LEARNING AND MOTIVATION (1974) 5, 369-379

Conditioned Inhibition is not the Symmetrical Opposite

of Conditioned Excitation: A Test of the
Rescorla-Wagner Model1

A. G. BAKER
Dalhousie University

Rescorla and Wagner’s ( 1972) linear associative model of learning pre-

dicts that a neutral stimulus will become weakly excitatory if it is paired
with a conditioned inhibitor and the compound is not reinforced. If the
inhibitor is not allowed to extinguish during these pairings, the model pre-
dicts that the neutral stimulus will become strongly excitatory. Rescorla
( 1971) has presented some evidence to support the former prediction of a
weak effect. Experiment I tested the latter prediction of a strong effect
using rats in a conditioned emotional response paradigm, and found no evi-
dence that the pairings made the neutral stimulus excitatory. Experiment II
replicated Rescorla’s results, but showed that even on a relearning test the
experimental group learned no faster than a novel stimulus control group.

Rescorla and Wagner (1972; Wagner & Rescorla, 1972) have recently
proposed a linear associative model to account for a wide range of phe-
nomena in Pavlovian conditioning and instrumental learning. The crucial
assumption of this model is that on a given trial the change in associative
strength of a particuIar stimulus is proportional to the difference between
the total associative strengths of all the stimuli present, including that
particular stimulus, and the asymptotic associative strength supportable
by the reinforcement used on that trial. Formally the model is given by
the expression AV~ = aA*fil ( h1 - B), where AV, is the change in asso-
ciative strength of stimulus A; P is the algebraic sum of the associative
strengths of all stimuli present; X~ is the asymptotic level of associative
strength supportable by the reinforcer used; and a,, and p1 are rate pa-
rameters for the stimulus and reinforcer, respectively.
One of the model’s most interesting features is its implication that Pav-
Iovian excitatory and inhibitory processes are symmetrical. This sym-

’ This research was supported by grant APA 259 from the National Researc,h
Council of Canada to N. J. Mackintosh. The author would like to thank N. J.
Mackintosh and G. Hall for their many comments on this manuscript. Requests for
reprints should be sent to the author at the Laboratory of Experimental Psychology,
University of Sussex, Falmer, Brighton, England, BN192Y.
369
Copyright @ 1974 by Academic Press, Inc.
All rights of reproduction in any form reserved.
370 A. G. BAKER

metry arises from the fact that, according to Rescorla and Wagner, both
excitation and inhibition arise from the same fundamental mechanism,
namely, that the associative strengths of the individual elements of a
compound are incremented or decremented according to the difference
between the components’ individual associative strengths and X. In a
typical procedure for studying excitatory conditioning a neutral cue is
paired with the unconditioned stimulus. Over a number of trials the
neutral cue will come to elicit a conditioned response. Excitatory con-
ditioning is explained directly from Rescorla and Wagner’s model. When
the neutral cue is first paired with the reinforcer the difference between
its associative strength and h is large and relatively large increments of
excitation (positive associative strength) will be conditioned to it. Over
a number of trials these increments will become smaller, as the difference
between h and the cue’s total associative strength decreases until an
asymptote is reached in which the cue’s associative strength approaches h.
In a typical procedure for studying conditioned inhibition, one cue
(A) is reinforced until it becomes excitatory, and a second cue (B) is
then paired with A and the resulting compound is not reinforced. Over
a number of trials B should become a conditioned inhibitor. Conditioned
inhibition may be measured by showing that the pairing of B with an-
other excitatory stimulus (C) reduces the response elicited by C, or by
showing that subsequent conditioning to B is retarded when B is rein-
forced (Rescorla, 1969). Rescorla and Wagner’s (1972) model requires
one supplementary assumption to explain conditioned inhibition. Res-
corla and Wagner chose to make the assumption that nonreinforcement
will support no associative strength (i.e., x0 for nonreinforcement is 0).
Once this assumption is made the explanation follows easily. During
excitatory training A acquires positive association strength until it ap-
proaches h,. When the compound AB is nonreinforced, its associative
value is positive, but since X, for nonreinforcement is 0 both A and B
will have their associative strengths decremented. Because B starts with
no associative strength it acquires negative associative strength during
this process. This negative associative strength is defined by Rescorla
and Wagner (1972) as conditioned inhibition. When B is paired with
an excitatory cue C, the resulting compound BC will have a lower asso-
ciative value than C alone and therefore will elicit less responding. Simi-
larly, in subsequent conditioning to B alone, B will be slow to develop
positive associative strength.
The prediction of the model tested in the present experiments arises
from its symmetry of inhibition and excitation. If a neutral cue is paired
with an excitor and the compound is not reinforced, the neutral cue will
become an inhibitor. It follows that, if a neutral cue is paired with an
inhibitor and the compound is not reinforced, the neutral cue should
 INHIBITION AND EXCITATION 371

