and physiological feedback in fear research: A review with special reference to
subject characteristics. Behavior Therapy, 4, 491 – 505.
Bottjer, S. W., & Arnold, A. P. (1997). Developmental plasticity in neural circuits for a learned behavior. Annual Review of
Neuroscience, 20, 459 – 481.
Bowler, P. J. (1983). The eclipse of Darwinism: Anti-Darwinian evolution theories in the decades around 1900. Baltimore, MD7
The Johns Hopkins University Press.
Brewin, C. R., Dalgleish, T., & Joseph, S. (1996). A dual representation theory of Posttraumatic Stress Disorder. Psychological
Review, 103, 670 – 686.
Brown, H. A. (1973). Role of expectancy manipulation in systematic desensitization. Journal of Consulting and Clinical
Psychology, 41, 405 – 411.
Chemtob, C., Roitblat, H. L., Hamada, R. S., Carlson, J. G., & Twentyman, C. T. (1988). A cognitive action theory of posttraumatic stress disorder. Journal of Anxiety Disorders, 2, 253 – 275.
Commons, M. L., Grossberg, S., & Staddon, J. E. R. (1991). Neural network models of conditioning and action. Hillsdale, NJ7
Lawrence Earlbaum Associates.
Cooke, G. (1968). Evaluation of the efficacy of the components of reciprocal inhibition psychotherapy. Journal of Abnormal
Psychology, 73, 464 – 467.
Craighead, W. E. (1973). The role of muscular relaxation in systematic desensitization. In R. D. Rubin, J. P. Brady, & J. D.
Henderson (Eds.), Advances in Behavior Therapy, vol. 4 (pp. 177 – 197). New York7 Academic Press.
Creamer, M., Burgess, P., & Pattison, P. (1992). Reaction to trauma: A cognitive processing model. Journal of Abnormal
Psychology, 101, 452 – 459.
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
91
Crowder, J. E., & Thornton, D. W. (1970). Effects of systematic desensitization, programmed fantasy and bibliotherapy on a
specific fear. Behaviour Research and Therapy, 8, 35 – 41.
Darwin, C. (1859). The origin of species: By means of natural selection or the preservation of favoured races in the struggle for
life. New York7 Appelton.
Davis, S. F., & Palladino, J. J. (2000). Psychology (3rd ed.). Upper Saddle River, NJ7 Prentice-Hall.
Davison, G. C. (1966). Anxiety under total curarization: Implications for the role of muscular relaxation in the desensitization of
neurotic fears. Journal of Nervous and Mental Disease, 143, 443 – 448.
Davison, G. C. (1968). Systematic desensitization as a counterconditioning process. Journal of Abnormal Psychology, 73, 91.
Davison, G. C., & Wilson, G. T. (1972). Critique of desensitization: Social and cognitive factors underlying the effectiveness of
Wolpe’s procedure. Psychological Bulletin, 78, 28 – 31.
Davison, G. C., & Wilson, G. T. (1973). Process of fear-reduction in systematic desensitization: Cognitive and social
reinforcement factors in humans. Behavior Therapy, 4, 1 – 21.
DeRubeis, R. J., Evans, M. D., Hollon, S. D., Garvey, M. J., Grove, W. M., & Tuason, V. B. (1990). How does cognitive
therapy work? Cognitive change and symptom change in cognitive therapy and pharmacotherapy for depression. Journal of
Consulting and Clinical Psychology, 58, 862 – 869.
Donahoe, J. W. (1991). The selectionist approach to verbal behavior: Potential contributions of neuropsychology and
connectionism. In L. J. Hayes, & P. N. Chase (Eds.), Dialogues on verbal behavior: The first international institute on
verbal relations (pp. 1 – 19). Reno, NV7 Context Press.
Donahoe, J. W. (1997). The necessity of neural networks. In G. E. Stelmach, P. A. Vroon, J. W. Donahoe, & V. P. Dorsel (Eds.),
Advances in psychology. Neural-Network Models of Cognition: Biobehavioral Foundations, vol. 121 (pp. 1 – 19).
Amsterdam7 Elsevier.
Donahoe, J. W., & Palmer, D. C. (1994). Learning and complex behavior (pp. 18–27, 31–67). New York: Allyn and Bacon.
