Did

you know that

Pavlovian conditioning typically does not involve the learning of a new
conditioned reflex or stimulus–response (S–R) connection but rather the
learning of a new stimulus–stimulus (S–S) connection?
conditioned responding can be increased or decreased by changing the
value of the unconditioned stimulus (US), which is an intervention that
does not involve presenting the conditioned stimulus (CS)?
according to all contemporary models of learning, what you learn about
one stimulus depends on the associative value of other concurrently
Copyright © 2017. American Psychological Association. All rights reserved.

present stimuli?
a CS can lose associative strength even though it is paired with a US?
attentional theories assume that what happens on one trial determines how
much attention is devoted to the CS on the next trial?
many major theories of learning do not consider time in their formulations?
the absolute duration of the CS is not as important for learning as is the
ratio between the CS duration and the interval between successive US
presentations?
conditioned responding depends on the associative value of the CS
compared with the associative value of other cues that were present at the
time the CS was conditioned?

Originally, Pavlovian conditioning was considered to be a simple form of learning that

Domjan, Michael. The Essentials of Conditioning and Learning, American Psychological Association, 2017. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kuleuvenul/detail.action?docID=5116675.
Created from kuleuvenul on 2018-09-27 08:10:51.
 depended only on pairings of a conditioned stimulus (CS) with an unconditioned stimulus (US)
and resulted in the conditioning of a new reflex. This naive perspective has turned out to be
incorrect in many ways. Long-delay taste aversion learning, selective associations, and the
blocking effect all challenge the view that Pavlovian conditioning is a simple form of learning.
In this chapter, I continue to document the richness and complexity of Pavlovian conditioning
by focusing on the underlying mechanisms and theories that address this form of learning. I
discuss two major questions: (a) What is learned in Pavlovian conditioning, and (b) how is it
learned?

What Is Learned in Pavlovian Conditioning?

The signature outcome of a Pavlovian conditioning procedure is that the participant comes to
perform a conditioned response (CR) when the CS is presented. What mechanism is
responsible for this CR? There are two prominent alternatives. According to the first
mechanism, the CS comes to elicit the CR directly. This is called S–R learning and is the
simpler of the two mechanisms. S–R mechanisms dominated theories of learning up until the
“cognitive revolution” that swept over psychology in the 1970s. That revolution encouraged
more “cognitive” theories of learning and the possibility that through Pavlovian conditioning
the CS comes to activate a representation of the US. That US representation or memory in turn
generates the CR. This second mechanism is called S–S learning. Because both S–R and S–S
learning mechanisms can generate the CR, how can we distinguish between them?
According to the S–R learning mechanism, classical conditioning leads to the formation of
an association between the CS and the CR. As a result of this CS–CR association, presentation
of the CS activates the CR directly and automatically. Such S–R learning is what is implied by
the traditional notion that Pavlovian conditioning results in the learning of a new reflex
response to the CS.
How S–S learning generates conditioned responding is a bit more complicated. According
Copyright © 2017. American Psychological Association. All rights reserved.

to the S–S learning mechanism, Pavlovian conditioning results in the learning of CS–US
association. Once this association is acquired, presentation of the CS will activate a neural
representation of the US (see Figure 6.1). Expressed informally, this means that upon
encountering the CS, the participant will start thinking about the US. This activation of the US
representation does not generate a response automatically. Rather, what the participant will do
will depend on its motivation to respond to the US at that time.

Domjan, Michael. The Essentials of Conditioning and Learning, American Psychological Association, 2017. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kuleuvenul/detail.action?docID=5116675.
Created from kuleuvenul on 2018-09-27 08:10:51.
 FIG URE 6. 1

Distinction between stimulus–response (S–R) and stimulus–stimulus (S–S) learning. In S–R learning, a
direct connection or association is established between the conditioned stimulus (CS) and the conditioned
response (CR) such that the CR is elicited directly upon presentation of the CS. In S–S learning, the CS
activates a representation of the unconditioned stimulus (US), which in turn leads to the CR.

A powerful technique for differentiating between S–R and S–S mechanisms was
popularized by Robert Rescorla (1973) and is basically a test of performance. The test
involves evaluating the vigor of conditioned responding after the individual’s motivation to
respond to the US has been changed. In one type of experiment, for example, motivation to
respond to the US is reduced. This manipulation is called US devaluation (see Table 6.1).

T AB L E 6 . 1
Design and Predictions of US Devaluation Study
Copyright © 2017. American Psychological Association. All rights reserved.

