3
Principles and Applications
of Conditioning
Learning Objectives
After reading this chapter, you should be able to do the following:
• Recognize the principles of contiguity, frequency, and intensity and realize the extent to which
these principles are supported by experimental research.
• Define the four basic paradigms of classical conditioning: delay conditioning, trace conditioning,
simultaneous conditioning, and backward conditioning.
• Explain the phenomenon of sensory preconditioning and consider why the simultaneous
presentation of a CS and a US does not result in conditioning.
• Understand Rescorla’s research on contingency, and its suggestion that conditioning requires
more than just contiguity.
• Describe Garcia and Koelling’s research on taste-aversion learning, focusing on its implications
for the role of contiguity in conditioning and on the role of evolution in shaping conditioning.
• Discuss Kamin’s discovery of blocking, how he provided a cognitive account based on the
concept of surprise, and the implications of his work for contiguity.
• Identify the specific applications of conditioning principles, specifically in terms of how they
can be used to treat phobias, cigarette smoking, and alcoholism.
lie6674X_03_c03_087-114.indd 87
3/14/12 4:15 PM
CHAPTER 3
Section 3.1 The Laws of Association
We have seen that classical conditioning is not confined to relatively innocuous behaviors
such as salivation; classical conditioning affects some of the most important choices we
make, including what foods we eat, whether we become addicted to drugs, and whether
we feel certain emotions, such as fear or sexual arousal. If we understood the principles
of conditioning, therefore, it might allow us not only to better understand our behavior
but also, potentially, to change this behavior. Could we use conditioning, for example, to
reduce our fear in situations where it incapacitates us (for example, in job interviews or
dating)? Conversely, could we learn to increase our fear in situations where this emotion
might be advantageous—for example, could a smoker who wanted to quit make smoking
aversive by pairing the sensations of smoking with a painful consequence? In this chapter
we will try to answer these questions. We will begin by reviewing laboratory research
on what factors determine the strength of conditioning. We will then look at attempts to
apply these principles to problems such as phobias and alcoholism.
3.1 The Laws of Association
The British Associationists, sitting in their armchairs several centuries ago, identified a
number of laws of association, of which the most important were contiguity, frequency,
and intensity. We will begin our survey of the principles of conditioning by considering
the extent to which these laws have been supported by experiments.
Contiguity
The most important principle of association was
thought to be contiguity. The very concept of
an association—a bond between two events that
occur closely in time—implicitly assumes that
contiguity is necessary, and considerable effort
has been devoted to exploring the role of contiguity in classical conditioning.
The CS–US Interval
When played together as part of a piece,
musical notes are contiguous: that is, they
occur together in time.
lie6674X_03_c03_087-114.indd 88
As with most other aspects of conditioning, Pavlov was the first to investigate the role of contiguity in establishing a strong conditioned response.
He experimented with four different temporal
arrangements between the CS and the US; Figure
3.1 shows all four paradigms. (In learning, the
term paradigm is used to represent a standard or
typical sequence of events.) In delay conditioning, once the CS came on, it remained on until the
US was presented. In trace conditioning, the CS
was terminated before the US began.
3/14/12 4:15 PM
CHAPTER 3
Section 3.1 The Laws of Association
Figure 3.1: Paradigms for four varieties of classical conditioning
Delay
conditioning
US
Trace
conditioning
US
CS
CS
Simultaneous US
conditioning CS
Backward
conditioning
US
CS
Time
Pavlov was the first to examine the role of contiguity and utilized four different temporal arrangements
between the CS and the US. Here, the bars on the time line indicate periods during which a stimulus is
presented. Pavlov found that delay conditioning produced the strongest responding.
As the British Associationists would have predicted, Pavlov found that conditioning was
much stronger in the delay conditioning model, where the CS preceded the US, but both
were ultimately on at the same time. Subsequent research confirmed Pavlov’s findings.
In a typical study, Moeller (1954) looked at the effects of the CS–US interval on GSR (galvanic skin response) conditioning. He used a trace conditioning procedure in which a
brief burst of white noise (CS) was followed after a delay by a weak electric shock (US),
with the interval between the onset of the CS and the onset of the US set at either 250,
450, 1,000, or 2,500 milliseconds (ms). Moeller’s results are illustrated in Figure 3.2, which
shows that the strength of the conditioned response was greatest in the group with a 450ms gap, conditioning was weaker with a delay of 1,000 milliseconds (one second), and
virtually no conditioning occurred when the delay was increased to 2,500 milliseconds.
Both the optimum interval and the maximum interval that will sustain conditioning vary
somewhat for different responses (see Cooper, 1991, for a discussion of why this might
be), but as a general rule, the shorter the interval between the CS and US, the better the
conditioning.
lie6674X_03_c03_087-114.indd 89
3/14/12 4:15 PM
CHAPTER 3
Section 3.1 The Laws of Association
Figure 3.2: GSR conditioning as a function of the CS–US interval during training
CR strength
4
3
2
1
0
0
250
450
1000
2500
CS-US interval (milliseconds)
Moeller’s experiment in 1954 concluded that the shorter the interval between the CS and US, the better
the conditioning. This was determined by using a trace conditioning procedure in which a burst of noise
(CS) was followed by an electric shock (US). The interval between the CS and US was set at either 250,
450, 1,000, or 2,500 milliseconds.
Source: Adapted from Moeller, 1954.
Simultaneous and Backward Conditioning
You might have noticed one aspect of the data in Figure 3.2 that does not support this
claim, namely that when the CS–US interval was less than 450 ms, conditioning not only
didn’t improve, it became worse. Other experiments confirmed this finding: When the
CS and the US are presented in rapid succession, with delays of less than half a second,
conditioning is usually poor.
As with so many other aspects of conditioning, Pavlov was the first researcher to discover
this anomaly. One might think that a simultaneous conditioning procedure, in which the
CS and US come on at the same time (Figure 3.1) would produce the strongest conditioning, but Pavlov found virtually no conditioning in that arrangement. He found similarly
poor results with backward conditioning, in which the US is presented before the CS.
Even if the CS followed the US very closely, little conditioning occurred.
If contiguous stimuli are associated, as Pavlov and the British Associationists believed,
why is no association formed when a CS and a US are presented simultaneously? The
answer, it now appears, is that an association is formed—it’s just that this association does
not lead to the performance of a conditioned response. The clearest evidence that associations are formed when stimuli are presented simultaneously has come from research on a
lie6674X_03_c03_087-114.indd 90
3/14/12 4:15 PM
CHAPTER 3
Section 3.1 The Laws of Association
phenomenon called sensory preconditioning. In a typical experiment, neutral stimuli such
as a light and tone are presented together in an initial, preconditioning phase, and then one
of these stimuli—say the light—is paired with a US such as shock. The typical result is that
fear is conditioned not only to the light but also to the tone, even though the tone was not
present when shock was delivered. The implication is that when the tone and light were
presented together during the first phase, an association was formed between them:
Tone
light
When the light was later paired with shock, a second association would have been formed,
between the light and shock:
Light
shock
With the light and tone previously paired, subsequent presentation of the tone activated
the representation of the light in the brain, which in turn elicited fear:
Light
Fear
Tone
In most sensory preconditioning experiments, the sensory stimuli are presented sequentially in the preconditioning phase—in our example, the rats would first have heard the
tone and then seen the light. In a variant of this procedure reported by Rescorla (1980b),
however, the sensory stimuli—both tastes—were presented simultaneously. (The rats
simply drank a water solution flavored with the two tastes.) When one of the tastes was
then paired with illness, the rats developed an aversion to both tastes.
One possible interpretation of this result is that it was caused by generalization: When the
rats developed an aversion to one of the tastes, this aversion generalized to the other taste.
However, Rescorla showed that the first taste became aversive only if the tastes had been
presented together during preconditioning. This suggests that simultaneous presentation
forged an association between the two tastes, and it was this association that eventually
caused the aversion to be transferred from one taste to the other.
But if simultaneous presentation of a CS and US can result in the formation of an association, why doesn’t it result in conditioning? One possible explanation stems from the fact
that conditioning is an adaptive process whose purpose is to allow organisms to prepare
for forthcoming events. In most conditioning experiments the CS precedes the US, and the
CS thus allows the subject to take preparatory action. If a light is paired with a puff of air
to the eye, for example, then subjects can blink before the puff arrives, thereby protecting
their eye. If a light and an air puff are presented simultaneously, however, there is no time
to prepare. When responding would serve no purpose, as in simultaneous and backward
conditioning, no response is made. Put another way, conditioning seems to involve at
least two separate stages: In the first, an association is formed between the CS and the US;
lie6674X_03_c03_087-114.indd 91
3/14/12 4:15 PM
Section 3.2 Contingency
CHAPTER 3
in the second, presentation of the CS seems to trigger some kind of decision process that
determines whether that CS will elicit a response. Simultaneous conditioning thus provides us with our first hint that conditioning might not be as simple as it appears.
Frequency
A second variable that the British Associationists thought determined the strength of an
association between two events was the frequency of their pairing, or the number of times
they occur together over a period of time. Pavlov’s research on salivary conditioning
strongly supported this view (see Figure 2.3) and so has subsequent research. In general,
the strength of the conditioned response seems to
increase most during the early trials of conditioning, with the rate of increase gradually declining
as training continues, until performance eventually reaches a stable plateau, or asymptote.
Intensity
The third major principle proposed by the British
Associationists was that the strength of any association depends on the vividness or intensity
of the stimuli involved. Associations involving
emotional or traumatic events, for example, were
thought to be better remembered. If someone suffered intense pain while waiting for a wound to
be treated at a hospital, any future visit to that
hospital would elicit vivid memories of that pain.
Again, research on conditioning strongly supports this principle. Annau and Kamin (1961),
for example, found that the amount of fear conditioned to a tone depends on the intensity of
the shock that follows the tone (see Figure 2.11).
There is also evidence that the intensity of the
CS is of some importance, although this effect
appears weaker (see Grice, 1968). On the whole,
then, the armchair speculations of the British
Associationists have been impressively confirmed by research under controlled conditions.
Associative learning really does depend on contiguity, frequency, and intensity.
The frequency with which one counts
sheep in order to sleep may strengthen the
association between sheep and sleepiness.