Drobes, D. J., & Lang, P. J. (1995). Bioinformational theory and behavior therapy. In W. O’Donohue, & L. Krasner (Eds.),
Theories of behavior therapy: Exploring behavior change (pp. 229 – 257). Washington, DC7 American Psychological
Association.
Ellis, A., & Blau, S. (Eds.) (1998). The Albert Ellis reader: Rational emotive behavior therapy, a guide to well-being. Secaucus,
NJ7 Carol Pub. Group.
Ellis, A., & Whiteley, J. M. (Eds.) (1979). Theoretical and empirical foundations of rational-emotive therapy. Monterey, CA7
Brooks/Cole.
Elman, J. L. (1993). Learning and development in neural networks: The importance of starting small. Cognition, 48, 71 – 99.
Emmelkamp, P. M. G. (1982). Phobic and obsessive–compulsive disorders: Theory, research and practice (pp. 52 – 59). New
York7 Plenum.
Emmelkamp, P. M. G. (1994). Behavior therapy with adults. In A. E. Bergin, & S. L. Garfield (Eds.), Handbook of
psychotherapy and behavior change (4th ed.). New York7 Wiley.
Emmelkamp, P. M. G. (2001). Behavior therapy with adults. In L. Beutler (Ed.), Bergin and Garfield’s handbook of
psychotherapy and behavior change (5th ed.). New York7 Wiley.
Emmelkamp, P. M. G., & Felten, M. (1985). The process of exposure in vivo: Cognitive and physiological changes during
treatment of acrophobia. Behaviour Research and Therapy, 23, 219 – 223.
Eysenck, H. J. (1964). Experiments in behaviour therapy: Readings in modern methods of treatment of mental disorders
derived from learning theory. New York7 Macmillan.
Festinger, L. (1957). A theory of cognitive dissonance. Stanford, CA7 Stanford University Press.
Foa, E. B., & Kozak, M. J. (1986). Emotional processing of fear: Exposure to corrective information. Psychological Bulletin,
99, 20 – 35.
Foa, E. B., Steketee, G., & Rothbaum, B. O. (1989). Behavioral/cognitive conceptualization of post-traumatic stress disorder.
Behavior Therapy, 20, 155 – 176.
Fodor, J. (1997). Connectionism and the problem of systematicity (continued): Why Smolensky’s solution still doesn’t work.
Cognition, 62, 109 – 119.
Fodor, J., & McLaughlin, B. P. (1990). Connectionism and the problem of systematicity: Why Smolensky’s solution doesn’t
work. Cognition, 35, 183 – 204.
Fodor, J. A., & Pylyshyn, Z. W. (1988). Connectionism and cognitive architecture: A critical analysis. Cognition, 28, 3 – 71.
Forgas, J. P. (1999). Network theories and beyond. In T. Dalgleish, & M. Power (Eds.), Handbook of cognition and emotion
(pp. 591 – 611). New York7 Wiley.
92
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
Frank, J. D. (1961). Persuasion and healing. Baltimore, MD7 Johns Hopkins University Press.
Freeling, N. W., & Shemberg, K. M. (1970). The alleviation of test anxiety by systematic desensitization. Behaviour Research
and Therapy, 8, 293 – 299.
Garson, G. D. (1998). Neural networks: An introductory guide for social scientists. Thousand Oaks, CA7 Sage Publications.
Gluck, M. A., Meeter, M., & Myers, C. E. (2003). Computational models of the hippocampal region: Linking incremental
learning and episodic memory. Trends in Cognitive Sciences, 7, 269 – 276.
Goldfried, M. R. (1971). Systematic desensitization as training in self-control. Journal of Consulting and Clinical Psychology,
37, 228 – 234.
Guthrie, E. R. (1952). The psychology of learning. New York7 Harper.
Harris, J. D. (1943). Habituatory response decrement in the intact organism. Psychological Bulletin, 40, 385 – 422.
Hebb, D. O. (1949). The organization of behavior. New York7 Wiley.
Hebert, J. D. (2003). The science and practice of empirically supported treatments. Behavior Modification, 27, 412 – 430.
Heider, F. (1946). Attitudes and cognitive organization. Journal of Personality, 21, 107 – 112.
Heider, F. (1958). The psychology of interpersonal relations. New York7 Wiley.
Hergenhan, B. R., & Olson, M. H. (1993). An introduction to theories of learning (4th ed.). Englewood Cliffs, NJ7 Prentice Hall.