become excitatory. Since the inhibitor has negative associative strength,

and the asymptote for nonreinforcement is zero, nonreinforcement of the
compound will result in increments in the associative strengths of both
components, and hence the associative strength of the neutral cue should
become positive. The model predicts that this effect will not be strong if
the inhibitor is allowed to extinguish during these nonreinforced trials.
In fact, Rescorla (1971) has provided some evidence which suggests
that a neutral cue may become weakly excitatory under these conditions.
The aim of the present Expt I was to protect the inhibitor from extinc-
tion, thereby, according to the model, forcing the neutral cue to become
strongly excitatory.

EXPERIMENT I

The conditioned emotional response (CER) technique was used. .d

noise was first established as an excitatory cue (i.e., it was reinforced
until it controlled total suppression). Next, the noise was paired with a
light in compound. This compound was not reinforced and the light
eventually became a strong inhibitor. During this phase reinforced
noise trials were continued, presumably keeping the associative strength
of the noise near x,, and, according to the model, forcing the associative
strength of the light to near --A,. In the final stage, nonreinforced trials
to a tone-light compound were added. During this stage the noise and
light were prevented from extinguishing (as an excitor and an inhibitor
respectively) by continuing the presentation of reinforced noise trials
and nonreinforced noise-light trials.

Method
Subjects. The subjects were 16 male hooded Long Evans rats which
were adapted to the laboratory conditions for one week and then main-
tained at 80% of their free feeding weight for the remainder of the
experiment.
Apparatus. The four conditioning chambers were lever boxes (23 cm
high, 20 cm deep and 25 cm wide ) . One side wall of each was made of
transparent Perspex, the ceiling was made of white translucent Perspex
and the remaining three walls were made of sheet metal. A food cup and
a lever were located on one wall with a speaker immediately above and
between them. A 40-W bulb was located above the ceiling. Scrambled
shock (1 mA, 1 set) from a Grason-Stadler (model 700) shock gen-
erator could be delivered through the grid floor (3-mm grids, 1 cm
apart), the lever and the sheet metal walls. The three cues used were
noise (80 dB, SPL from a Grason-Stadler noise generator), tone (70 dB,
SPL, 1200 Hz, pulsed once a set) and light (the 40-W bulb was turned
on). The boxes were isolated in separate sound attenuating chambers,
372 A. G. BAKER