Consider, for example, a study of sexual Pavlovian conditioning that was conducted with
domesticated male quail (Holloway & Domjan, 1993). Brief exposure to a light CS was paired
with access to a female bird once a day. Initially, the visual CS did not elicit any significant
behavior. However, because the males were sexually motivated, they always readily
approached and copulated with the female that was released at the end of each conditioning
Domjan, Michael. The Essentials of Conditioning and Learning, American Psychological Association, 2017. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kuleuvenul/detail.action?docID=5116675.
Created from kuleuvenul on 2018-09-27 08:10:51.
 trial. With repeated conditioning trials, the males also started approaching the CS. After 10
conditioning trials, the CS elicited a strong approach or sign tracking response regardless of
where the males were at the start of the trial.
According to the S–R learning mechanism, conditioned responding reflects the
establishment of a direct connection between the CS and the CR. If such a direct connection
has been established, then changing the animal’s motivation to perform the unconditioned
response should not influence its conditioned responding. Thus, an S–R interpretation predicts
that once the quail has learned the sexual conditioned approach response, presentation of the
CS will elicit the CR even if the birds are no longer sexually motivated.
Holloway and Domjan (1993) tested the S–R prediction by reducing the sex drive of one
group of birds. Sexual motivation was reduced by changing the light cycle in the laboratory to
mimic winter conditions, when the birds do not breed. The results of the experiment are
summarized in Figure 6.2. Contrary to predictions based on the S–R mechanism, a reduction in
sexual motivation reduced conditioned responding to the visual CS.
Copyright © 2017. American Psychological Association. All rights reserved.

Domjan, Michael. The Essentials of Conditioning and Learning, American Psychological Association, 2017. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kuleuvenul/detail.action?docID=5116675.
Created from kuleuvenul on 2018-09-27 08:10:51.
 FIG URE 6. 2
Copyright © 2017. American Psychological Association. All rights reserved.

Effects of unconditioned stimulus (US) devaluation on sexual approach conditioned behavior. Three test
sessions were conducted at 1-week intervals after two groups of quail had acquired a conditioned
approach response. During the test phase, the sexual motivation of one group of birds was reduced. This
US devaluation produced a decrease in conditioned responding. Adapted from “Sexual Approach
Conditioning: Tests of Unconditioned Stimulus Devaluation Using Hormone Manipulations,” by K. S.
Holloway and M. Domjan, 1993, Journal of Experimental Psychology: Animal Behavior Processes, 19, p.
49. Copyright 1993 by the American Psychological Association.

The results summarized in Figure 6.2 indicate that S–S learning had occurred in the
experiment. S–S learning does not involve learning a specific CR. Rather, it involves learning
an association between the CS and the US. Once the CS–US association has been established,
presentation of the CS activates a representation of the US. This in turn leads to conditioned
responding, but only if the participants are motivated to respond to the US. In the quail
Domjan, Michael. The Essentials of Conditioning and Learning, American Psychological Association, 2017. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kuleuvenul/detail.action?docID=5116675.
Created from kuleuvenul on 2018-09-27 08:10:51.
 experiment, the CS elicited conditioned approach behavior, but only if the birds were sexually
motivated.
In the preceding experiment, motivation to respond to the US was reduced as a test for S–S
learning. Another approach is to increase motivation to respond to the US. This is called US
inflation. If conditioned responding is mediated by S–S learning, US inflation results in
increased responding to the CS.
Although studies of US devaluation are more common than studies of US inflation effects,
both types of results have been obtained in a variety of different forms of Pavlovian
conditioning (e.g., Delamater, Campese, LoLordo, & Sclafani, 2006; Fudim, 1978; Storsve,
McNally, & Richardson, 2012). These results indicate that Pavlovian conditioning typically
involves S–S learning rather than the learning of a new S–R reflex.
A particularly interesting implication of these findings is that one can alter responses to a
CS using procedures that involve manipulations that target the US rather than the CS itself.
Most clinical interventions that seek to reduce maladaptive CRs involve changing the
properties of the CS by using something like an extinction procedure (see Chapter 10, this
volume). US devaluation effects suggest an alternative avenue. Focusing on changing the value
of the US may be especially useful in clinical situations where the troublesome CS cannot be
easily identified or manipulated.

How Are Pavlovian Associations Learned?

We next turn to considering possible mechanisms involved in the learning of Pavlovian
associations. The modern era in theories of associative learning was launched by the discovery
of the blocking effect, which demonstrated that CS–US contiguity is not sufficient for learning.
The first and most influential modern learning theory was the Rescorla–Wagner model. Other
models and theories soon followed. These alternatives sought to explore different ways of
characterizing learning and to overcome some of the shortcomings of the Rescorla–Wagner
Copyright © 2017. American Psychological Association. All rights reserved.

model. However, the Rescorla–Wagner model remains the standard against which other
theories are evaluated.

THE RESCORLA–WAGNER MODEL

Because the blocking effect was critical in shaping the development of contemporary learning
theory, the basics of the blocking effect are reviewed in Figure 6.3. Participants first receive
one CS (A) paired with the US. After conditioned responding to A is well established, a new
stimulus (B) is added, and the A+B compound is paired with the US. Blocking is said to occur
if the presence of the previously conditioned stimulus A blocks the conditioning of the new
added stimulus B.

Domjan, Michael. The Essentials of Conditioning and Learning, American Psychological Association, 2017. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kuleuvenul/detail.action?docID=5116675.
Created from kuleuvenul on 2018-09-27 08:10:51.
 FIG URE 6. 3

Review of the design of a blocking experiment. In Phase 1, the experimental group gets stimulus A
conditioned to asymptote. In Phase 2, both the experimental and the control groups get stimuli A and B
presented simultaneously and paired with the unconditioned stimulus (US). Finally, both groups are tested
for responding to stimulus B.