Hoffman, R. E., & Dobscha, S. K. (1989). Cortical pruning and the development of schizophrenia: A computer model.
Schizophrenia Bulletin, 15, 477 – 490.
Holmbeck, G. N. (1997). Toward terminological, conceptual, and statistical clarify in the study of mediators and moderators:
Examples from the child-clinical and pediatric psychology literatures. Journal of Consulting and Clinical Psychology, 65,
599 – 610.
Huttenlocher, P. R. (1979). Synaptic density in human frontal cortex—Developmental changes and the effects of aging. Brain
Research, 163, 195 – 205.
Ilardi, S. S. (2002). The cognitive neuroscience perspective: A brief primer for clinical psychologists. The Behavior Therapist,
25, 49 – 52.
Jacobson, E. (1938). Progressive relaxation. Chicago7 University of Chicago Press.
Jones, J. C., & Barlow, D. H. (1990). The etiology of post-traumatic stress disorder. Clinical Psychology Review, 10,
299 – 328.
Kalat, J. W. (2001). Biological psychology (8th ed.). Australia7 Wadsworth.
Kandel, E. R. (1991). Cellular mechanisms of learning and the biological basis of individuality. In E. R. Kandel, J. H. Schwartz,
& T. M. Jessell (Eds.), Principles of neural science (pp. 1009 – 1031). Norwalk, CT7 Appleton and Lange.
Kass, W., & Gilner, F. H. (1974). Drive level, incentive conditions and systematic desensitization. Behaviour Research and
Therapy, 12, 99 – 106.
Kazdin, A. E., & Wilcoxon, L. A. (1976). Systematic desensitization and nonspecific treatment effects. Psychological Bulletin,
83, 729 – 758.
Kazdin, A. E., & Wilson, G. T. (1978). Evaluation of behavior therapy: Issues, evidence, and research strategies. Cambridge,
MA7 Ballinger.
Kheriaty, E., Kleinknecht, R. A., & Hyman, I. E., Jr. (1999). Recall and validation of phobia origins as a function of a structured
interview versus the Phobia Origins Questionnaire. Behavior Modification, 23, 61 – 78.
Kleinknecht, R. A. (1994). Acquisition of blood, injury and needle fears and phobias. Behaviour Research and Therapy, 32,
817 – 823.
Kolb, B., & Whishaw, I. Q. (1998). Brain placticity and behavior. Annual Review of Psychology, 49, 43 – 64.
Krasne, F. (2002). Neural analysis of learning in simple systems. In H. Pashler, & R. Gallistel (Eds.), Stevens’ handbook of
experimental psychology (3rd ed.). Learning, Motivation, and Emotion, vol. 3 (pp. 131 – 200). New York7 Wiley.
Kuhn, T. S. (1996). The structure of scientific revolutions. Chicago7 University of Chicago Press.
Lader, M. H., & Mathews, A. M. (1968). A physiological model of phobic anxiety and desensitization. Behaviour Research and
Therapy, 6, 411 – 421.
Lang, P. J. (1977). Imagery in therapy: An information processing analysis of fear. Behavior Therapy, 8, 862 – 886.
Lang, P. J. (1979). A bio-information theory of emotional imagery. Psychophysiology, 16, 495 – 512.
Lang, P. J. (1985). The cognitive psychophysiology of emotion: Fear and anxiety. In A. H. Turner, & J. Maser (Eds.), Anxiety
and the anxiety disorders (pp. 131 – 170). Hillsdale, NJ7 Lawrence Erlbaum Associates.
Levis, D. J. (2002). Implosive therapy. In M. Hersen, & W. Sledge (Eds.), Encyclopedia of Psychotherapy, vol. 2 (pp. 1 – 6).
New York7 Elsevier Science.
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
93
Lynch, G. (2000). Memory consolidation and long-term potentiation. In M. S. Gazzaniga (Ed.), The new cognitive
neurosciences (2nd ed.) (pp. 139–157 ). Cambridge, MA7 The MIT Press.
Mahoney, M. J. (1974). Cognition and behavior modification. Cambridge, MA7 Lippincott.
Marcus, G. F. (1998). Can connectionism save constructivism? Cognition, 66, 153 – 182.
Marcus, G. F. (1998). Rethinking eliminative connectionism. Cognitive Psychology, 37, 243 – 282.
Marcus, G. F. (1999). Connectionism: With or without rules? Trends in Cognitive Sciences, 3, 168 – 170.