and sound was further masked by the fan motors in these chambers.
Automatic programming and recording equipment was located in a
separate room.
Procedure. All training and conditioning sessions were 1 hr long. All
stimuli were 1 min in duration and responses during each, and for the
minute preceding each, were recorded, The main measure of condition-
ing was Kamin’s (Annau & Kamin, 1961) suppression ratio, i.e., the ratio
of responses in each trial to the sum of the responses during that trial
and the minute immediately preceding. With this measure, a ratio of 0
denotes complete suppression during the CS while a ratio of 5 signifies
no suppression to the CS.
The experiment consisted of four phases plus one pretest and one test
period. Phase 1 (days l-7) consisted of shaping and VI training.
On day 1 the animals were placed in the conditioning chambers for
1 hr. During the first 30 min, 30 pellets were delivered on a random time
schedule. During this stage each bar press caused a pellet to be delivered.
Each rat was removed after his 50th response. Those rats which did not
respond 50 times were hand shaped the next day. On days 3-7 the sub-
jects were placed on a random interval food schedule in which on the
average of once a minute, a reinforcement was primed and was delivered
by the next response. During all remaining sessions this reinforcement
schedule remained in effect.
Phase 2, the phase of excitatory conditioning consisted of days 8 and 9
during which twice a session the noise was presented and was followed
immediately by a shock. Phase 3, which lasted for days 9-33 was the
inhibitory training phase. On each day of phase 3 there was one rein-
forced noise trial and three trials during which the noise and the light
were presented together (noise-light trials). The following two days
(days 34-35) were the pretest. On each of these days four tone trials
occurred.
For Phase 4 the subjects were divided into two groups (Group 12 and
Group 24) in order to monitor the course of any excitation that might
accrue to the tone during this phase of training. Group 12 received 12
days of training (days 36-17) during which, on each day, they received
a reinforced noise trial, a noise-light trial and three tone-light trials none
of which were reinforced. Group 24 received the same treatment for 24
days. After this phase each group was presented with, as the second test,
four tone trials.

Results
The results of the last two days of Phase 3 and all of Phase 4 are
plotted in Fig. 1. It can be seen that at the end of Phase 3 there was
 INHIBITION AND EXCITATION 373

g .“‘r&q>+&

0 Y-4: _a
\*_-
.b 3 )I’ G12 G2L
3 N --.I -
it NL i--l -
:: .l- TL --: -
L? 0
12 24
2 Session Blocks
FIG. 1. Mean suppression ratio plotted over two session blocks for the last two
sessions of Phase 3 and for all of Phase 4 of Expt 1.

considerable suppression to the noise and only slight suppression to the

noise-light compound. During Phase 4 the noise continued to control
strong suppression while the noise-light compound controlled little
suppression.
The results of the three individual test days are presented in Fig. 2. It
is apparent that the tone controlled considerable suppression on the first
day of the pretest, but by the second day controlled little suppression.
This decrease was significant in both groups (t( 7) = 5.50, p 2 0.01 in
Group 12; t( 7) = 6.58; p < 0.01 in Group 24). On the final test day the
tone appeared to control even less suppression than on the last pretest
day. This trend was not significant for Group 12 ( t( 7) = .855, p 2 0.10)
but was significant for Group 24 (t( 7) = 2.77, p 2 0.05). If suppression
ratios to the tone-light compound for the last day of Phase 4 are com-
pared with the suppression ratios to the tone it is apparent that the

,’
/

-
1 2 Test
Pretest
FIG. 2. Mean suppression ratio on the four nonreinforced tone trials for the tuw
pretest sessions and the test sessions of Expt 1.
374 A. G. BAKER

animals were less suppressed to the tone than they had been to the tone-
light compound. This difference was significant for both groups (t( 7) =
2.55, p 5 0.05 for Group 12; t( 7) = 4.84, p 5 0.01 for Group 24).

Discussion
It will be recalled that Rescorla and Wagner’s (1972) theory predicts
that over a number of trials a cue paired with an inhibitory cue should
acquire a positive associative value, which in the present case would be
observed as suppression of responding to the tone. The results of this
experiment provide no support for this prediction. It is apparent that
over the course of the experiment the tone was quite excitatory when
first presented, probably due to generalization between it and the noise,
but over the pretest days and Phase 4 this excitation decreased. It should
be pointed out that it is possible that the procedure employed in the
present experiment did cause the tone to develop weak excitatory prop-
erties, but that the present experiment did not detect these. However,
this undetectable level of excitation would be very weak and Rescorla
and Wagner’s model predicts strong excitation.
It is worthwhile at this point to consider possible alternative expla-
nations of the present data. The first falls within Rescorla and Wagner’s
theoretical framework. It may be pointed out that while the light was
quite inhibitory, it never reached an associative value of -h because
there was always some suppression to the noise-light compound. The
theory further predicts that the light should, over the course of Phase 4,
achieve a new equilibrium associative value that is somewhat less in-
hibitory. That this may be the case is shown by a nonsignificant increase
in suppression between Phase 3 and Phase 4. It follows, that the equi-
librium associative value of the tone should fall somewhat short of being
equal and opposite to that of the light. Finally, Rescorla and Wagner
claim no connection between responding and associative strength except
that they are monotonically related. It is, therefore, possible to argue
that the present procedures were insufficiently powerful to make the
tone excitatory enough to be measured. This interpretation seems un-
likely, and if made would require Rescorla and Wagner to adopt rather
an awkward response connection rule, which would rob the model of a
considerable amount of its simplicity and appeal.
Another explanation of the present experiment calls upon the principle
of compounding. It has been suggested (cf. Razran, 1965) that when a
compound is reinforced a large number of times the compound remains
excitatory but the elements alone no longer elicit responding. If this had
happened in the present experiment (i.e., if the noise-light and tone-
light compounds were treated as wholes and not as pairs of elements),
 INHIBITIOK AND EXCITATION :375