(Presumably, this would also hold true for
sheep who count people to fall asleep!)
3.2 Contingency
Until the 1960s, all the available evidence converged on a coherent and satisfying picture
of conditioning in which the foundation stone was contiguity: If two events are contiguous—that is, occur closely together in time—then an association will be formed between
lie6674X_03_c03_087-114.indd 92
3/14/12 4:16 PM
CHAPTER 3
Section 3.2 Contingency
them. The strength of this association might be
modulated by other factors such as the intensity
of the stimuli involved. Fundamentally, though,
conditioning appeared to be a simple process in
which associations were automatically formed
between contiguous events. In 1966, however,
two landmark papers were published that posed
a fundamental challenge to traditional views of
the role of contiguity and unleashed an intellectual ferment—revolution would not be too
strong a word—that is still continuing.
The Concept of Contingency
The first of these papers was the work of Robert
Rescorla, then a graduate student at the University of Pennsylvania. In his paper, Rescorla sugThis rice farmer depends on heavy rainfall
gested that contiguity between two events was
to irrigate crops. In deciding whether to
not sufficient for conditioning; something more
pay for a weather forecasting service,
was needed. Specifically, he suggested that a CS
it would be important for the farmer
must not only be contiguous with a US but must
to consider not only the probability of
also be an accurate predictor of the occurrence of
rain when it was forecast but also the
the US. To understand what he meant by this,
probability when it was not forecast.
let’s take a look at the following example: Suppose you were in a room where you occasionally heard a tone that lasted for two minutes.
And further suppose that you also occasionally received electric shocks. (Not a pleasant
example, but it will be useful for reasons that will become clear.) Figure 3.3 shows two
possible variants of this situation. In situation A, a shock is always presented at some
point while the tone is on, but
shock is never presented in the
tone’s absence. In this situation
the tone is a good predictor: It
warns that you that a shock is
imminent.
Now consider situation B. Here
too, shocks occur during the
tone, but shocks also occur in the
absence of the tone. Indeed, the
likelihood of receiving a shock
is just as great in the absence
of the tone as in its presence.
In this situation the tone has no
predictive value: When a tone
is on, you are no more likely to
receive a shock than when it is
off, and therefore, the tone does
not help you predict when you
will receive a shock.
lie6674X_03_c03_087-114.indd 93
The presence of dark storm clouds in the sky is often a sign
that rain will soon fall; although dark clouds do not always
signal rain, there is a high likelihood that if one occurs, the
other will soon follow. And, equally important, when clouds
are not present, rain is not likely to occur. When both of these
conditions are satisfied, so that rain is much more likely in the
presence of clouds than in their absence, we say that there is a
high level of contingency between the two events.
3/14/12 4:16 PM
CHAPTER 3
Section 3.2 Contingency
Figure 3.3: CS-US contingencies
CS
US
Perfect
predictor
A
Random
group
B
In this experiment, Rescorla studied the effects of the predictor between the US and the CS; he
suggested that the CS must be contiguous and act as a predictor of the occurrence of the US.
Although we have not shown it in this figure, it is also possible to imagine an intermediate
situation in which the tone had some predictive value but the prediction was not perfect.
For example, suppose the shock occurred in both the presence and absence of the tone, but
was more likely when the tone was present. Clearly the tone in this situation would have
some predictive value, although you wouldn’t be certain about what was going to happen.
What these examples illustrate is that the predictive value of a CS can vary widely—at
one extreme, a tone might be a perfect predictor of when shock will occur; at the other, it
might be no help at all. It would be quite useful, therefore, if we had some way of measuring predictive value. In fact, there are several such measures, but one of the most useful is a mathematical statistic called a contingency. Because contingencies are defined in
terms of probabilities, however, we need to start by quickly reviewing what we mean by
a probability.
A probability is just a mathematical expression of the likelihood that an event will occur. If
there is no chance of an event occurring, its probability is said to be 0; if the event is certain
to occur, its probability is said to be 1.0. Suppose that a tone was presented 100 times, and
that every one of these presentations was followed by a shock. In that case, the probability
of a shock following the tone would be 1.0. Let us further suppose that the shock never
occurs in the absence of the tone. The probability of a shock in the absence of the tone
would then be 0. This is the situation shown in Figure 3.3A—shock would be much more
likely when the tone was on than when it was off.
Now consider the situation shown in Figure 3.3B, in which shocks occurred in the absence
of the tone as well as in its presence. If the probability of a shock in the absence of the tone
was the same as in its presence, the tone would have no predictive value; its onset would
not signal any greater probability of shock.
lie6674X_03_c03_087-114.indd 94
3/14/12 4:16 PM
CHAPTER 3
Section 3.2 Contingency
Even though tone is followed by shock equally often in both of our examples, its predictive value would be very different. One way of capturing this idea is to say that the predictive value of a CS depends on the extent to which the probability of the US changes when
the CS is present. And that is what a contingency statistic measures. The contingency
between a CS and a US is defined as the difference between the probability of the US when
the CS is present and when it is absent (Allan, 1980). The formula would look like this:
Contingency
=
(
(– (
probability of US
in presence of CS
probability of US
in absence of CS
(
The greater the difference—the more the probability of the US in the presence of the CS
exceeds that in its absence—the greater the contingency.
You can check how well you understand the concept of contingency by considering the
following hypothetical example. Suppose that you are a farmer who has just moved to a
new county, and you need to be able to predict the probability of rain to decide whether
to plant your corn. A salesperson for a weather forecasting company approaches you and
tells you that the company has developed a new forecasting system that is far more accurate than any existing method. As proof, the salesperson shows you evidence that last
year the company predicted rain on 100 days and it actually rained on 95 of those days.
Should you buy the new forecasting service?
The answer, from the point of view of contingency, is no—or, at least, not necessarily. To
determine the value of the company’s predictions, you need to know not only the probability of rain when it was forecast, but also the probability when it was not forecast. Suppose, for example, that you had just moved to an area where it always rains on 95 days out
of 100. In this case, the company’s predictions would clearly be of very little aid in deciding whether rain was imminent. To evaluate the accuracy of any forecast or prediction, in
other words, you need to consider not only how often the predicted event occurs when it
is predicted but also how often it occurs when it is not predicted. If these probabilities are
similar, then the prediction will not help you very much.
The Role of Contingency in Conditioning
A subject in a classical conditioning experiment faces a problem similar to that of the
farmer who wants to predict rain. Consider a rat that suddenly becomes ill. If this illness
were caused by food it had eaten earlier, it would obviously be advantageous for the rat to
avoid that food in the future. In searching for a cue that could predict illness, however, the
rat (like the farmer) might be seriously misled if it relied solely on contiguity. Just because
the rat becomes ill after eating lima beans, for example, doesn’t necessarily mean it was
the lima beans that made the rat ill; if the rat becomes ill on days when it doesn’t eat lima
beans as well as on days when it does, there would be no point to its avoiding lima beans
in the future. In seeking to identify the true cause of an event, in other words, animals and
humans would do better if they considered the contingency between two events as well as
their contiguity.
lie6674X_03_c03_087-114.indd 95
3/14/12 4:16 PM
CHAPTER 3
Section 3.2 Contingency
Rescorla, as mentioned earlier,
was the first to wonder about
the role of contingency. What
would happen, he asked, if a
tone and shock were presented
contiguously, as in most fearconditioning experiments, but
the shock was also presented in
the absence of the tone, thereby
eliminating their contingency?
Although Rescorla’s initial work
on contingency was published in
1966, we will look at the results
of an experiment he reported in
Taste is a very refined sense, as this professional coffee taster
1968. Figure 3.4 illustrates the
can attest. It allows us to detect foods that may be spoiled
design of this experiment. In the
simply by the way they taste: for example, spoiled milk tastes
random group, rats received a
sour and contaminated apple juice can have a vinegary flavor.
series of tones and shocks delivIt turns out that taste cues are more readily associated with
ered totally at random. Subillness than visual cues.
jects in the contingency group
received tones whenever their counterparts in the random group did, and they also received
some—but not all—of the shocks delivered to subjects in the random group. Specifically,
they received the shocks given to the random group while the tone was present but not
when the tone was absent. Both groups thus received the same number of tones and the
same number of pairings of the tone with the shock.
Figure 3.4: Rescorla’s contingency experiment of 1968
2 min
tone
shock
Random
group
Contingency
group
In this figure a tone is indicated by a pink bar and a shock by an orange bar. An orange bar inside a pink
bar indicates that the shock occurred while the tone was on. In the Random group, shocks were presented
at totally random intervals, sometimes during the tone and sometimes in its absence. The Contingency
group also received the shocks presented during the tone, but not those presented in its absence.
lie6674X_03_c03_087-114.indd 96
3/14/12 4:16 PM
Section 3.3 Preparedness
CHAPTER 3
How should conditioning in the two groups compare? If conditioning depends simply on
contiguity, then conditioning should be equal, because both groups had the same number
of tone-shock pairings. If contingency also matters, however, we should expect conditioning only in the contingency group. In accordance with this prediction, Rescorla found powerful conditioning in the contingency group and none whatsoever in the random group.
In other experiments reported in his 1968 paper, Rescorla manipulated the degree of contingency between the CS and the US and found that conditioning depended on the precise
level of contingency: The greater the contingency, the stronger the conditioning. In one
sense, this is hardly surprising; it is just a fancy way of saying that conditioning depends
on the extent to which the CS is a good predictor of the US.
3.3 Preparedness
The second seminal paper of 1966 was by Garcia and Koelling, and they also challenged
the assumption that any two events that were contiguous would be associated. In particular, these researchers challenged the idea that it did not matter what stimulus was chosen
as a CS. Pavlov had claimed, “Any natural phenomenon chosen at will may be converted
into a conditioned stimulus . . . any visual stimulus, any desired sound, any odor, and
the stimulation of any part of the skin” (1928,
p. 86). Subsequent research almost universally
supported Pavlov’s position—until, that is, the
publication of Garcia and Koelling’s paper.