Marks, I. (1975). Behavioral treatments of phobic and obsessive compulsive disorders: A critical appraisal. In M. Hersen, R. M.
Eisler, & P. M. Miller (Eds.), Progress in Behavior Modification, vol. 1. New York7 Academic Press.
Martin, K. C., Bartsch, D., Bailey, C. H., & Kandel, E. R. (2000). Molecular mechanisms underlying learning-related longlasting synaptic plasticity. In M. S. Gazzaniga (Ed.), The new cognitive neurosciences (2nd ed.). Cambridge, MA7 The MIT
Press.
Martin, S. J., Grimwood, P. D., & Morris, R. G. M. (2000). Synaptic placticity and memory: An evaluation of the hypothesis.
Annual Review of Psychology, 23, 649 – 711.
Martindale, C. (1991). Cognitive psychology: A neural-network approach. Pacific Grove, CA7 Brooks/Cole.
Matzel, L. D. (2002). Learning mutants. In H. Pashler, & R. Gallistel (Eds.), Stevens’ handbook of experimental psychology
(3rd ed.). Learning, Motivation, and Emotion, vol. 3 (pp. 201 – 238). New York7 Wiley.
Mayr, E. (1982). The growth of biological thought: Diversity, evolution, and inheritance. Cambridge, MA7 Harvard University
Press.
McAllister, W. R., & McAllister, D. E. (1995). Two-factor fear theory: Implications for understanding anxietybased clinical phenomena. In W. O’Donohue, & L. Krasner (Eds.), Theories of behavior therapy: Exploring behavior
change (pp. 145 – 171). Washington, DC7 American Psychological Association.
McClelland, J. L., & Rumelhart, D. E. (1987). Parallel distributed processing: Explorations in the microstructure of cognition.
Psychological and Biological Models, vol. 2. Cambridge, MA7 MIT Press.
McGlynn, F. D. (1973). Graded imagination and relaxation as components of experimental desensitization. Journal of Nervous
and Mental Disease, 156, 377 – 385.
McGlynn, F. D. (2002). Systematic desensitization. In M. Hersen, & W. Sledge (Eds.), Encyclopedia of Psychotherapy,
vol. 2 (pp. 755 – 764). New York7 Elsevier Science.
McGlynn, F. D., Mealiea Jr., W. L., & Landau, D. L. (1981). The current status of systematic desensitization. Clinical
Psychology Review, 1, 149 – 179.
McLeod, P., Plunkett, K., & Rolls, E. T. (1998). Introduction to connectionist modelling of cognitive processes. New York7
Oxford University Press.
Meichembaum, D. (1974). Cognitive behavior modification. Morristown, NJ7 General Learning Press.
Menzies, R. G., & Clarke, J. C. (1995). The etiology of phobias: A nonassociative account. Clinical Psychology Review, 15,
23 – 48.
Miller, C. (2002). Flooding. In M. Hersen, & W. Sledge (Eds.), Encyclopedia of Psychotherapy, vol. 1 (pp. 809 – 813). New
York7 Elsevier Science.
Miller, H. R., & Nawas, M. M. (1970). Control of aversive stimulus termination in systematic desensitization. Behaviour
Research and Therapy, 8, 57 – 61.
Morgan, W. G. (1973). Nonnecessary conditions or useful procedures in desensitization: A reply to Wilkins. Psychological
Bulletin, 79, 373 – 375.
Mowrer, O. H. (1960). Learning theory and the symbolic processes. New York7 Wiley.
Nawas, M. M., Welsch, W. V., & Fishman, S. T. (1970). The comparative effectiveness of pairing aversive imagery with
relaxation, neutral tasks and muscular tension in reducing snake phobia. Behaviour Research and Therapy, 6, 63 – 68.
Noll, J. P. (1995). The matching law as a theory of choice in behavior therapy. In W. O’Donohue, & L. Krasner (Eds.), Theories
of behavior therapy: Exploring behavior change (pp. 129 – 144). Washington, DC7 American Psychological Association.
O’Reilly, R. C., & Munakata, Y. (2000). Computational explorations in cognitive neuroscience: Understanding the mind by
simulating the brain. Cambridge, MA7 MIT Press.
Oreskes, W. (1999). Rejection of continental drift. New York7 Oxford University Press.