then the prediction derived from the model is that no excitation will
accrue to either compound or to the tone, There are several reasons
why it is unlikely that this interpretation is valid. First the slight sup-
pression to the noise-light compound did not decrease over Phase -1.
Secondly, although it is difficult to think of an empirical test of the
compounding hypothesis the following procedure was adopted. After
the final tone test was completed, the subjects were given one day of
tone-shock pairings. On the next day they were presented with two
unreinforced trials each of the tone and the tone-light compound (i.e.,
a summation test for inhibition to the light; Rescorla, 1969). On the test
day the animals were completely suppressed to the tone (mean sup-
pression ratio = .027) and unsuppressed to the tone-light compound
(mean suppression ratio = .462). These results are inconsistent with the
compounding hypothesis. Finally, while Booth and Hammond (1971)
and Rescorla ( 1972) have demonstrated compounding, Rescorla ( 1972 )
showed that it is unlikely to occur “naturally” although it does occur
when subjects are “forced” to compound the rues in situations such as
a discrimination between a compomld and its elements. Further, other
studies (i.e., Baker, 1969) have failed to find compounding at all, imply-
ing that the phenomenon is not overly robust. Hence, there appears to be
little support for a compounding interpretation of the present results. 111
short, it seems that there is no viable description for the present results.
except that they are inconsistent with Rescorla and Lf’agner’s model.
Further, the results seem entirely consistent with the attentional notiwl
that an animal will soon stop attending to a stimulus which signals
nothing about his environment.

EXPERIMENT II

Experiment 1 failed to confirm the prediction, derived from Rescorla

and Wagner’s (1972) model, that a neutral stimulus presented in con-
jlmction with an inhibitor will acquire excitatory properties. Rescorla
( 1971) has reported data consistent with this prediction, and it there-
fore becomes necessary to reconcile the two sets of results. Rescorla
(1971) used a savings test to measure the weak excitation that had sup-
posedly been conditioned to the neutral stimulus during its pairings with
the inhibitory stimulus. In this test the rate of acquisition of suppression
was found to be slightly higher in the experimental group than in various
control groups. A feature co~n~no~~ to each of the control groups, ho\v-
ever, w;ls that during the immediate pre-test phase they were repeatedly
exposed to the neutral stimulus either alone or in compound with another
neutral stimulus. It seems possible that these control procedures in them-
sel\~ might hinder the acquisition of suppression bv the control ,gm~qw.
376 A. G. BAKER

Rescorla’s results, therefore, might reflect not enhanced acquisition in

the experiment group but a retardation in the control groups. Experi-
ment II was carried out to test this notion. This experiment attempts
to replicate Rescorla’s main control group (I-U) and his experimental
group (I-P) and in addition adds an extra control group which does not
receive repeated exposure to the neutral stimulus.