Taste-Aversion Learning
Their experiment had its origins in naturalistic
observations of animal behavior—in particular,
in observations of a phenomenon in rats called
bait-shyness. Rats, it turns out, resist human
efforts to exterminate them. When they encounter a novel food, they tend to take only the smallest taste at first; if it turns out to be poisoned
bait but they survive, they rarely if ever touch
that food again. Classical conditioning provides
a possible explanation for the rats’ avoidance of
the bait: Ingestion of the poisoned bait produces
nausea, and this reaction becomes conditioned to
the smells and tastes that precede the nausea. On
future occasions, the rats avoid the bait because
its odor or taste makes them ill. As we saw in
Chapter 2, this phenomenon is known as tasteaversion learning.
As plausible as this explanation is, it cannot
account for one aspect of the rats’ behavior.
lie6674X_03_c03_087-114.indd 97
If this mouse ate a piece of poisoned
cheese and became ill, it would be
extremely reluctant to eat the same food
in the future, no matter how much cheese
it was tempted with; this phenomenon is
known as “taste-aversion learning.”
3/14/12 4:16 PM
Section 3.3 Preparedness
As part of Garcia and Koelling’s research on taste-aversion
learning, rats were given flavored water to drink from a tube
and then made ill.
CHAPTER 3
Although the poisoned rats later
avoided the bait, they showed
no reluctance to return to the
place where they had been poisoned to consume other foods
found there. If associations form
between any contiguous events,
then we should expect place
cues to be associated with illness as readily as taste and odor
cues, but this did not appear to
be happening. Was it possible
that the rats could associate
nausea with tastes, but not with
visual cues?
To test this hypothesis under
controlled laboratory conditions, Garcia and Koelling allowed rats to taste distinctly flavored water from a drinking
tube that was wired so that every lick produced not only water but a brief noise and light
flash. Following exposure to this taste-noise-light compound, the rats received a dose of
radiation sufficient to make them ill. Then, on a test trial, the rats were exposed to each
of the compound stimuli separately, to determine which ones had become aversive. A
lick produced either the flavored water or plain water plus the noise-light compound. As
shown in Figure 3.5a, the rats were now very reluctant to drink the flavored water, but
they had no such compunctions about the bright-noisy water. These naturalistic observations suggested that nausea could be conditioned to gustatory cues but not visual ones.
lie6674X_03_c03_087-114.indd 98
3/14/12 4:16 PM
CHAPTER 3
Section 3.3 Preparedness
Water intake
Figure 3.5: Water intake before and after conditioning
pre-cond
post-cond
“bright-noisy” water
Water intake
a.
X ray
b.
Electric
shock
pre-cond
post-cond
Time
Water intake before (pre) and after (post) conditioning: (a) when X rays were used as the US; (b) when
shock was used as the US. The green bars represent intake of the flavored water; the orange bars
represent intake of plain water when licking produced a noise and light.
Source: Based on Garcia & Koelling, 1966
An alternative explanation, however, was possible: Perhaps the noise and light used in
the experiment were simply too faint to be detected, so conditioning would not have
occurred with any US. To test this hypothesis, Garcia and Koelling repeated their experiment with the same compound CS, but with electric shock as the US instead of X rays.
The results for the suppression test are shown in Figure 3.5b, which illustrates that the
audiovisual stimulus produced suppression of drinking and the taste stimulus had no
effect. We thus face this strange situation in which nausea cannot be conditioned to a
noise, nor fear to a taste, even though each of these conditioned stimuli is easily associated with the other US.
Subsequent research established that it is possible to associate taste with shock and noise
with illness, but it is much more difficult, requiring many more trials (for example, Best,
Best, & Henggeler, 1977). Seligman (1970) coined the term preparedness to refer to the
fact that we seem prepared to associate some CS–US combinations more readily than
others.
lie6674X_03_c03_087-114.indd 99
3/14/12 4:16 PM
Section 3.3 Preparedness
CHAPTER 3
Implications
Garcia and Koelling’s experiment has proved to be one of the most influential studies on
learning ever published. In part, this is because their study provided the first clear evidence
for the existence of taste-aversion learning, a process that plays a major role in determining
food preferences. (See Chapter 2.) In addition to its practical significance, taste-aversion
learning proved to have important implications for how psychologists view learning, and,
indeed, for our understanding of scientific discovery. Before we complete our discussion
of Garcia and Koelling’s work, therefore, we will look briefly at these implications.
The Role of Contiguity in Associative Learning
According to the traditional view, all that matters in conditioning is contiguity: If two events
are contiguous, then they will be associated. The evidence for preparedness, however, clearly
shows that this is not the case. In the taste-aversion experiment, noise was just as contiguous with illness as taste was, but this contiguity did not result in learning. Contiguity, therefore, is not sufficient for learning to take place. Other evidence has shown that contiguity
is not even necessary. In the Garcia and Koelling experiment, there was a delay of at least
20 minutes between the presentation of the taste and the animals’ becoming ill; in a subsequent, memorable experiment
by Etscorn and Stephens (1973),
conditioning occurred despite a
delay of 24 hours. Clearly, conditioning is not due simply to the
linking of events that happen to
occur contiguously: Some other
process or processes must be
involved. Garcia and Koelling’s
experiment thus contributed to
a major theoretical shift in the
way we view conditioning—
from a simple process to one of
considerable sophistication and
complexity. We will examine
this shift in greater detail in sub- Just because two events occur together in time doesn’t mean
sequent chapters.
that a link will be forged between them.
The Uniformity of Conditioning
Pavlov, and most of the Western psychologists who followed him, viewed conditioning as
an entirely general process. No matter what CS was paired with what US, the same associative process would be involved, and the principles of conditioning would thus also be the
same. The principles of taste-aversion learning, however, are not the same as those of, say,
salivary conditioning. As we have seen, it is easy to associate a light with food but very difficult to associate that same light with illness. Moreover, the role of contiguity is also different: In salivary conditioning, the longest CS–US interval at which conditioning will occur
is on the order of minutes, whereas in taste-aversion learning it can be as long as 24 hours.
And whereas salivary conditioning is a fairly slow process, requiring many trials, strong
taste aversions can be learned in just one or two trials. (For a review, see Domjan, 1980.)
lie6674X_03_c03_087-114.indd 100
3/14/12 4:16 PM
Section 3.3 Preparedness
CHAPTER 3
These differences should not be exaggerated. Regarding contiguity, for example, it is true
that the longest interval at which conditioning will occur is longer in taste-aversion learning than in salivary conditioning, but shorter intervals still produce stronger conditioning
(Andrews & Braveman, 1975). Nevertheless, the principles of conditioning clearly do vary
for different responses, if only in degree.
The Adaptive Value of Preparedness
Why should this be? To answer this question, it is helpful to begin by considering why
classical conditioning occurs at all.
The Value of Conditioning
In discussing Pavlov’s research, we referred repeatedly to his view that the process of
conditioning has evolved because it helps animals survive in their natural environments.
One way of thinking about conditioning is as a means of identifying stimuli that cause
or predict important events: If an animal knows where food is available, for example, or
which of the other animals in its vicinity is likely to attack it, then it can use this information to guide appropriate action. Culler (1938) expressed this view with some eloquence.
[Without a signal] the animal would still be forced to wait in every case for the
stimulus to arrive before beginning to meet it. The veil of the future would hang
just before his eyes. Nature began long ago to push back the veil. Foresight proved
to possess high survival-value, and conditioning is the means by which foresight is
achieved. (Culler, 1938, p. 136)
Salivary conditioning provides one example of the advantages of foresight: If a dog knows
when food is coming, it can begin to salivate beforehand, and this will allow it to consume
the food more quickly—not a small advantage when predators or other hungry dogs are
around. Similarly, if a rat learns to freeze whenever it sees a predator, this freezing may
enhance its chance of escaping
detection and thus surviving.
In a world where intense competition exists over every food
source, a hyena’s ability to salivate to certain cues before food
presents itself provides an adaptive advantage.
lie6674X_03_c03_087-114.indd 101
In these examples, we can
only speculate about the functional value of the conditioned
response, but in some cases
we have direct evidence. One
example concerns the conditioning of sexual arousal. In an
experiment by Zamble, Hadad,
Mitchell, and Cutmore (1985),
male rats were given access to
a sexually receptive female. In
one group, the female’s appearance was preceded by a signal;
in the other group, it was not.
When the female’s appearance
3/14/12 4:16 PM
Section 3.4 Blocking
CHAPTER 3
was signaled, males initiated and completed copulation more quickly. However unromantic such behavior might seem, a male that approaches a female faster is likely to have
an advantage over competing suitors, and a male that finishes more quickly will have
spent less time in a position in which it is highly vulnerable to attack. The conditioning
of sexual arousal will give this male a significant advantage in reproducing, thus ensuring that its genes—including those responsible for the susceptibility to conditioning of
sexual arousal—will be passed on to succeeding generations (Hollis, Pharr, Dumas, Britton, & Field, 1997).
The Value of Preparedness
The principles of taste-aversion
learning seem to differ from
other forms of conditioning,
because these variations contribute to the species’ survival. Consider a rat that became ill after
eating rancid meat. If the only
learning system it possessed was
an all-purpose mechanism that
associated all contiguous events,
then it would have developed an
aversion to all the stimuli present when it became ill. In other
words, the rat might have been A rat would be unlikely to associate illness with a singing bird;
equally likely to develop an taste would be a more adaptive predictor of illness in this case.
aversion to the sound of a drill
or to the smell of flowers or perhaps to a bird that was singing nearby. If it thereafter tried
to escape every time it heard a singing bird, it would be more likely to die of exhaustion
than to prosper. The pressures of natural selection would thus favor rats that associated illness with preceding tastes, rather than with irrelevant lights or sounds (see also Wilcoxon,
Dragoin, & Kral, 1971; Beecher, 1988).
3.4 Blocking
The 1960s were a difficult time for the principle of contiguity. First, Rescorla showed that
temporal contiguity between a CS and a US is not sufficient to ensure conditioning; the CS
must also be a good predictor of the US. Then Garcia and Koelling showed that even valid
predictors are not always conditioned. In 1969, a third event undermined still further the
traditional view of contiguity and suggested an alternative analysis to replace it. This
event was the publication of a paper by Leo Kamin.
lie6674X_03_c03_087-114.indd 102
3/14/12 4:16 PM
CHAPTER 3
Section 3.4 Blocking
Kamin’s Research on Blocking
Kamin (1969) gave rats fear-conditioning trials in
which two stimuli, a noise (N) and a light (L), were
paired with an electric shock. In a control group,
the noise and the light came on together, remained
on for three minutes, and were immediately followed by the shock. To assess conditioning to
the light, Kamin used a conditioned emotional
response (CER) test in which the light was presented while the rats pressed a bar to obtain food.