Why does the presence of the previously conditioned stimulus A block the acquisition of
responding to stimulus B? Kamin (1969), who originally identified the blocking effect,
explained the phenomenon by proposing that a US must be surprising to be effective in
producing learning. If the US is signaled by a CS that you learned about previously, the US will
not be surprising and therefore will not stimulate the “mental effort” needed for the formation
of an association. Expected events are things the participant has already learned about. Hence,
expected events will not activate processes leading to new learning. To be effective, the US
must be unexpected or surprising.
The idea that the effectiveness of a US is determined by how surprising it is formed the
basis of the Rescorla–Wagner model (Rescorla & Wagner, 1972; Wagner & Rescorla, 1972).
With the use of this model, the implications of the concept of US surprisingness were extended
Copyright © 2017. American Psychological Association. All rights reserved.

to a wide variety of conditioning phenomena. The Rescorla–Wagner model had a huge impact
on the field of conditioning and learning (Siegel & Allan, 1996) and continues to be used in a
variety of areas of psychology, computer science, and neuroscience.
What does it mean to say that something is surprising? By definition, an event is surprising
if it is different from what is expected. If you expect a small gift for your birthday and get a car,
you will be very surprised. This is analogous to an unexpectedly large US. Likewise, if you
expect a car and receive a box of candy, you will also be surprised. This is analogous to an
unexpectedly small US. According to the Rescorla–Wagner model, an unexpectedly large US is
the basis for excitatory conditioning or increases in associative value. In contrast, an
unexpectedly small US is the basis for inhibitory conditioning or decreases in associative
value. A critical component of the model is the assumption that how surprised you are by a US
depends on all of the cues that are present on a conditioning trial.
Strong conditioned responding indicates a strong expectation that the US will occur,
whereas weak conditioned responding indicates a low expectation of the US. Using the
Domjan, Michael. The Essentials of Conditioning and Learning, American Psychological Association, 2017. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kuleuvenul/detail.action?docID=5116675.
Created from kuleuvenul on 2018-09-27 08:10:51.
 magnitude of the CR as a proxy for US expectancy, we can infer that the US is highly surprising
at the beginning of training and not at all surprising by the end when conditioned responding
has reached an asymptote or limit. Thus, distance from the asymptote of learning may be used
as a measure of US surprise.
The basic ideas of the Rescorla–Wagner model are expressed mathematically by using λ to
represent the asymptote of learning possible with the US that is being used and V to represent
the associative value of the stimuli that precede the US. The surprisingness of the US will then
be (λ − V). According to the Rescorla–Wagner model, the amount of learning on a given trial is
assumed to be proportional to (λ − V), or US surprisingness. The value of (λ − V) is large at
the start of learning because V (the associative value of the stimuli preceding the US) is close
to zero at this point. Hence, substantial increments in associative strength occur during early
conditioning trials. As the associative value of the cues that precede the US increases, the
difference term (λ − V) will get smaller, and less additional learning will occur.
Learning on a given conditioning trial is the change in the associative value of a stimulus.
This change can be represented as ΔV. Using these symbols, the idea that learning depends on
the surprisingness of the US can be expressed as follows:

ΔV = k(λ − V).

In this equation, k is a constant related to the salience of the CS and US and (λ − V) is a

measure of US surprise. ΔV = k(λ − V) is the fundamental equation of the Rescorla–Wagner
model.

Application to the Blocking Effect

The basic ideas of the Rescorla–Wagner model clearly predict the blocking effect. In applying
the model, it is important to keep in mind that expectations of the US are based on all of the
Copyright © 2017. American Psychological Association. All rights reserved.

cues available to the organism during the conditioning trial. As illustrated in Figure 6.3, the
blocking design first involves extensive conditioning of stimulus A so that the participants
acquire a perfect expectation that the US will occur based on the presentation of stimulus A.
Therefore, at the end of Phase 1, VA equals the asymptote of learning (VA = λ). In Phase 2,
stimulus B is presented together with stimulus A, and the two CSs are followed by the US.
According to the Rescorla–Wagner model, no conditioning of stimulus B will occur in Phase 2
because the US is now perfectly predicted by the presence of stimulus A: (λ − VA+B) = 0.
The control group receives the identical training in Phase 2, but for them the presence of
stimulus A does not lead to an expectation of the US. Therefore, the US is surprising for the
control group in Phase 2 and produces new learning.