Method
Subjects and Apparatus. The subjects were 24 hooded male Long
Evans rats maintained in the same manner as in Expt I. The apparatus
was identical to that used in Expt. I. The stimuli were the same as in
Expt I except that the noise was decreased in intensity by 5 dB to 75 dB.
(Pilot studies showed that inhibition was more rapidly acquired at this
noise level.) The shocks were 1.3 mA and of 1-set duration.
Procedure. The animals were trained to press the bar in the same man-
ner as in Expt I. During Phases 2 and 3 the subjects received 2 days of
excitatory training to the noise (2 noise-shock trials a day) followed by
12 days of inhibitory training to the light, (1 reinforced noise trial and 3
nonreinforced noise-light trials per day). After Phase 3 the subjects were
given 2 days of pretest to the tone, during which they received 4 tone
trials per day. For Phase 4 the subjects were divided into 3 equal groups
(Group TL, Group T, and Group 0). For the 12’ days of Phase 4 all of
the groups received 1 reinforced noise trial and 1 nonreinforced noise-
light trial per day. Each day, in addition to this, Group TL received
3 nonreinforced tone-light trials, Group T received 3 nonreinforced tone
trials, and Group 0 received no additional trials.
After Phase 4 all groups received a savings test for acquisition of
suppression to the tone. This consisted of two days during each of which
the subjects received 4 presentations of the tone. The second and fourth
trials each day were followed by shock while the first and third were
unreinforced.
Results. The results of the first 3 phases were similar to those of Expt I.
By the end of Phase 3 the mean suppression ratio to the noise-light com-
pound was .497, .481, and ,430 for group TL, Group T, and Group 0,
respectively. These means were not significantly different from one an-
other ( F( 2,21) = 1.097). The mean suppression ratios to the noise trial
were .054, .063, and .089. These also did not differ significantly (F(2,2I)
= 0.506). All groups were mildly suppressed to the tone on the first tone
pretest day with mean suppression ratios of .340, .346, and .300, respec-
tively. On the second day of pretest there was no sign of suppression for
the three groups with mean suppression ratios of ,506, .5I3, and .495
respectively. There was no significant difference between groups on
 INHIBITION AND EXCITATION 37

either the first or second day of the pretest (F( 2,21) = 0.162, and
F( 2,Zl) = 0.038, respectively). During the fourth phase of the experi-
ment the difference in suppression ratios to the noise-light compound
and to the noise were maintained for the three groups. The mean sup-
pression ratios for the noise-light compound on the last day of Phase 4
were ,469, .511, and ,507, respectively ( F( 2,21) = .047), and for the noise
were .012, .Oll and .013 (F( 2,21) = 0.041). The results of the test phase
are shown in Fig. 3. There were no significant differences between groups
except on trial 4 (F( 2,21) = 4.41). T wo orthogonal contrasts were tested
using the method of ScheffC ( Winer, 1971) .
The first contrast compared the mean of Group T with the means of
groups TL and 0. The second compared the mean of Group TL with
that of Group 0. The first contrast was significant (F( 2,21) = 4.35) and
the second was not ( F( 2,21) = 0.06) thereby implying that the differ-
ence found was due to the mean of Group T being larger than those of
groups TL and 0, which did not differ from one another.

Discussion
The results of the first two trials of the test phase confirm the results
of Expt I; the present procedure, contrary to the predictions of Rescorla
and Wagner’s model, did not cause the neutral cue to become excitatory.
However of more relevance to this discussion are the results for the
various groups on trial 4 of the test phase. First, it should be pointed out
that the results of group T and group TL are similar to those of Res-
corla’s groups I-U and I-P, and therefore represent a replication of his
major finding in a somewhat different situation. It is this difference in
acquisition of suppression after the first shock that Rescorla interpreted

1 2 3 1
Blocks of 2 Trials
FIG. 3 Mean suppression ratios to the tone plotted in two trial blocks over thr
two tone test days for groups TL, T and 0 of Expt 2.
378 A. G. BAKER