The suppression ratio for the light was 0.05, indicating substantial fear conditioning. (Recall from
Chapter 2 that a suppression ratio of 0.50 indicates
no fear and zero indicates maximal fear.) In other
words, fear was strongly conditioned to the light
for the rats in the control group.
Just as a hand can block the light from the
sun, prior conditioning to a stimulus can
block conditioning to a second conditioned
stimulus that now accompanies it.
blocking group
control group
Kamin was interested primarily in a second
group, though. The subjects in this second group
received the same pairings of the noise-light
compound with shock, but these compound trials were preceded by trials in which the noise by
itself was paired with shock.
Pretraining
Conditioning
N
NL
shock
NL
shock
shock
This first phase produced substantial fear conditioning to the noise. For subjects in the
blocking group, therefore, the noise already elicited fear when the compound trials began.
What effect should we expect this to have on conditioning to the light?
According to a contiguity analysis, fear should be conditioned to the light in both groups,
because, in both, the light was repeatedly and contiguously paired with the shock. The
results for the two groups, however, proved to be very different. The suppression ratio
in the control group was 0.05, meaning they showed substantial fear conditioning to the
light; however, the ratio for subjects in the blocking group ? those who were given preliminary conditioning to the noise ? was 0.45, a statistic only barely distinguishable from the
0.50 level representing no fear. In other words, prior conditioning to the noise had blocked
conditioning to the light. Kamin called this phenomenon, in which prior conditioning to
one element of a compound prevents conditioning to the other element, blocking.
lie6674X_03_c03_087-114.indd 103
3/14/12 4:16 PM
Section 3.4 Blocking
CHAPTER 3
Surprise!
To account for blocking, Kamin proposed an intriguing explanation. When an important
event such as shock occurs, he said, animals search their memories to identify cues that
could help to predict the event in the future. Imagine that a rat foraging for food in a forest
is suddenly attacked by an owl. If the rat survives the attack, it will search its memory to
identify cues that preceded the attack, helping it avoid such an event in the future. If the
rat had seen the owl in a tree just before the attack, for example, then the next time it saw
an owl it would dive for cover.
Kamin’s first assumption, then,
was that unconditioned stimuli trigger memory searches
for predictive cues. His second assumption was that such
searches require effort. In tasteaversion
conditioning,
for
example, we have seen that animals may develop an aversion
to foods consumed as much as
24 hours before they became
ill, indicating that any memory
search must cover events spread
over at least this time period.
Such a search would require According to Kamin’s theory, if a bird had already been
considerable time and effort, conditioned to fly away when it heard a gunshot, it would
and Kamin speculated that in probably not be conditioned to fly away on seeing a hunter,
order to save energy, subjects if it now saw the hunter at the same time as it heard the
would scan their memories gunshot; prior conditioning to the first cue (the noise) would
only if the US were unexpected block conditioning to the second cue (the hunter).
or surprising. If the US were
expected, then by definition some cue predicting its occurrence must already have been
available, so that no further search would be needed.
To see how this analysis can account for blocking, consider first the control group that
received only the compound trials. The first shock would have been unexpected and
would have triggered a memory search for the cause. The rats would remember the preceding noise and light, and thus both cues would be associated with the shock.
Similarly in the blocking group, presentation of the shock during the preliminary phase
would have surprised the rats and thus triggered a memory search in which the rats recalled
the noise and associated it with the shock. When the noise was then presented as part of the
noise-light compound, the rats would have expected the shock to follow and hence would
not have been surprised. As a result, they would not have searched their memories and
thus would not have learned about the relationship between the light and the shock.
According to Kamin, then, blocking occurs because the US is already expected. To test
this analysis, he used an ingenious design in which he changed the US used during the
compound trials so that its presentation would come as a surprise. As before, a noise was
paired with shock during preconditioning, but during conditioning the shock presented
lie6674X_03_c03_087-114.indd 104
3/14/12 4:16 PM
CHAPTER 3
Section 3.4 Blocking
at the end of each noise-light compound was unexpectedly followed by a second shock
five seconds later:
Noise
Shock Noise-Light
Shock . . . Shock
During pretraining, the rats would have learned that the noise was followed by a shock,
which would have produced strong fear conditioning to the noise; on the compound trials,
therefore, no conditioning to the light would have occurred, because the rats would have
expected the first shock. The second shock, however, would have been a surprise. The rats
should therefore search their memories for possible causes, notice the light, and associate
it with the shock. And that is what Kamin found: When the light was later presented on
its own, it produced powerful fear in the group that had received two shocks. To sum up,
conditioning seems to depend on whether the US is surprising, as any change in a US that
makes it surprising—making it more aversive or less aversive—produces conditioning.
Implications
Research on contingency, preparedness, and now blocking have shattered the traditional
view of conditioning, which posited that contiguity between a CS and a US was sufficient
for the formation of an association. Instead, conditioning seems to be concentrated on
stimuli that are good predictors of a US—either because, in the evolutionary history of
the species, these stimuli have proven to be valid predictors (preparedness), or because
they currently provide useful information (contingency). Blocking fits the same pattern,
as fear was conditioned to the stimulus that was the best predictor of the US over time—
the noise that preceded every shock, rather than the light that preceded only some. (See
also Wagner, Logan, Haberlandt, & Price, 1968.)
Conditioning, in other words, turns out to be a
beautifully adaptive system that targets the cues
most likely to be the true causes or predictors of
important events.
We now know that conditioning is not a
simple or automatic process; it involves
both memory and attention.
lie6674X_03_c03_087-114.indd 105
Kamin not only identified a crucial problem—how
does the brain identify the best predictor of the US,
if not by contiguity?—he also provided a persuasive solution. Conditioning, in his account, is not
limited to cues that are physically present when
a US is encountered; instead, the use of memory
allows the search for predictors to be broadened
to cues that occurred seconds, minutes, or—in
the case of taste aversions—even hours earlier.
Memory thus plays a central role in conditioning,
and this realization helped to produce a fundamental shift in how psychologists viewed conditioning. It was understood now to be a far more
complex process, one that also involved cognitive
processes such as memory and attention. Conditioning might not be as complex as language, but
neither was it as primitive as previously believed.
We will discuss this shift further in Chapter 4, but
3/14/12 4:16 PM
Section 3.5 Applications
CHAPTER 3
if you would like to find more material dealing specifically with the role of memory in conditioning, some excellent research is available in papers by Revusky (1971), Wagner, Rudy,
and Whitlow (1973), Bouton and Nelson (1998), and Manns, Clark, and Squire (2002).
3.5 Applications
By studying conditioning in the highly controlled environment of the laboratory, Pavlov
and his successors hoped to tease apart the complex processes involved in the formation
of associations. The laboratory, though, was not an end in itself—the ultimate goal was
always to apply the knowledge gained in the laboratory to helping people in real life. In
this section, we will examine attempts that have been made to apply conditioning principles to such problems as phobias, cigarette smoking, and alcoholism.
Phobias
The first speculations about the possibility of applying classical conditioning principles
to practical problems appeared in the study by Watson and Rayner (1920), discussed in
Chapter 2, in which they conditioned “little Albert” to fear a rat by pairing the rat with
presentations of a loud noise. At the end of their published report, they suggested that
fear conditioning in children might explain many of the phobias and anxieties found in
adults. If so, it might also be possible to use conditioning principles to eliminate these fears.
The Origin of Phobias
One way to assess the claim that adult phobias are
the result of conditioning is to interview phobics
about the circumstances that led to their phobias.
On the whole, studies that have done this have
supported the claim. In one such study, by Öst
and Hugdahl (1981), 58% of those interviewed
were able to recall traumatic incidents that triggered their phobias. What, though, of the 42%
who could not—how did their phobias arise? In
some cases the cause appeared to be vicarious
learning, in which individuals learn that a stimulus is dangerous because they see someone else
being injured. In one such case, a boy developed
a severe dental phobia when he accompanied a
friend to the dentist and the dentist’s drill accidentally punctured his friend’s cheek (Öst, 1985,
cited by Barlow & Durand, 1995).
At first glance vicarious learning and conditioning might appear to be alternative explanations,
but vicarious learning can be understood as a kind
of conditioning if we assume that animals and
lie6674X_03_c03_087-114.indd 106
According to the principle of vicarious
learning, we can develop a phobia not only
by experiencing pain ourselves but also by
witnessing others experiencing that pain.
Seeing a scene like the one above might
not encourage future trips to dentists.
3/14/12 4:17 PM
Section 3.5 Applications
CHAPTER 3
humans are innately programmed to become distressed when they see another member
of their species hurt. If a young monkey sees his mother becoming frightened when she
encounters a snake, for example, it would clearly be advantageous for the infant to learn
to associate snakes with fear. In the course of evolution, the sight of others’ distress could
thus have become an unconditioned stimulus for anxiety—indeed, the boy who saw his
friend injured became so distressed that he ran from the dentist’s office. (For more direct
evidence that the sight of others in distress can lead to the conditioning of fear, see Mineka
& Cook, 1986, and Gerull & Rapee, 2002.)
What about cases in which phobics cannot recall any traumatic incident, whether involving themselves or others? One possibility is that such incidents occurred but were forgotten. This might at first seem implausible—surely someone who experienced a trauma
severe enough to produce a phobia would remember it?—but people’s memories for painful incidents are surprisingly poor. In one study cited by Loftus (1993), a survey of 1,500
people who had been hospitalized within the preceding year revealed that 25% could not
recall this hospitalization! Moreover, memory seems to be particularly poor for incidents
experienced when we are young, which is when many phobias develop. (See, for example, Henry, Moffitt, Caspi, Langley, & Silva, 1994.)
The issue of how phobias arise is still controversial, but it does look as if a very substantial
proportion of specific phobias—those involving specific stimuli such as snakes and spiders—really are due to conditioning, as Watson and Rayner had suggested. (For divergent
views on this issue, see Mineka & Öhman, 2002, and Poulton & Menzies, 2002.)