Loss of Associative Value Despite Pairings With the US

The Rescorla–Wagner model is consistent with such fundamental facts of classical
Domjan, Michael. The Essentials of Conditioning and Learning, American Psychological Association, 2017. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kuleuvenul/detail.action?docID=5116675.
Created from kuleuvenul on 2018-09-27 08:10:51.
 conditioning as acquisition and the blocking effect. However, much of the importance of the
model has come from its unusual predictions. One such prediction is that under certain
circumstances the conditioned properties of stimuli will decline despite continued pairings
with the US. That is highly counterintuitive. Why should a CS lose associative value if it
continues to be paired with the US? The Rescorla–Wagner model predicts that stimuli will
lose associative value when they are paired with the US if there is an overexpectation of that
US.
The design of a US overexpectation experiment is outlined in Figure 6.4. In Phase 1,
stimuli A and B are paired with the same US (e.g., one pellet of food) on separate trials. This
continues until each of stimuli A and B predict perfectly the one food pellet US, or VA = VB =
λ. Phase 2 is then initiated. In Phase 2, stimuli A and B are presented simultaneously for the
first time, and the A+B stimulus compound is followed by the original one food pellet US.

FIG URE 6. 4

Design of the overexpectation experiment. In Phase 1, participants receive stimuli A and B, each paired
with the unconditioned stimulus (US; one food pellet). In Phase 2, stimuli A and B are presented together,
creating an expectation of more than the one pellet US. As a consequence, the associative values of
stimuli A and B each decrease in Phase 2.

When stimuli A and B are presented simultaneously at the start of Phase 2, the expectations
Copyright © 2017. American Psychological Association. All rights reserved.

based on the individual stimuli are assumed to add together, with the result that two food
pellets are predicted as the US (VA + VB = 2 λ). This is an overexpectation, because the US
remains only one food pellet. Thus, there is a discrepancy between what is expected (two
pellets) and what occurs (one pellet).
At the start of Phase 2, the participants find the US surprisingly small. To bring their
expectations of the US in line with what actually happens in Phase 2, the participants have to
decrease their expectancy of the US based on the individual stimuli A and B. Thus, stimuli A
and B are predicted to lose associative value despite continued presentations of the same US.
The loss of associative value is predicted to continue until the sum of the expectancies based
on A+B equals one food pellet. The predicted loss of CR to the individual stimuli A and B in
the overexpectation experiment is highly counterintuitive but has been verified repeatedly (e.g.,
Kehoe & White, 2004; Lattal & Nakajima, 1998; Sissons & Miller, 2009).

Domjan, Michael. The Essentials of Conditioning and Learning, American Psychological Association, 2017. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kuleuvenul/detail.action?docID=5116675.
Created from kuleuvenul on 2018-09-27 08:10:51.
 Conditioned Inhibition
The Rescorla–Wagner model treats the development of conditioned inhibition as another
illustration of the consequences of US over-expectation. Consider, for example, the standard
inhibitory conditioning procedure (see Figure 5.5). This procedure involves trials when the US
is presented (reinforced trials) and trials when the US is omitted (nonreinforced trials). On
reinforced trials, a conditioned excitatory stimulus (CS+) is paired with the US. On
nonreinforced trials, the CS+ is presented together with the conditioned inhibitory stimulus (CS
−).

FIG URE 6. 5
Copyright © 2017. American Psychological Association. All rights reserved.

Predicted associative values of conditioned excitatory stimulus (CS+) and conditioned inhibitory stimulus
(CS−) during the course of conditioned inhibition training (left panel) and extinction (right panel). During
conditioned inhibition training, when the CS+ is presented alone, it is paired with the unconditioned stimulus
(US); in contrast, when the CS+ is presented with the CS−, the US is omitted. The net associative value of
CS+ and CS− is the sum of the associative values of the individual stimuli. During extinction, the CSs are
presented alone, and the US never occurs.

To apply the Rescorla–Wagner model to the conditioned inhibition procedure, it is helpful

Domjan, Michael. The Essentials of Conditioning and Learning, American Psychological Association, 2017. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kuleuvenul/detail.action?docID=5116675.
Created from kuleuvenul on 2018-09-27 08:10:51.
 to consider reinforced and nonreinforced trials separately. To accurately anticipate the US on
reinforced trials, the CS+ has to gain conditioned excitatory properties. Excitatory conditioning
involves the acquisition of positive associative value and ceases once the organism predicts
the US perfectly on each reinforced trial. This is illustrated in the left panel of Figure 6.5.
On nonreinforced trials, both the CS+ and CS− occur. Once the CS+ has acquired some
degree of conditioned excitation (because of its presentation on reinforced trials), the organism
will expect the US whenever the CS+ occurs, even on nonreinforced trials. However, the US
does not happen on nonreinforced trials. This creates overexpectation of the US, similar to the
example in Figure 6.4. To accurately predict the absence of the US on nonreinforced trials, the
associative value of the CS+ and the value of the CS− have to sum to zero (the value
represented by no US). Given the positive associative value of the CS+, the only way to
achieve a net zero expectation of the US on nonreinforced trials is to make the associative
value of the CS− negative. Thus, the Rescorla–Wagner model explains conditioned inhibition
by assuming that the CS− acquires negative associative value (see the left panel of Figure 6.5).