as being caused by weak excitation conditioned to the neutral cue. The

results of Expt II put this interpretation in doubt.
It is possible to suggest a number of reasons why conditioning pro-
ceded relatively slowly in Group T in the present experiment and in
Rescorla’s Group I-U and other control groups. It is well known that
nonreinforced exposure to a neutral stimulus will retard future condition-
ing to that stimulus (e.g., Lubow & Moore, 1959). It is quite likely that
such latent inhibition would occur in the final phase before test in both
the present Group T and in Rescorla’s Group I-U thereby accounting
for these groups slightly retarded acquisition of suppression during the
test phase.
While the previous explanation may seem appealing it is weakened
considerably by the fact that Rescorla (1971) reported the results of
two other control groups both of which received nonreinforced TL
trials in the final pretest stage. The results of these groups were very
similar to those of Rescorla’s group I-U. However a second alternative
explanation lies in the fact that in the present experiment and most
probably in Rescorla’s experiment also there was some generalization
between the excitor and the neutral cue at the beginning of the final
conditioning phase. Nonreinforced presentation of the neutral cue in
the control groups probably extinguished this generalized excitation.
However in Rescorla’s and the present experiment’s experimental groups
and also in Group 0 of the present experiment, this excitation may not
have extinguished completely: in the experimental groups, the neutral
cue was paired with an inhibitor thereby protecting it from extinction,
and in Group 0 the tone was not presented in this final stage. Hence
the difference in acquisition of suppression may merely have repre-
sented the residue of generalized excitation.
It should be pointed out that regardless of which, if either, of these
alternative explanations is correct does not affect the main finding of
the present experiments, namely that continued nonreinforced pairings
of a strong inhibitor and an initially neutral cue do not turn the neutral
cue into a strong excitor. Even on a relearning test evidence for any
effect let alone a strong one was equivocal. In conclusion then it seems
that Rescorla and Wagner’s treatment of inhibition as the symmetrical
opposite of excitation, while attractively parsimonious, requires some
modificaion to cope with the present results.

REFERENCES

ANNAU, Z., & KAMIN, L. J. The conditioned emotional response as a function of the
intensity of the US. Journal of Comparative and Physiological Psychology, 1961,
54, 428-432.
 INHIBITIOS AND EXCITATIO?i 379

BAKER, T. W. Component strength in a compound CS as a function ol number of

acquisition trials. Journal of Experimental Psychology, 1969, 79, 3-17-352.
BOOTH, J. II., & HA~IMOND, L. J. Configural conditionmg: greater fear in rats to thr
compound than component through overtraining of the conlpound. JozrrnaI o/
Experimental Psychology, 1971, 87, 255-262.
Luuow, R. E., 8r MOORE, A. V. Latent inhibition: the effect of nonreinforced pre-
exposure to the conditional stimulus. Journal of Comparatice and Physiologicd
Psychology, 1959, 52, 416-419.
HALHAN, C. H. S. Empirical codifications and spxific theoretical implications of
compound-stimulus conditioning: Perception. In W. F. Prokasy (Ed. ), Cla.rsica/
conditioning. New York: Appleton-Century-Crofts, 1965. Pp. 226-248.
RESCORLA, R. A. Pavlovian conditioned inhibition. Psyc/fo/ogical Bulletin, 1969, 72,
77-94.
RESCOIILA, R. A. Variation in the effectiveness of reinforcement and nonreinforcement
following prior inhibitory conditioning. Leanrittg and Motication, 1971, 2,
113-123.
RESCOHLA, R. A. Configural conditioning in discrete trial bar pressing. Journal of
Cotnparatitie and Physiological Psychology, 1972, 79, 307-317.
RESCOKLA, R. A., CL WAGNER, A. R. A theory of Pavlovian conditioning: variations in
the effectiveness of reinforcement and nonreinforcement. In .h. H. Black & \\F. F.
Prokasy (Eds.), Classical conditioning 11. New York: Appleton-Century-Crofts.
1972. Pp. 64-99.
WAGNEH, A. R., & RESCORLA, R. A. Inhibition in Pavlovian Conditioning: An applica-
tion of a theory. In R. A. Boakeg and M. S. Halliday (Eds.), inhibition ant/
Learning. London: Academic Press, 1972. Pp. 301-336.
WISEH, B. J. Statistical principles in experimental design. (2nd Ed.) New York:
hlcCraw-Hill, 1971.

Received August 13, 1973

Revised December 23, 1973