Systematic Desensitization
What, then, of Watson and Rayner’s other claim, that if phobias are caused by conditioning
then it should also be possible to use conditioning principles to overcome them? One of
their suggestions was to associate the stimulus that elicited fear
with a pleasurable experience
such as eating or sexual stimulation. The pleasant feelings elicited by these events would be
incompatible with fear, they reasoned, so that if these reactions
could be conditioned, then fear
might be suppressed. This is, of
course, the counterconditioning
procedure originally described
by Pavlov.
Mary Cover Jones (1897-1987) was dubbed the “mother of
behavior therapy” by her colleague Joseph Wolpe due to her
research on the “deconditioning” of the fear reaction; her
study of a three-year-old boy named Peter who had a fear of
rabbits has been referenced more extensively than any other
aspect of her work.
lie6674X_03_c03_087-114.indd 107
The first human application of
this counterconditioning strategy was in an experiment by
Mary Cover Jones (1924). One of
her subjects, a boy named Peter,
was terrified of rabbits, and, following Watson and Rayner’s
3/14/12 4:17 PM
Section 3.5 Applications
CHAPTER 3
suggestion, she resolved to introduce the rabbit
while Peter was engaged in the pleasurable activity of eating. She introduced the rabbit gradually
over a period of days, on the eminently reasonable assumption that simply dropping the rabbit
on to Peter’s lap while he was eating would not
have produced the desired effect. She kept the
rabbit at a distance at first and then moved it progressively closer to the boy’s chair. The result was
nothing short of spectacular, as Peter not only lost
all fear of the rabbit but actively began to seek out
opportunities to play with it.
Despite the impressive success of this treatment,
there was little further research until the mid
1950s, when Joseph Wolpe reported a therapy
he had developed called systematic desensitisation (1958). Wolpe’s technique was similar to
that of Jones, except that his counterconditioning procedure used relaxation rather than eating
as the response. In addition, instead of actually
presenting the fear stimuli, he asked his patients This Moulin Rouge cabaret dancer looks
extremely comfortable with a snake
to imagine them. A therapist using Wolpe’s techdraped over her shoulders; however, a
nique would ask patients to describe situations person with a snake phobia, known as
that frightened them and then would arrange ophidiophobia, would experience intense
these stimuli in a hierarchy based on their aver- fear and possibly even panic attacks at the
siveness. A patient who had a fear of snakes, for touch of a snake and would benefit from
example, might find the idea of looking at a toy a counterconditioning procedure such as
snake to be only somewhat threatening, while systematic desensitization.
the idea of coming across a snake in the woods
might be even more frightening, and the idea of picking up a live snake perhaps the most
frightening situation of all. These images involving snakes would then be arranged in
a hierarchy of ascending order according to their ability to produce fear or anxiety. The
therapist would also train the patient in special techniques to encourage deep relaxation
(see Wolpe & Lazarus, 1966). Typically, a patient would start with the lowest stimulus
in the hierarchy (the one that produced the least amount of fear) and try to relax while
visualizing that scene. Only when the patient reported complete relaxation while imagining that scene would the therapist ask the patient to visualize the next scene, and so on.
Wolpe reported remarkable success in eliminating phobias with this technique, and subsequent studies have largely confirmed his claims. In a study by Paul (1969), for example,
students who had severe anxieties about public speaking were treated with either systematic desensitization or insight-oriented psychotherapy (which focuses on identifying the
cause of the phobia). When examined two years later, 85% of those given desensitization
showed significant improvement relative to pre-treatment levels, compared with 50% in
the psychotherapy group and only 22% in an untreated control group. The effectiveness
of systematic desensitization varies depending on the phobia being treated, but it is one
of the most effective treatments currently available for phobias involving specific objects
such as snakes or blood, or activities such as flying (Borden, 1992; Thyer & Birsinger, 1994).
lie6674X_03_c03_087-114.indd 108
3/14/12 4:17 PM
CHAPTER 3
Section 3.5 Applications
Exposure Therapy
Exposure therapy has sometimes been used
to treat individuals who are afraid to fly;
the procedure might begin by showing the
client pictures of airplanes and gradually
increasing his exposure, until eventually he
is able to fly with little or no anxiety.
One limitation to the effectiveness of systematic
desensitization is that the conditioned stimulus
is imagined rather than experienced directly. In
some cases, patients have overcome their fear
of an imagined stimulus such as a snake only to
find themselves still fearful when they encountered a snake in real life. To overcome this problem, many therapists now use exposure therapy,
in which patients are exposed to the actual stimuli that frighten them. As in systematic desensitization, exposure is gradual, starting with
situations that elicit minimal fear and advancing
only gradually to more frightening situations.
Patients are still encouraged to relax, but this
element of the treatment typically receives less
emphasis because of the difficulties of remaining
fully relaxed while engaged in physical activities such as moving toward a snake. Exposure is
thus closer to straightforward extinction, in contrast to systematic desensitization’s emphasis on
counterconditioning, but it too has proven very
effective (for example, Öst, Stridh, & Wolf, 1998;
Barlow, Raffa, & Cohen, 2002).
Aversion Therapy
A second major application of conditioning principles has been aversion therapy, in which the goal
is not to eliminate fear but rather to harness it to
produce avoidance of a harmful situation. This
principle is by no means new, with some of the
most imaginative—and gruesome—applications
stemming from ancient times. Pliny the Elder, for
example, recommended a treatment for alcoholism
that consisted of covertly putting the putrid body
of a dead spider in the bottom of the alcoholic’s
tankard. When the drinker would innocently tip
the contents into his mouth, the resulting revulsion
and nausea supposedly would deter him from ever
drinking again. A somewhat more modern example
(technically, at any rate) involved the treatment of
a 14-year-old boy who wanted to give up smoking
(Raymond, 1964). The boy was given injections of
apomorphine, a drug that produces intense nausea,
and each injection was timed so that it would take
effect while the boy was in the middle of smoking.
Here’s an excerpt from Raymond’s study:
lie6674X_03_c03_087-114.indd 109
Pliny the Elder (23 AD–79 AD) was a Roman
naturalist and natural philosopher who
wrote Naturalis Historia, an encyclopedic
work that collected much of the knowledge
of his time.
3/14/12 4:17 PM
3.5 Applications
CHAPTER 3
On the first occasion he was given an injection of apomorphine 1/20g, and
after seven minutes he was told to start smoking. At eleven minutes he
became nauseated and vomited copiously. Four days later he came for the
second treatment, and said that he still had the craving for cigarettes, but had
not in fact smoked since the previous session because he felt nauseated when
he tried to light one. . . .Two months later he left school and started working.
He said he had “got a bit down” at work and wanted to “keep in with the
others,” so he had accepted a proffered cigarette. He immediately felt faint
and hot, and was unable to smoke. It is now a year since his treatment, and
his parents confirm that he no longer smokes. (Raymond, 1964, p. 290)
Although Raymond’s results were highly impressive, early attempts to apply his procedures to problems such as smoking and alcoholism were less successful. In retrospect, the
main problem in these early studies was probably the unconditioned stimulus used. Raymond used apomorphine, which was seen to be very effective; however, because apomorphine is a dangerous drug that requires medical supervision, many of the early follow-up
studies used electric shock instead. As we saw in our discussion of preparedness, stimuli
such as the taste of alcohol or the odor of cigarette smoke are difficult to associate with
shock, and this could account for the higher failure rate in the studies that followed Raymond’s initial research (Lamon, Wilson, & Leaf, 1977). Once research on taste-aversion
learning in rats made this problem clear, researchers switched to USs that would be easier
to associate. For alcoholism, nausea-inducing drugs such as Antabuse are now used.
A further problem in the early studies was that even where treatment was effective initially,
patients often relapsed when treatment was discontinued. The cause was probably discrimination learning, as patients would have rapidly learned that whereas drinking alcohol in
the clinical setting was followed by illness, drinking in the neighborhood bar or with friends
had no such consequences. Rather than learning not to drink, they simply learned not to
drink in the presence of the experimenter! More recent studies have therefore incorporated
other forms of training to help patients cope with temptation once treatment has ceased.
One approach has been to provide counseling during treatment to teach strategies for
coping with the urge to smoke or drink when it arises. Another approach has been to provide posttreatment “booster” sessions to help maintain the aversion established during
treatment. In one study using this approach, Boland, Mellor, and Revusky (1978) paired
alcohol with lithium during treatment and arranged additional conditioning trials after
patients had been discharged. When they assessed their patients six months after discharge, they found that 50% of the chronic alcoholics in the treatment group were still
abstinent, compared with only 12% of the controls.
The use of multicomponent treatments in which aversion therapy has been combined with
other approaches has contributed to an improvement in the long-term effectiveness of aversion therapy (Hall, Rugg, Tunstall, & Jones, 1984; O’Farrell et al., 1996). In a review, Elkins
(1991) reported that approximately 60% of alcoholics treated with aversion therapy were
still abstinent one year after treatment, an impressive result for a problem that is notoriously
difficult to treat. However, this does not mean that aversion therapy is always appropriate.
The need for hospitalization means that aversion therapy for alcoholism is expensive, and
its unpleasant nature leads to higher drop-out rates during treatment. Where milder forms
of treatment are possible, therefore, they are preferred. For patients suffering from chronic
alcoholism, however, aversion therapy appears to be an effective alternative.
lie6674X_03_c03_087-114.indd 110
3/14/12 4:17 PM
Review Questions
CHAPTER 3
Summary and Review
•
•
•
•
•
•
•
Until the mid 1960s, psychologists believed that conditioning was a fundamentally simple process in which an association is formed whenever two stimuli are
presented together. (As shown by sensory preconditioning, it is not necessary
that one of these stimuli be a US—any salient stimuli that occur together may
become associated.) The strength of the association was determined by the contiguity of the stimuli, their intensity, and the frequency of their pairing.
The first major challenge to this belief came from Rescorla’s research on contingency: Contiguous pairings of a tone with shock did not result in conditioning if
the shock also occurred in the absence of the tone.
An experiment by Garcia and Koelling also showed that contiguous pairings of a
CS with a US does not necessarily result in conditioning—the outcome depends
on which CS is paired with which US. Rats will readily associate illness with a
taste, but not with a noise or light. Both animals and humans seem prepared to
form some CS–US associations more easily than others, a predisposition that
probably reflects an evolutionary history in which some stimuli proved to be
more likely causes of illness than others.