Extinction of Conditioned Excitation and Inhibition

In an extinction procedure, the CS is presented repeatedly without the US. (I have a lot more to
say about extinction in Chapter 10.) Let us consider predictions of the Rescorla–Wagner model
for extinction. These predictions are illustrated in the right panel of Figure 6.5. When the CS+
is first presented without the US in extinction, there will be an overexpectation of the US
(because the US is predicted but does not occur). With continued presentation of the CS+ by
itself, the expectation elicited by the CS+ will gradually be brought in line with the absence of
the US by gradual reductions in the associative value of the CS+. This process will continue
until the associative value of the CS+ is reduced to zero.
The Rescorla–Wagner model predicts an analogous scenario for extinction of conditioned
Copyright © 2017. American Psychological Association. All rights reserved.

inhibition. At the start of extinction, the CS− has negative associative value. This may be
thought of as creating an underprediction of the US: The organism predicts less than the zero
US that occurs on extinction trials. To bring expectations in line with the absence of the US, the
negative associative value of the CS− is gradually reduced until the CS− ends up with zero
associative strength.

Problems With the Rescorla–Wagner Model

The Rescorla–Wagner model has stimulated a great deal of research and led to the discovery
of many new and important phenomena in classical conditioning (Siegel & Allan, 1996). Not
unexpectedly, however, the model has also encountered some difficulties since it was
proposed in 1972 (Miller, Barnet, & Grahame, 1995).
One of the difficulties with the model that became evident early on is that its analysis of
the extinction of conditioned inhibition is incorrect. As I pointed out in the previous section
Domjan, Michael. The Essentials of Conditioning and Learning, American Psychological Association, 2017. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kuleuvenul/detail.action?docID=5116675.
Created from kuleuvenul on 2018-09-27 08:10:51.
 (see Figure 6.5), the model predicts that repeated presentations of a conditioned inhibitor (CS
−) by itself will lead to loss of conditioned inhibition. This, however, does not occur (Witcher

& Ayres, 1984; Zimmer-Hart & Rescorla, 1974). In fact, some investigators have found that
repeated nonreinforcement of a CS− can enhance its conditioned inhibitory properties (see,
e.g., DeVito & Fowler, 1987; Hallam, Grahame, Harris, & Miller, 1992). Curiously, an
effective procedure for reducing the conditioned inhibitory properties of a CS− does not
involve presenting the CS− at all. Rather, it involves extinguishing the excitatory properties of
the CS+ with which the CS− was presented during inhibitory training (Best, Dunn, Batson,
Meachum, & Nash, 1985; Lysle & Fowler, 1985).
Another difficulty is that the Rescorla–Wagner model views extinction as the opposite of
acquisition, or the return of the associative value of a CS to zero. However, as I discuss in
Chapter 10, a growing body of evidence indicates that extinction is not simply the reversal of
acquisition. Rather, extinction appears to involve the learning of a new relationship between
the CS and US (namely, that the US no longer follows the CS).

ATTENTIONAL MODELS OF CONDITIONING

Given that classical conditioning has been studied for about a century, a comprehensive theory
must account for many diverse findings. No theory has been entirely successful in
accomplishing that goal. Nevertheless, interesting new ideas about classical conditioning
continue to be proposed and examined. Some of these proposals supplement the Rescorla–
Wagner model. Others are incompatible with the model and move the theoretical debate in
other directions.
North American psychologists have favored learning mechanisms like the Rescorla–
Wagner model that focus on changes in the surprise value or effectiveness of the US. In
contrast, British psychologists have approached phenomena such as the blocking effect by
postulating changes in how well the CS commands attention. The general assumption is that for
Copyright © 2017. American Psychological Association. All rights reserved.

conditioning to occur, participants must pay close attention to the CS. Procedures that disrupt
attention to the CS are expected to disrupt learning as well (Mitchell & Le Pelley, 2010).
How noticeable a stimulus is, or how much attention it commands, is called the salience
of the stimulus. Attentional theories differ in their assumptions about what determines the
salience of a CS on a given trial. Pearce and Hall (1980), for example, assumed that how much
attention a participant devotes to the CS on a given trial is determined by how surprising the
US was on the preceding trial (see also Hall, Kaye, & Pearce, 1985; McLaren & Mackintosh,
2000). Individuals have a lot to learn if the US was surprising to them on the preceding trial.
Therefore, under such conditions, CS salience will increase and they will pay closer attention
to the CS on the next trial. In contrast, if a CS was followed by an expected US, attention to
that CS will decrease.
An important feature of attentional theories is that they assume that the surprisingness of
the US on a given trial alters the degree of attention commanded by the CS on future trials. For
example, if Trial 10 ends in a surprising US, the salience of the CS will increase from Trial 10
Domjan, Michael. The Essentials of Conditioning and Learning, American Psychological Association, 2017. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kuleuvenul/detail.action?docID=5116675.
Created from kuleuvenul on 2018-09-27 08:10:51.
 to Trial 11. Thus, US surprisingness is assumed to have only a prospective or proactive
influence on attention and conditioning. This is an important difference from US–reduction
models such as the Rescorla–Wagner model, in which the surprisingness of the US on a given
trial determines what is learned on that same trial.
The assumption that the US on a given trial can change what is learned about a CS on the
next trial has received experimental support (e.g., Mackintosh, Bygrave, & Picton, 1977).
However, this same assumption has made it difficult for attentional models to explain other
findings. In particular, attentional models cannot explain the blocking that occurs on the first
trial of Phase 2 of the blocking experiment (see, e.g., Azorlosa & Cicala, 1986; Balaz,
Kasprow, & Miller, 1982; Dickinson, Nicholas, & Mackintosh, 1983). According to
attentional models, blocking occurs because in Phase 2 of the blocking experiment, the lack of
surprisingness of the US reduces attention to the added CS. However, such a reduction in
salience can occur only after the first Phase 2 trial. Therefore, attentional models cannot
explain the blocking that occurs on the first trial of Phase 2 of the blocking experiment.