Öst, L. -G., & Hugdahl, D. (1981). Acquisition of phobias and anxiety response patterns in clinical patients. Behaviour
Research and Therapy, 19, 439 – 447.
Packard, M. G., & Knowlton, B. J. (2002). Learning and memory functions of the basal ganglia. Annual Review of
Neuroscience, 25, 563 – 593.
94
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
Plaud, J. J., (Eds.). (1998). From behavior theory to behavior therapy. Boston7 Allyn and Bacon.
Plaud, J. J., & Vogelantz, N. D. (1997). Back to the future: The continued relevance of behavior theory to modern behavior
therapy. Behavior Therapy, 28, 403 – 414.
Proctor, R. W., & Capaldi, E. J. (2001). Empirical evaluation and justification of methodologies in psychological science.
Psychological Bulletin, 127, 759 – 772.
Quartz, S. R. (1993). Neural networks, nativism, and the plausibility of constructivism. Cognition, 48, 223 – 242.
Rachman, S. (1965). Studies in desensitization: I. The separate effects of relaxation and desensitization. Behaviour Research
and Therapy, 3, 245 – 252.
Rachman, S. (1980). Emotional processing. Behaviour Research and Therapy, 18, 51 – 60.
Rachman, S. (1990). Fear and courage. New York7 W.H. Freeman.
Rachman, S., & Bichard, S. (1988). The overprediction of fear. Clinical Psychology Review, 8, 303 – 313.
Read, S. J., (Eds.). (1998). Connectionist models of social reasoning and social behavior. Mahwah, NJ7 Lawrence Earlbaum.
Rolls, E. T., & Treves, A. (1998). Neural networks and brain function. Oxford, UK7 Oxford University Press.
Rosen, G. M., & Davison, G. C. (2003). Psychology should list empirically supported principles of change (ESPs) and not
credential trademarked therapies or other treatment packages. Behavior Modification, 27, 300 – 312.
Rumelhart, D. E., & McClelland, J. L. (1986). Parallel distributed processing: Explorations in the microstructure of cognition.
Foundations, vol. 1. Cambridge, MA7 MIT Press.
Rychlak, J. F. (Ed.). (1976). Dialectics: Humanistic rationale for behavior and development. New York7 S. Karger.
Rychlak, J. F. (1981). Philosophy of science for personality theory (2nd ed.). Malabar, FL7 Krieger Pub.
Rychlak, J. F. (1994). Logical learning theory: A human teleology and its empirical support. Lincoln, NE7 University of
Nebraska.
Sanderson, W. C. (2003). Why empirically supported treatments are important. Behavior Modification, 27, 290 – 299.
Schwartz, J. M., Stoessel, P. W., Baxter Jr., L. R., Martin, K. M., & Phelps, M. E. (1996). Systematic changes in cerebral
glucose metabolic rate after successful behavior modification treatment of obsessive–compulsive disorder. Archives of
General Psychiatry, 53, 109 – 113.
Shultz, T. R., & Lepper, M. R. (1996). Cognitive dissonance reduction as constraint satisfaction. Psychological Review, 103,
219 – 240.
Shapiro, D. A. (1995). Finding out how psychotherapies help people change. Psychotherapy Research, 5, 1 – 21.
Sherrington, C. S. (1961). Integrative action of the nervous system. New Haven, CT7 Yale University Press. Original work
published 1906.
Spiegler, M. D., & Guevremont, D. C. (1993). Contemporary behavior therapy. Pacific Grove, CA7 Brooks/Cole.
Spitzer, M. (1999). The mind within the net: Models of learning, thinking, and acting. Cambridge, MA7 The MIT Press.
Taylor, S. (1995). Stimulus estimation and the overprediction of fear: A comment on two studies. Behaviour Research and
Therapy, 33, 699 – 700.
Taylor, S. (2002). Systematic desensitization. In M. Hersen, & W. Sledge (Eds.), Encyclopedia of Psychotherapy, vol. 2
(pp. 755 – 759). New York7 Elsevier Science.
Taylor, S., & Rachman, S. (1994). Stimulus estimation and the overprediction of fear. British Journal of Clinical Psychology,
33, 173 – 181.
Telch, M. J., Valentiner, D., & Bolte, M. (1994). Proximity to safety and its effects on fear prediction bias. Behaviour Research
and Therapy, 32, 747 – 751.
Thompson, R. F., & Spencer, W. A. (1966). Habituation: A model phenomenon for the study of neuronal substrates of behavior.