Kamin provided a third demonstration that contiguous pairings do not always
produce conditioning. He paired a noise-light compound with shock and found
that no fear was conditioned to the light if the noise had previously been paired
with shock—conditioning fear to one element of a compound blocked conditioning to the other element. Kamin’s explanation was that when we encounter a US,
we search our memories to identify possible predictors, but because this search
is effortful, we search only when necessary. Specifically, we search only if we are
surprised by the US—if we expected it, then an adequate predictor must have
already been available.
Together, these results provided convincing evidence that conditioning is not
simply a matter of associating stimuli that occur together. Even in the simplest
conditioning situations, learning seems to depend on cognitive processes such as
memory and attention.
Fears are often acquired through conditioning, and psychologists have used
conditioning principles to eliminate these fears. Exposure therapy and systematic
desensitization have both proven highly effective.
Conditioning principles have also been used to treat alcoholism, by pairing the
taste of alcohol with illness. Aversion therapy has been effective when administered in hospitals, but other techniques are sometimes needed to ensure that this
aversion is maintained once patients return to their normal lives.
Review Questions
1. Why did simultaneous and backward conditioning seem to pose problems for
the principle of contiguity? How can these apparent anomalies be explained?
2. How did Rescorla disentangle the roles of contiguity and contingency in
conditioning?
3. How did Garcia and Koelling show that the conditioning of a stronger aversion
to a taste than to a light was not simply the result of greater salience of the taste
as a conditioned stimulus?
lie6674X_03_c03_087-114.indd 111
3/14/12 4:17 PM
Concept Check
CHAPTER 3
4. How might classical conditioning contribute to an animal’s survival? Why might
it be better not to associate a US with all the stimuli that precede it?
5. How did Kamin account for blocking?
6. Is contiguity necessary or sufficient for conditioning? What is the relevant
evidence?
7. How could the Pavlovian concepts of generalization and counterconditioning be
used to account for the success of systematic desensitization?
8. Can conditioning principles account for the development of phobias?
Concept Check
1. Food is presented for three seconds, followed by a two-second tone. This is an
example of
a.
b.
c.
d.
simultaneous conditioning.
delay conditioning.
trace conditioning.
backward conditioning.
2. The contingency between a CS and a US is determined by
a.
b.
c.
d.
the time between the beginning of the CS and the beginning of the US.
the probability of the US.
the number of pairings of the CS and the US.
the probability of the US in the presence of the CS and in the absence of the CS.
3. Research on taste-aversion learning suggests that contiguity is ____ for classical
conditioning.
a.
b.
c.
d.
necessary
sufficient
necessary and sufficient
neither necessary nor sufficient
4. To explain blocking, Kamin proposed that
a.
b.
c.
d.
all salient stimuli elicit memory searches.
only unexpected stimuli elicit memory searches.
all salient stimuli elicit attention.
only unexpected stimuli elicit attention.
5. Exposure therapy is potentially a better treatment for phobias than systematic
desensitization because
a.
b.
c.
d.
it requires fewer trials.
it involves real rather than imagined stimuli.
it allows more scope for relaxation.
all of the above
Answers: 1) c, 2) d, 3) d, 4) b, 5) b
lie6674X_03_c03_087-114.indd 112
3/14/12 4:17 PM
Key Terms
CHAPTER 3
Key Terms
asymptote In mathematics, a stable value
that a curve on a graph approaches but
never quite reaches. As used in learning,
it generally describes the level of performance at which improvement ceases, so
that further training would produce no
additional improvement.
aversion therapy A procedure for eliminating a behavior by conditioning fear to
stimuli associated with the performance of
that behavior.
backward conditioning A procedure in
which first a US is presented, then a CS.
blocking A phenomenon in which prior
conditioning to one element of a compound
prevents conditioning to other elements.
contiguity Literally, proximity or closeness. In learning, the principle of contiguity
says that the formation of an association
between two events depends on their
closeness in time. A stronger version is that
contiguity is both necessary for the formation of an association (the events must be
contiguous to be associated) and sufficient
(any events that are contiguous will be
associated).
contingency A measure of the extent to
which two events occur together, or covary, over time. A contingency coefficient
is a mathematical statistic determined by
two probabilities—the probability that a
US will occur in the presence of a CS, and
the probability that it will occur in the
absence of the CS. If a US is more likely
to occur in the presence of a CS than in its
absence, we say that there is a contingency
between them.
lie6674X_03_c03_087-114.indd 113
delay conditioning A procedure in which
a CS is presented and then continues until
a US is presented. (In some experiments
the CS terminates when the US starts; in
others, the two overlap.) Some definitions
further restrict this term, confining it to
situations in which there is also a long
interval between CS onset and US onset.
exposure therapy A treatment for phobias
in which phobics are exposed to phobic
stimuli and given an opportunity to learn
that these stimuli are no longer followed
by traumatic events (extinction). Exposure
starts with stimuli that elicit low levels
of fear and gradually progresses to more
frightening situations.
frequency The number of times an event
occurs. In classical conditioning, the
strength of conditioning depends on how
often a CS is paired with a US.
intensity In classical conditioning, this usually refers to the strength of a stimulus—
for example, how bright a light is. Conditioning is stronger when the US is intense.
preparedness The tendency to associate
some CS–US combinations more readily
than others. Other terms for this phenomenon include relevance, selective association,
and associative bias.
sensory preconditioning A procedure in
which two neutral stimuli are presented
together and subsequently one of them is
paired with an unconditioned stimulus.
The typical result is that responding is
conditioned not only to the conditioned
stimulus but also to the stimulus that was
paired with it during the first phase.
3/14/12 4:17 PM
Key Terms
CHAPTER 3
simultaneous conditioning A procedure
in which a CS and a US are presented at
the same time.
trace conditioning A procedure in which
a CS is presented but then terminated
before presentation of the US.
systematic desensitization A therapy for
phobias based on counterconditioning.
Patients visualize fear-evoking stimuli
while relaxing, to associate the stimuli
with relaxation instead of fear.
vicarious learning The acquisition of new
behaviors arising from observation of others’ experiences. In vicarious conditioning,
conditioning occurs to a CS as a result of
seeing someone else receive pairings of
that CS with a US.
lie6674X_03_c03_087-114.indd 114
3/14/12 4:17 PM
4
Theories of Conditioning
Learning Objectives
After reading this chapter, you should be able to do the following:
• Describe the Rescorla-Wagner model and how the authors translated Kamin’s cognitive
account of conditioning into a more associative account based on a mathematical equation.
• Understand how mathematical models work, and how they can be used to explain known
phenomena such as conditioning, extinction, and blocking.
• Predict new mathematical models, such as the overexpectation effect.
• Identify challenges to the model from phenomena such as latent inhibition and configural
learning, and how the model could be modified to account for them.
• Differentiate between Pavlov’s substitution theory and Tolman’s concept of expectation and
describe some of the experimental research that supports each.
• Define and explore the two-system hypothesis, which proposes that both views were correct,
as two different learning systems emerged in the course of evolution.
• Explain the role of awareness in conditioning and the related form of learning called “causal
learning.”
• Examine the seemingly more sophisticated form of learning called causal learning, and the
possibility that it might be based on the same associative processes that underlie conditioning.
lie6674X_04_c04_115-152.indd 115
4/9/12 8:18 AM
Section 4.1 The Rescorla-Wagner Model
CHAPTER 4
The phenomenon of classical conditioning is basically very simple: If a CS and a US are
repeatedly presented together, the CS will eventually begin to elicit the same response
as the US does by itself. Pavlov proposed an equally simple theory to account for this
evidence, namely that whenever two centers in the brain are active simultaneously, the
connection between them will be strengthened. In essence, all that matters is contiguity:
If a CS and a US occur together in time, they will be associated. This account is delightfully simple, and until the 1960s it was used to explain virtually all the known facts about
conditioning. Research on contingency, preparedness, and blocking, however, posed a
fundamental challenge to the notion that simple contiguity is sufficient. In the case of
contingency, for example, Rescorla showed that conditioning would not occur if a US was
equally likely to occur in the absence of a CS as in its presence. The fact that a CS and US
occur together, in other words, does not guarantee conditioning, and thus conditioning
must involve more than simply linking brain centers that are active simultaneously.
In this chapter we will consider what this “more” might be and examine current theories
about what really happens when a CS and a US occur together. We will see that although
Pavlov was not totally wrong, conditioning involves a much more intricate web of processes than a simple contiguity explanation suggests.
4.1 The Rescorla-Wagner Model
Recall from Chapter 2 the idea that in order for conditioning to occur, the CS must be an
accurate predictor of the occurrence of the US (contingency). In fact, Rescorla’s research
revealed that animals are remarkably sensitive to the probability of the US both in the
presence of the CS and in its absence. The obvious way to account for this sensitivity is to
assume that animals are somehow capable of computing probabilities. If rats sometimes
receive shocks in the presence of a tone and sometimes in its absence, for example, they
might count how many shocks occur while the tone is present and also assess how much
time has elapsed. Using this data, they could determine the average probability of the
shock in the presence of the tone and, in a similar fashion, compute the shock’s probability in the tone’s absence. Finally, they could compare the two probabilities to determine
whether the tone signals an increase in the likelihood of shock.
It is possible that animals do carry out the complex processes implicit in this account—
measuring time, counting events, and computing probabilities.—but many consider it
unlikely. In 1972, however, two psychologists published a theory that offered a much simpler account. Robert Rescorla and Allan Wagner, from Yale University, offered an account
for almost every major aspect of conditioning—the occurrence of conditioning itself,
extinction, blocking, the effects of contingency, and so on. And they achieved all this using
only a single, simple equation!
The Rescorla-Wagner model has proved to be one of the most remarkable and influential
models in psychology, and we therefore will begin our exploration of theories of conditioning by examining it in some detail. Before we begin, it might be worth noting that
some of the following sections are difficult and may require careful rereading. This might
seem to contradict the previous claim that the model is simple, but once you understand
lie6674X_04_c04_115-152.indd 116
3/14/12 4:21 PM
Section 4.1 The Rescorla-Wagner Model
CHAPTER 4
it, you’ll see that it really does involve only a few simple assumptions. Because the model
is stated in mathematical form, you may have to master unfamiliar symbols and concepts
before it all begins to make sense. Mastering this new terminology may not be easy, but
the potential reward is an insight into how a few simple assumptions can explain what
seems to be a bewildering array of unrelated facts.