TEMPORAL FACTORS AND CONDITIONED RESPONDING

Neither the Rescorla–Wagner model nor attentional models were designed to explain the
effects of time in conditioning. However, time is obviously a critical factor. One important
temporal variable is the CS–US interval. As I noted in Chapter 5, procedures with longer CS–
US intervals produce less responding (see Figure 5.2). This relation appears to be primarily a
characteristic of responses closely related to the US (e.g., focal search). If behaviors that are
ordinarily farther removed from the US are measured (e.g., general search), responding is
greater with procedures that involve longer CS–US intervals. Both findings illustrate that the
duration of the CS is an important factor in conditioning.
Another important temporal variable is the interval between successive trials. Generally,
more conditioned responding is observed with procedures in which trials are spaced farther
Copyright © 2017. American Psychological Association. All rights reserved.

apart. Of greater interest, however, is the fact that the intertrial interval and the CS duration act
in combination to determine responding. Numerous studies have shown that the critical factor
is the relative duration of these two temporal variables rather than the absolute value of either
one by itself (Balsam & Gallistel, 2009).
Consider, for example, an experiment by Holland (2000). The experiment was conducted
with laboratory rats, and food presented periodically in a cup was the US. Presentations of the
food were signaled by an auditory CS. Initially the rats went to the food cup only when the
food was delivered. However, as conditioning proceeded, they started going to the food cup as
soon as they heard the auditory CS. Thus, nosing of the food cup (a form of focal search)
developed as the CR. Each group was conditioned with one of two CS durations, either 10
seconds or 20 seconds, and one of six intertrial intervals (ranging from 15 seconds to 960
seconds). Each procedure could be characterized in terms of the ratio (I/T) between the
intertrial interval (I) and the CS duration, which Holland called the trial duration (T).
The results of the experiment are summarized in Figure 6.6. Time spent nosing the food
Domjan, Michael. The Essentials of Conditioning and Learning, American Psychological Association, 2017. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kuleuvenul/detail.action?docID=5116675.
Created from kuleuvenul on 2018-09-27 08:10:51.
 cup during the CS is shown as a function of the relative value of the intertrial interval (I) and
the trial duration (T) for each group of subjects. Notice that conditioned responding was
directly related to the I/T ratio. At each I/T ratio, the groups that received the 10-second CS
responded similarly to those that received the 20-second CS.

FIG URE 6. 6

Percent time rats spent nosing the food cup during an auditory conditioned stimulus (CS) in conditioning
with either a 10-second or 20-second trial duration (T) and various intertrial intervals (I) that created I/T
ratios ranging from 1.5 to 48.0. Data are shown in relation to responding during baseline periods when the
CS was absent. From “Trial and Intertrial Durations in Appetitive Conditioning in Rats,” by P. C. Holland,
2000, Animal Learning & Behavior, 28, p. 125. Copyright 2000 by Springer. Adapted with permission.
Copyright © 2017. American Psychological Association. All rights reserved.

Various interpretations have been offered for why conditioned responding is so strongly
determined by the I/T ratio (Gallistel & Gibbon, 2000; Jenkins, Barnes, & Barrera, 1981).
However, they all capture the notion that the I/T ratio determines how well the CS reduces
ambiguity about the next occurrence of the US (Balsam & Gallistel, 2009). The CS reduces
ambiguity about the US if it provides better information about the US than the background cues
of the experimental situation. With a high I/T ratio, the participant spends much more time in
the experimental context (I) than in the presence of the CS (T) before the US occurs. This
makes the CS much more informative about the next occurrence of the US than background
contextual cues, and therefore the CS comes to elicit a high level of conditioned responding.
The informational advantage of the CS over background contextual cues is lost if durations of I
and T are similar. As a consequence, less conditioned responding develops to the CS with low
I/T ratios.
Domjan, Michael. The Essentials of Conditioning and Learning, American Psychological Association, 2017. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kuleuvenul/detail.action?docID=5116675.
Created from kuleuvenul on 2018-09-27 08:10:51.
 Informational models of learning that were designed to explain the effects of the I/T ratio
are based on experiments that involve multiple conditioning trials (so that the organism can
learn about the durations of I and T and the rates of US delivery during each of these intervals).
However, there is substantial evidence that Pavlovian conditioning can occur in a single trial.
One-trial learning occurs readily in studies of fear conditioning, taste-aversion learning, and
sexual conditioning. Examples of one-trial learning are challenging for informational models of
conditioning.