Psychological Bulletin, 73, 16 – 43.
Tierney, K. J. (1995). Molar regulatory theory and behavior therapy. In W. O’Donohue, & L. Krasner (Eds.), Theories of
behavior therapy: Exploring behavior change (pp. 97 – 128). Washington, DC7 American Psychological Association.
Thagard, P. (2000). Coherence in thought and action. Cambridge, MA7 The MIT Press.
Tryon, W. W. (1993). Neural networks: I. Theoretical unification through connectionism. Clinical Psychology Review, 13,
341 – 352.
Tryon, W. W. (1993). Neural networks: II. Unified learning theory and behavioral psychotherapy. Clinical Psychology Review,
13, 353 – 371.
Tryon, W. W. (1995a). Neural networks for behavior therapists: What they are and why they are important. Behavior Therapy,
26, 295 – 318.
Tryon, W. W. (1995b). Resolving the cognitive behavioral controversy. The Behavior Therapist, 18, 83 – 86.
W.W. Tryon / Clinical Psychology Review 25 (2005) 67–95
95
Tryon, W. W. (1996). Yes—Neural network learning theory can resolve the behavioral-cognitive controversy. The Behavior
Therapist, 19(70), 72 – 73.
Tryon, W. W. (1998). A neural network explanation of posttraumatic stress disorder. Journal of Anxiety Disorders, 12, 373 – 385.
Tryon, W. W. (1999). A bidirectional associative memory explanation of posttraumatic stress disorder. Clinical Psychology
Review, 19, 789 – 818.
Tryon, W. W. (2000). Behavior therapy as applied learning theory. The Behavior Therapist, 23, 131 – 134.
Tryon, W. W. (2002). Expanding the explanatory base of behavior analysis via modern connectionism: Selectionism as a
common explanatory core. The Behavior Analyst Today, 3, 104 – 118.
Tryon, W. W. (2002). Network models contribute to cognitive and social neuroscience. American Psychologist, 57, 728.
Van Egeren, L. F. (1971). Psychophysiological aspects of systematic desensitization: Some outstanding issues. Behaviour
Research and Therapy, 9, 65 – 77.
Van Egeren, L. F., Feather, B. W., & Hein, P. L. (1971). Desensitization of phobias: Some psychophysiological propositions.
Psychophysiology, 8, 213 – 228.
Waters, W. F., McDonald, D. G., & Koresko, R. L. (1972). Psychophysiological responses during analogue systematic
desensitization and non-relaxation control procedures. Behaviour Research and Therapy, 10, 381 – 393.
Watts, F. N. (1979). Habituation model of systematic desensitization. Psychological Bulletin, 86, 627 – 637.
Wilkins, W. (1971). Desensitization: Social and cognitive factors underlying the effectiveness of Wolpe’s procedure.
Psychological Bulletin, 76, 311 – 317.
Wilkins, W. (1972). Getting it together with Davison and Wilson. Psychological Bulletin, 78, 32 – 36.
Wilkins, W. (1973). A rejoinder to Morgan. Psychological Bulletin, 79, 376 – 377.
Wilkins, W. (1973). Expectancy of therapeutic gain: An empirical and conceptual critique. Journal of Consulting and Clinical
Psychology, 40, 69 – 77.
Wilson, E. O. (1998). Consilience: The unity of knowledge. New York7 Knopf.
Wolpe, J. (1958). Psychotherapy by reciprocal inhibition. Stanford, CA7 Stanford University Press.
Wolpe, J. (1969). The practice of behavior therapy. Elmsford, NY7 Pergamon Press.
Wolpe, J. (1995). Reciprocal inhibition: Major agent of behavioral change. In W. O’Donohue, & L. Krasner (Eds.), Theories of
behavior therapy: Exploring behavior change (pp. 23 – 57). Washington, DC7 American Psychological Association.
Wolpe, J., & Lazarus, A. A. (1966). Behavior therapy techniques: A guide to the treatment of neuroses. New York7 Pergamon.
Wright, L. M., Holborn, S. W., & Rezutek, P. E. (2002). An experimental test of stimulus estimation theory: Danger and safety
with snake phobic stimuli. Behaviour Research and Therapy, 40, 911 – 922.
Yates, A. J. (1975). Theory and practice in behavior therapy (pp. 152 – 182). New York7 Wiley.