The Importance of Surprise
One powerful impetus for the Rescorla-Wagner model came from Kamin’s work on blocking. As we saw in Chapter 3, Kamin found that when a noise-light compound was followed
by shock, no fear was conditioned to the light if fear had previously been conditioned to
the noise. From the perspective of contiguity, this
result was bewildering: Why, when the light was
paired with an electric shock, was fear not conditioned to it, regardless of the noise?
If you became ill several hours after eating
peanut butter, you might wonder if it had
been the cause, but if this continued to
happen every time you ate it, your belief
that it was the cause would strengthen,
until eventually you were certain.
Kamin’s explanation was that when we encounter an important event, we search our memories
to identify stimuli that might have caused or
predicted it—if we know that a shock is coming, we at least have the possibility of preparing
for it. However, this kind of memory search uses
scarce cognitive resources that might be needed
for other purposes—an animal may need to stay
vigilant, for instance, for the appearance of a possible predator. So Kamin assumed that memory
searches occur only if the US is a surprise. In the
example of the noise-light compound, the rats
had already learned that noise was a predictive
cue for shock and thus would not have been surprised by the shock when it was associated with
both the noise and the light. They would not have
needed to search their memories again for a predictor and thus would not have learned about the
relationship between the light and the shock. In
sum, Kamin’s theory thus proposed that whether
or not conditioning will occur depends crucially
on whether the occurrence of the US surprises us.
To see how Rescorla and Wagner built on this idea, let’s suppose a painful rash suddenly appears on your face. It would be useful to be able to predict this type of event,
so let’s imagine that when the rash first appears, you search your memory for possible
causes. You remember having eaten a peanut butter and jelly sandwich earlier in the day.
Could that have caused the rash? You would become even more suspicious if another
rash appeared after you ate some peanut butter cups. At that point you might have felt
that peanut butter was the cause, but because you loved peanut butter, you were reluctant
to accept this. So a few days later you ate something else that contained peanut butter,
and the rash returned.
lie6674X_04_c04_115-152.indd 117
3/14/12 4:21 PM
Section 4.1 The Rescorla-Wagner Model
CHAPTER 4
Figure 4.1 plots how your expectation that eating peanut butter was the cause of your
rash might have changed with experience. At first you would have no expectation that it
would cause a rash, but after each new experience your expectation would have become
stronger, until eventually you were certain.
Figure 4.1: Rescorla-Wagner model and conditioning
The expectation that peanut butter causes a rash would become stronger each time you ate peanut
butter and then developed a rash.
We have assumed that expectations increase rapidly at first but then more slowly, and the
likely reason for this is surprise. The first time you noticed the rash, it would have come as
a complete surprise, and when you remembered eating the peanut butter, you would have
formed a tentative belief that it was the cause. The next time you ate peanut butter, therefore,
you would have been half-expecting illness to follow. When it did, you would not have been
nearly so surprised, and as a result you would not have needed to alter your expectation as
much. If an expectation is completely wrong, it makes sense to modify it substantially, but
the more accurate the expectation is, the less we need to adjust it. As your expectation of a
rash increased over trials, therefore, you would have needed to modify it less and less each
time, until eventually your initially-tentative expectation hardened into certainty.
This intuition—that how much we adjust our expectations depends on how surprised we
are—was Rescorla and Wagner’s fundamental insight. Where Kamin had suggested that
surprise determines whether conditioning occurs, Rescorla and Wagner now proposed
that surprise also determines how much conditioning occurs: The greater the surprise,
the greater the conditioning. For example, the first time you received a jolt of static electricity when you touched a metal door knob in your house, there would be a substantial
increase in your fear of touching that knob. The hundredth time you approached the door,
however, you would already have a high level of anxiety, and yet another shock would be
unlikely to produce much of an increase.
lie6674X_04_c04_115-152.indd 118
3/14/12 4:21 PM
Section 4.1 The Rescorla-Wagner Model
CHAPTER 4
However, Rescorla and Wagner wanted to avoid
mentalistic explanations of the kind we have been
developing here. We have no way of knowing
what a rat is thinking, and, as we saw in Chapter 1, there can also be problems in inferring the
thoughts and emotions of humans. Rescorla and
Wagner therefore wanted to express their ideas in
more neutral terminology. When a CS is paired
with a US, they said, an association or connection will be formed between them; they didn’t
speculate about what thoughts or feelings might
accompany this association.
A Mathematical Model
Because they wanted to be able to predict the
amount of conditioning that took place, they
made a second change to Kamin’s theory: They
expressed their ideas in mathematical form. This
can make their model appear intimidating, so as
we discuss their model in more detail, hold onto
the fact that underlying the equations is really
a simple idea—how much conditioning occurs
each time we encounter a US depends simply on
how much we are surprised by it.
In the early history of psychology,
mentalists used introspection to try to
understand the mind. This led to many
theories about the mind’s structure,
although introspection didn’t provide
the kind of clear evidence needed to say
which ones were correct.
The Learning Curve
As we have seen, Rescorla and Wagner assumed that when a CS and US occur together
an association will be formed. They used the symbol V to represent the strength of this
association. They further assumed that if these CS–US pairings were repeated (recall our
discussion of frequency in Chapter 3), the strength of the association would increase in
roughly the manner shown in Figure 4.2a. As you can see by looking at the top graph, the
more pairings, or trials, that occur for a particular CS and US, the stronger their association becomes. However, it becomes clear after looking at Figure 4.2b that this increase
in associative strength is not constant over trials. On the first trial, V increases by a substantial amount. Over successive trials, the increase in V on each trial gets progressively
smaller, until eventually V approaches a stable value.
lie6674X_04_c04_115-152.indd 119
3/14/12 4:21 PM
CHAPTER 4
Section 4.1 The Rescorla-Wagner Model
Figure 4.2: Increases in associative strength according to Rescorla-Wagner model
1.0
0
1
2
3
4
5
6
7
8
9
1.0
0
1
2
3
4
5
6
7
8
9
Associative strength (V) increases over number of conditioning trails (n) according to the Rescorla-Wagner
model. Figure 4.2a shows a typical learning curve; Figure 4.2b shows the same curve, indicating the
change in associative strength on each trial (∆V) and the asymptotic value of associative strength (Vmax).
lie6674X_04_c04_115-152.indd 120
4/9/12 8:19 AM
CHAPTER 4
Section 4.1 The Rescorla-Wagner Model
Rescorla and Wagner used the symbol ∆V to represent the change in associative strength on each
trial (∆, or delta, is the mathematical symbol for
change). The change in associative strength produced by the first trial was ∆V1, the change on
trial 2 was ∆V2, and so on. As we saw in Chapter
3, a stable value that a curve approaches but never
quite reaches is called an asymptote, and we will
use the symbol Vmax to represent the asymptotic
value of V (Figure 4.2b).
Quantifying Surprise
We can summarize the model to this point by
saying that associative strength increases over
trials until it reaches a stable maximum value; in
mathematical terms, V increases by ∆V on each
trial until it approaches Vmax. For example, if we
wanted to condition fear to the sound of a tone and
we decided to do this by pairing tone with shock,
the associative strength between them would
As one can see by looking at the image
above, the two ties in a set of railroad tracks increase rapidly during the first few pairings, or
trials. As the pairings continued, the strength of
will run next to each other but never meet.
the conditioning would continue to increase, but
In geometry, an asymptote is a line that a
curve approaches but never quite reaches;
the amount of increase—how much fear increased
there is always a gap between the two.
on any given trial—would be smaller each time.
Eventually, each new increase would be so small
that for all practical purposes fear would have reached a maximum level and would not
increase any further. (For the purposes of strict accuracy it would be more accurate to say
that fear would approach this asymptotic level but would never quite reach it.) At this
point, further pairings would not produce any significant increase in fear.
If we want to predict how much conditioning will occur, then, we need a formula to predict ∆V. A number of formulas were possible; in choosing one, Rescorla and Wagner were
guided by their assumption that the amount of conditioning depends on the amount of
surprise. To quantify surprise, they focused on the relationship between V and Vmax. At
the beginning of conditioning, when associative strength is low, we are not expecting the
US and so will be surprised when it occurs. When associative strength is high, on the other
hand, we will be expecting the US and hence are less surprised when it occurs. So, when
associative strength is low (and thus when V is far below its maximum value), we are
very surprised; when associative strength is high (and thus when V is close to Vmax), we
are much less surprised. The difference between V and Vmax, therefore, provides us with
a useful index of surprise: The closer V is to Vmax, the less we are surprised when the US
is presented.
Figure 4.3 illustrates this point by focusing on two trials, one that occurs early on in the
conditioning process and one that occurs late.
lie6674X_04_c04_115-152.indd 121
3/14/12 4:21 PM
CHAPTER 4
Section 4.1 The Rescorla-Wagner Model
Figure 4.3: Quantifying surprise based on the relationship between V and Vmax
V
This figure shows the relationship between V and Vmax early and late in conditioning. Early in
conditioning (point 1), the difference between V and Vmax is great, and so surprise is strong. Later
(point 2), the difference is much less, and thus surprise is much lower.
Early in conditioning (point 1), there will be a large difference between V and Vmax, and
substantial conditioning will occur. As conditioning proceeds, however (point 2), the difference between V and Vmax will become smaller, and the occurrence of the US will occasion less surprise. So, how surprised we are depends on how far V is from Vmax. Another
way of saying this is that surprise depends on how different the value of V is from the
value of Vmax: The greater the difference in their values, the more we are surprised.
Putting all of these ideas together, the notion that the amount of conditioning depends
on the amount of surprise can potentially be translated into mathematical form by saying that the amount of conditioning on any trial n (∆Vn) will depend on the difference
between V and Vmax:
∆Vn ≈ Vmax − Vn
where
Vn = the strength of the association at the beginning of trial n
∆Vn = the change in the strength of the association produced by trial n
lie6674X_04_c04_115-152.indd 122
3/14/12 4:21 PM
Section 4.1 The Rescorla-Wagner Model
CHAPTER 4
Parameters
In our presentation of the model to this point, we have talked as if the learning curve
shown in Figures 4.2 and 4.3 is found in all conditioning curves, but this is not quite true.