THE COMPARATOR HYPOTHESIS

Studies of the I/T ratio and informational models of learning have emphasized that conditioned
responding depends not only on what happens during the CS but also on what happens in the
experimental situation in general. The idea that both of these factors influence what we
observed in conditioning experiments has been developed in greater detail by Ralph Miller
and his collaborators in the comparator hypothesis (R. R. Miller & Matzel, 1988; Stout &
Miller, 2007).
The comparator hypothesis is similar to informational models in assuming that conditioned
responding depends on the relationship between the target CS and the US, as well as on the
relationship between other cues in the situation (e.g., the background context) and the US. The
associative strength of other cues present during training with the target CS is especially
important. Another constraint of the comparator hypothesis is that it only allows for the
formation of excitatory associations with the US. Whether conditioned responding reflects
excitation or inhibition is assumed to be determined by the relative strengths of excitation
conditioned to the target CS compared with the excitatory value of the contextual cues that
were present with the target CS during training.
The comparator process is represented by the balance in Figure 6.7. In this figure, a
comparison is made between the excitatory value of the target CS and the excitatory value of
Copyright © 2017. American Psychological Association. All rights reserved.

the other cues that are present during the training of that CS. If CS excitation exceeds the
excitatory value of the contextual cues, the balance of the comparison will be tipped in favor of
excitatory responding to the target CS. As the excitatory value of the other cues becomes
stronger, the balance of the comparison will become less favorable for excitatory responding.
In fact, if the excitatory value of the contextual cues becomes sufficiently strong, the balance
may eventually tip in favor of inhibitory responding to the target CS.

Domjan, Michael. The Essentials of Conditioning and Learning, American Psychological Association, 2017. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kuleuvenul/detail.action?docID=5116675.
Created from kuleuvenul on 2018-09-27 08:10:51.
 FIG URE 6. 7

Illustration of the comparator hypothesis. Responding to the target conditioned stimulus (CS) is
represented by the reading of the balance. If the excitatory value of the target CS exceeds the excitatory
value of the other cues present during training of the target CS, the balance tips in favor of excitatory
responding. As the associative value of the contextual cues increases, the comparison becomes less
favorable for excitatory responding and may tip in favor of inhibitory responding.
Copyright © 2017. American Psychological Association. All rights reserved.

Unlike informational models, the comparator hypothesis emphasizes associations rather

than time. A simplified version of the comparator hypothesis, presented in Figure 6.8, involves
three different associations. The first association (Link 1 in Figure 6.8) is between the target
CS (X) and the US. The second association (Link 2) is between the target CS (X) and the
comparator contextual cues. Finally, there is an association between the comparator stimuli
and the US (Link 3). With all of these links in place, when the CS is presented it activates the
US representation directly (through Link 1) and indirectly (through Links 2 and 3). A
comparison between the direct and indirect activations of the US representation determines the
degree of excitatory or inhibitory responding that is observed.

Domjan, Michael. The Essentials of Conditioning and Learning, American Psychological Association, 2017. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kuleuvenul/detail.action?docID=5116675.
Created from kuleuvenul on 2018-09-27 08:10:51.
 FIG URE 6. 8

The associative structure of the comparator hypothesis. The target conditioned stimulus (CS) is
represented as X. Excitatory associations result in activation of the unconditioned stimulus (US)
representation, either directly by the target (Link 1) or indirectly (through Links 2 and 3). From “Comparator
Copyright © 2017. American Psychological Association. All rights reserved.

Mechanisms and Conditioned Inhibition: Conditioned Stimulus Preexposure Disrupts Pavlovian

Conditioned Inhibition but Not Explicitly Unpaired Inhibition,” by B. X. Friedman, A. P. Blaisdell, M. Escobar,
and R. R. Miller, 1998, Journal of Experimental Psychology: Animal Behavior Processes, 24, p. 454.
Copyright 1998 by the American Psychological Association.