The overall shape of the curve—increasing over trials, but at a declining rate—is indeed
uniform, or at least roughly so, but the asymptotic level of conditioning can vary, and
so too can the speed of conditioning. In discussing taste-aversion learning, for example,
we noted that such aversions develop very quickly, whereas salivary conditioning generally requires many trials for conditioning to reach
its peak. To allow the model to account for variations in the speed of conditioning, Rescorla and
Wagner added a constant, c, to their equation. The
complete statement of the equation was thus:
∆Vn = c(Vmax − Vn)
In mathematics, a constant in an equation is called
a parameter. Suppose, for example, that a person’s weight was always 20 times as great as their
height. If we used the symbol H to represent height
and W to represent weight, then we could express
this relationship with the following equation:
W = 20H
If a woman was 5 feet tall, her weight would
be 100 pounds; if she was 6 feet tall, her weight
would be 120 pounds, and so on. The values for
height and weight would thus vary, but the value
of 20 would always be the same. It would be a
constant, and in mathematics any constant in an
equation is called a parameter.
If we use a linear equation such as weight
= 2 × height to predict a boy’s weight,
“2” is the constant value, or parameter, in
the equation. Simply plugging in a value
for “height” and multiplying this value
by a constant of 2 will give us a value for
“weight.”
The Rescorla-Wagner equation actually has two
parameters: c and Vmax. Vmax determines the asymptotic level of conditioning, the level attained after
many pairings. If a tone is paired with a 100-volt
shock, for example, the asymptotic level of fear is much greater than if the shock is only
20 volts. Knowing this, Rescorla and Wagner specified that the value of Vmax depends on
the intensity of the US—the value of Vmax used in the equation is greater when a 100-volt
shock is used than when the shock is 20 volts. The other parameter in the Rescorla-Wagner
model, c, determines the speed of conditioning—the greater the value of c, the larger will be
the change in associative strength on each trial. Thus, the faster conditioning will reach its
asymptote more quickly as the value of c increases.
If you come across a statement of the model in journal articles, you might not recognize it,
as the symbols Rescorla and Wagner used to represent these parameters are not the same
as the ones used here; we’ve altered the symbols to make the model easier to follow. And
lie6674X_04_c04_115-152.indd 123
3/14/12 4:21 PM
Section 4.2 The Rescorla-Wagner Model: Evaluation
CHAPTER 4
one final note—should you want to play with the model to see what it might predict in
different situations, note that the value of Vmax depends on what US is used, especially
its intensity, and the value of c depends on both the CS and the US and must be between
0 and 1. Note that the value of c in our version of the equation must lie between 0 and 1.
4.2 The Rescorla-Wagner Model: Evaluation
We now have an equation with which we can predict the precise change in associative
strength on any trial. To test the model, it might seem that all we need to do is present
a series of CS–US trials, calculate the predicted value of V for each trial, and see if our
predictions are correct. However, to calculate the model’s predictions for learning on any
trial, we need to know what values of c and Vmax to use.
Suppose, for example, that we ran a salivary conditioning experiment in which a light (CS)
was paired with 20 grams of food (US); what values of c and Vmax should we use in order
to predict the outcome? One solution would be to run a pilot experiment using 20 grams
of food, see what values of c and Vmax produce the most accurate prediction, and then
use these values in future experiments. If we found that setting
Vmax at 7 produced accurate
predictions when the US is 20
grams of food, for example, then
we could then use this value in
any further applications involving this US. When a theory has
several parameters, however,
this process turns out to be considerably more complex than
it sounds, and in the entire history of learning theory there has
only been one sustained effort
to estimate parameters in this
A “pilot study” is a small-scale, preliminary study that is
way (Hull, 1943). When this
often conducted prior to a larger study in order to test design
effort failed, after more than a
parameters, assess feasibility, and address other practical
decade of effort, it convinced
matters related to the research.
many conditioning theorists that
mathematical models were more
trouble than they were worth. Given this history, Rescorla and Wagner decided not to try to
determine the appropriate values for c and Vmax; instead, they used totally arbitrary values!
The use of arbitrary values might seem pointless, because the model will generate different quantitative predictions—for example, how many drops of saliva to expect—depending on which values are used. And as we have no way of knowing which of these values
might be correct, we have no way of deciding which prediction to believe. The model’s
quantitative predictions are thus of no value, but it turns out that the model can still
make some interesting qualitative predictions. For example, suppose that a dog received
lie6674X_04_c04_115-152.indd 124
3/14/12 4:21 PM
Section 4.2 The Rescorla-Wagner Model: Evaluation
CHAPTER 4
pairings of a tone with food. We could not predict how many drops of saliva would be
observed, but regardless of what values we used for c and Vmax, the model would always
predict that salivation would increase as training continued. So, although we could not
predict the number of drops of saliva, we could still make qualitative predictions about
whether salivation would increase or decrease.
This might still seem a waste of time—we hardly need a sophisticated mathematical model
to tell us that conditioning will increase over trials—but Rescorla and Wagner were able
to show that even simple statements of this kind can lead to interesting and unexpected
predictions.
Explaining the Old
To see how this can happen, we will first consider how the model accounts for relatively
straightforward phenomena such as conditioning and extinction. Then, once the basic
operations of the model are a bit clearer, we will turn to some of its more striking predictions. To begin, though, let us take a look at how the model accounts for the basic shape of
the learning curve during conditioning.
Conditioning
Suppose that we repeatedly
paired a tone with food, as in
the hypothetical experiment
whose results are illustrated by
the learning curve presented in
Figure 4.2. To see what sort of
results the model might predict
in this situation, let us arbitrarily
assume that the values of c and
Vmax are as follows:
c = 0.30
Vmax = 1.0
Although the rise and fall of the tides may change the exact
location of a shoreline, its basic curve is always the same. In a
similar way, although the values of c and Vmax vary depending
on the type of conditioning experiment being conducted, the
predicted shape of the learning curve remains the same.
If so, how much learning should
we expect? At the beginning
of trial 1, associative strength
would be zero, because the CS has never been paired with the US before. The amount of
conditioning on that trial would therefore be:
∆V = c (Vmax − V1) = 0.30 (1.0 − 0) = 0.30 (1.0) = 0.30
At the beginning of trial 2, the strength of the association would thus be 0.30: Trial 1
started with a strength of zero, and its strength was then increased (∆V1) by 0.30, giving
a new strength of 0.30. The change in associative strength produced by the second trial
would then be:
lie6674X_04_c04_115-152.indd 125
3/14/12 4:21 PM
Section 4.2 The Rescorla-Wagner Model: Evaluation
CHAPTER 4
∆V2 = c (Vmax − V2) = 0.30 (1.0 − 0.30) = 0.30 (.70) = 0.21
Since the associative strength of V at the beginning of trial 2 (V2) was 0.30, and it was
increased by 0.21 on that trial, the value of V at the beginning of trial 3 would be:
V3 = V2 + ∆V2 = 0.30 + 0.21 = 0.51
The predicted values for V for the first four trials are shown in Table 4.1. As you can see,
they correspond to the values plotted in Figure 4.2.
Table 4.1: Using the Rescorla-Wagner Model to Predict Conditioning
Trial
Vn
𝚫Vn = c (Vmax − Vn)
1
0.00
∆V1 = 0.30 (1 − 0.00) = 0.30
2
0.30
∆V2 = 0.30 (1 − 0.30) = 0.21
3
0.51
∆V3 = 0.30 (1 − 0.51) = 0.15
4
0.66
∆V4 = 0.30 (1 − 0.66) = 0.10
Our success in predicting these hypothetical data is perhaps not too surprising (especially
when you consider that the calculations were done first and the graph simply plots these
calculations!), but it does indicate the capacity of the model to generate learning curves
of the shape found in most conditioning experiments. The predicted shape of the curve is
the same, moreover, regardless of what values of c and Vmax are used. These parameters
alter the height of the asymptote and the speed with which it is reached, but in all cases
the basic shape of the curve remains the same. (You might find it useful to verify this for
yourself by working through some calculations using other values. You can use any value
for Vmax, but the value of c must lie between 0 and 1.0.)
Extinction
What about other aspects of conditioning? For example, can the model explain decreases
in responding as well as increases? Yes, and it does so using exactly the same equation
used to predict conditioning. The key to understanding how one equation can predict
diametrically opposite results lies in Vmax. We have said that Vmax is the strength of the
association that would be produced if a CS and US were paired repeatedly. In extinction,
we know that the level of conditioning reached after extended training is zero (in other
words, there is no longer an association between the CS and the US). The value of Vmax on
any trial in which a US is not presented, therefore, must also be zero.
To see the implications of this, suppose that after the third conditioning trial in our previous example we began to present the CS by itself. On the first extinction trial, V would
have an initial value of 0.66 (see Table 4.1), but as a result of nonreinforcement on that trial,
its associative strength would be changed by
lie6674X_04_c04_115-152.indd 126
3/14/12 4:21 PM
CHAPTER 4
Section 4.2 The Rescorla-Wagner Model: Evaluation
∆V1 = c (Vmax − V1) = 0.30 (0 − 0.66) = 0.30 (−0.66) = −0.198
The strength of the association, in other words, would be decreased by approximately
0.20, and its new strength would be
V2 = 0.66 – 0.20 = 0.46
Each extinction trial would decrease associative strength further, until eventually V would
approach its asymptotic value of zero. Using only a single equation, therefore, the model
can predict extinction as well as conditioning.
Blocking
We can also use the model to explain blocking. Before doing so, however, we need to consider how conditioning is affected if two stimuli instead of just one are present on a trial.
We said earlier that conditioning on any trial depends on how surprising the US is, which
in turn depends on how much the subject expected the US to occur. Rescorla and Wagner
assumed that if two conditioned stimuli, a and b,
were presented together, the subject would take
both stimuli into account in estimating the likelihood of the US. Specifically, they proposed that
the association or expectation at the beginning of
a trial would be the sum of the strengths of each
of the stimuli present...
Purchase answer to see full
attachment