It is important to note that the comparator hypothesis makes no assumptions about how
associations become established. Rather, it describes how CS–US and context–US
associations determine responding to the target CS. Thus, unlike US-modification and
attentional models, the comparator hypothesis is a theory of performance, not a theory of
learning.
An important corollary to the comparator hypothesis is that the comparison between CS–
US and context–US associations is made at the time of testing for conditioned responding. As a
consequence of this assumption, the comparator hypothesis makes the unusual prediction that
extinction of context–US associations after training of a target CS will enhance responding to
Domjan, Michael. The Essentials of Conditioning and Learning, American Psychological Association, 2017. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kuleuvenul/detail.action?docID=5116675.
Created from kuleuvenul on 2018-09-27 08:10:51.
 that target CS. This prediction has been confirmed repeatedly (e.g., Blaisdell, Gunther, &
Miller, 1999). US modification and attentional theories of learning cannot explain such results.
The fact that postconditioning extinction of contextual cues enhances responding to a target
CS indicates that responding to a target CS can be altered by changing the properties of
comparator cues. This type of result is called a revaluation effect. Research on the comparator
hypothesis has identified a growing number of revaluation effects. One of the more provocative
revaluation effects concerns the blocking effect.
In the critical phase of the blocking experiment, a new stimulus (B) is conditioned in the
presence of a previously conditioned CS (A). Because stimulus A is present when stimulus B
is being conditioned, stimulus A serves as the comparator for stimulus B. According to the
comparator hypothesis stimulus B will not elicit much conditioned responding because its
comparator (stimulus A) has high excitatory strength, tipping balance away from stimulus B. If
the lack of responding to stimulus B reflects this type of comparison, a revaluation
manipulation might unmask responding to stimulus B. In particular, the comparator hypothesis
predicts that participants will increase responding to stimulus B if the comparator (stimulus A)
is extinguished. Interestingly, this prediction has been confirmed in several experiments
(Blaisdell et al., 1999; Boddez, Baeyens, Hermans, & Beckers, 2011).
The comparator hypothesis has also been tested in studies of conditioned inhibition. The
hypothesis attributes inhibitory responding to situations in which the association of the target
CS with the US is weaker than the association of contextual cues with the US. The contextual
cues in this case are the stimuli that provide the excitatory context for inhibitory conditioning.
Interestingly, the hypothesis predicts that extinction of these conditioned excitatory stimuli after
inhibitory conditioning will reduce inhibitory responding. Thus, the comparator hypothesis is
unique in predicting that extinction of conditioned inhibition is best accomplished not by
presenting the CS− alone but by extinguishing the CS+ cues that provided the excitatory context
for inhibitory conditioning. This unusual prediction has been confirmed in several studies
(Best et al., 1985; Lysle & Fowler, 1985). (For additional revaluation effects, see McConnell,
Copyright © 2017. American Psychological Association. All rights reserved.

Urushihara, & Miller, 2010; Miguez, Witnauer, & Miller, 2012.)

Summary
Many instances of Pavlovian conditioning reflect the learning of an S–S association rather than
an S–R association. This conclusion is supported by experiments showing that the vigor of
conditioned behavior can be increased or decreased by changes in the value of the US (US
inflation or devaluation) after acquisition. A variety of mechanisms have been proposed to
explain Pavlovian learning. The Rescorla–Wagner model elaborated on the idea that the
suprisingness of the US is the driving force that produces learning and led to the discovery of
numerous novel learning phenomena. Attentional models addressed the same wide range of
phenomena as the Rescorla–Wagner model, but they had some of the same difficulties as that
model. Informational models focused on temporal variables in conditioning procedures, such
Domjan, Michael. The Essentials of Conditioning and Learning, American Psychological Association, 2017. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kuleuvenul/detail.action?docID=5116675.
Created from kuleuvenul on 2018-09-27 08:10:51.
 as the I/T ratio. The comparator hypothesis has been extended to a wider range of phenomena,
but it is a theory of performance rather than learning. It does not provide an explanation of how
associations are acquired in the first place. All of these models have been important in
directing our attention to previously ignored aspects of classical conditioning and each has
identified important novel conditioning variables and manipulations.

Suggested Readings
Balsam, P. D., & Gallistel, C. R. (2009). Temporal maps and informativeness in associative
learning. Trends in Neuroscience, 32, 73–78. http://dx.doi.org/10.1016/j.tins.2008.10.004
Delamater, A. R., Campese, V., LoLordo, V. M., & Sclafani, A. (2006). Unconditioned stimulus
devaluation effects in nutrient-conditioned flavor preferences. Journal of Experimental
Psychology: Animal Behavior Processes, 32, 295–306.
Hogarth, L., Dickinson, A., & Duka, T. (2010). Selective attention to conditioned stimuli in
human discrimination learning: Untangling the effects of outcome prediction, valence,
arousal, and uncertainty. In C. J. Mitchell & M. E. Le Pelley (Eds.), Attention and
associative learning (pp. 71–97). Oxford, England: Oxford University Press.
McLaren, I. P. L., & Mackintosh, N. J. (2000). An elemental model of associative learning: I.
Latent inhibition and perceptual learning. Animal Learning & Behavior, 28, 211–246.
http://dx.doi.org/10.3758/BF03200258
Stout, S. C., & Miller, R. R. (2007). Sometimes-competing retrieval (SOCR): A formalization
of the comparator hypothesis. Psychological Review, 114, 759–783.
http://dx.doi.org/10.1037/0033-295X.114.3.759 [Correction published in 2008,
Psychological Review, 115, 82.]

Technical Terms
Copyright © 2017. American Psychological Association. All rights reserved.

Asymptote
Comparator hypothesis
I/T ratio
Overexpectation
S–R learning
S–S learning
Salience
US devaluation
US inflation

Domjan, Michael. The Essentials of Conditioning and Learning, American Psychological Association, 2017. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kuleuvenul/detail.action?docID=5116675.
Created from kuleuvenul on 2018-09-27 08:10:51.