SUCCESSIVE HABIT REVERSAL LEARNING
BY THE SPECTACLED CAIMAN
JOHN TAYLOR WILLIAMS, JR.
A DISSERTATION PRESENTED TO THE GRADUATE COUNCIL OF
THE UNIVERSITY OF FLORIDA
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS 'OR THE
DEGREE OF DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
The author wishes to express his appreciation to the chairman
of his supervisory committee, Dr. B. N. Bunnell, and to the other
members of the committee, Dr. Pierce Brodkorb, Dr. H. S. Pennypacker,
Dr. C. M. Levy, and Dr. W. B. Webb. This study was conducted at the
Veterans Administration Hospital in Augusta, Georgia.
TABLE OF CONTENTS
ACKNOWLEDGMENTS . . . . . . . . . . ii
LIST OF TABLES . . . . . . . . iv
LIST OF FIGURES . . . . .... .. .. . . ... v
1 INTRODUCTION . . . . . . . . . . 1
Phylogenetic Review . . . . . . . . 2
Methodological and Interpretive Considerations . 8
2 EXPERIMENT 1--CUE DOMINANCE . . .... . . . 13
Method . . . . . . . . . . . 13
Results and Discussion . . . . . . ... 16
3 EXPERIMENT 2--REVERSAL LEARNING . . . . ... 18
Method . . . . . . . . . . . 18
Results . . . . . . . . .. . . 21
4 DISCUSSION . . . . . . . . . . . 27
5 SUMMARY ................... ... 32
REFERENCES .. .. . .. . . . . . . . . 34
APPENDIX . . . . . . . . ... . . ... . 37
BIOGRAPHICAL SKETCH . . . . . . . ...... 39
LIST OF TABLES
1. Analysis of Variance of Errors to Criterion for C06 .... .22
2. Analysis of Variance of Errors to Criterion for C-12 . .. 23
LIST OF FIGURES
1. Maze with interchangeable black-white panel and with
goal box at left . . . . . . . .... ..... 14
2. Top view of maze and major dimensions . . . . .. 15
3. Median number of errors per reversal for the first five
reversals . . .. . . . . . . ... .19
4. Mean number of errors per reversal averaged across pairs
of reversals . . . . . . . . ... ..... 20
5. Within-problem learning averaged across sets of four
consecutive reversals for group C012 . . . . .... 24
6. Within-problem learning averaged across sets of four
reversals for group C06 . . . . . . . .... 25
7. Per cent of first trials of each day correct as a
function of the number of consecutive correct trials
at the end of the previous day . . . . . .... 26
In a successive habit reversal problem a subject is given a series
of two-choice discrimination in which the valences of the positive and
negative stimuli are reversed periodically. The primary focus of inter-
est is usually the course of improvement, if any, across successive
reversals. This problem is viewed by many as a similar, but simpler,
type of set learning when compared with the object-quality discrimina-
tion set learning studied by Harlow (1949). As Warren (1965a) has
pointed out, there is an orderly progression in ability to show inter-
problem improvement in an object-quality set learning situation when one
compares mammals at various stages of phylogenetic development. This
progression is most marked when comparisons are made among various
In addition Co its apparent similarity to a traditional learning-
set situation, the successive habit reversal problem has gained the
interest of comparative psychologists since Bitterman announced that
fish either showed interproblem improvement that was qualitatively
different than rats (Wodinsky & Bitterman, 1957) or that they showed no
interreversal improvement whatsoever (Bitterman, Wodinsky, & Candland,
1958). These findings have sparked a number of studies designed to help
determine what changes in learning ability are associated with the
evolutionary development of more complex nervous systems.
A review of the findings phylogenetically and a discussion of some
of the methodological variations may help to evaluate the importance of
studies of habit reversals to the comparative psychologist and suggest
further experimentation which may be of value.
Mammals.--As warren (1965b) has pointed out, every mammalian species
which has been studied has shown progressive improvement across suc-
cessive reversals. Schusterman (1962) studied chimpanzees and found
"extremely efficient" learning, even though the Ss only received three
reversals. He trained his Ss to a criterion of 12 consecutive correct
responses, using a correction procedure. The most striking aspect of his
findings was the great number of errors made on the first reversal
(-57.8); after this, it would have been most surprising if there had
not been a great improvement.
House and Zeaman (1959) studied reversal learning of low-grade
retardates on a position discrimination with a correction procedure and
candy for a reward. The Sa were trained to a criterion with conditional
limits regarding trials per day and reversals during the course of a
day's testing. There was negative transfer during the first four re-
versals, but the Ss had approached single error reversing after six
Warren and Warren (1962) trained two horses and one raccoon on con-
founded (visual and spatial) problems for food reward and found rapid
improvement. These Ss were trained to a criterion of 11 correct in 12
trials and learned one reversal per day, using a noncorrection procedure.
The raccoon learned 20 reversals (four with no errors), one horse
learned nine, and the other horse was given only six reversals before
testing was terminated.
Kittens of various ages have been tested (Warren & Warren, 1967)
for food reward on a confounded problem. A criterion of ten consecutive
correct trials was used with a noncorrection procedure. There was no
systematic variation based upon age, and the results were consistent
with studies of older cats (Cronholm, Warren, & Hara, 1960) in that
there was an increase in errors on the first reversal and a rapid de-
crease on subsequent reversals to an asymptote of about six errors per
In general, the studies with rats agree with the studies conducted
with other mammals. There is usually, but not always, an increase in
errors on the first few reversals and then a decrease so that there is
a high level of efficiency after a dozen or so reversals. North (1950a,
1950b) performed the first systematic studies of successive habit re-
versal learning in the rat, partly as a result of Buytendijk's (1930)
report that the rat could learn to reverse after a single error. However,
North was unable to get single error reversals in a T maze, even after
60 reversals. Procedurally, North reversed his Ss after 6, 18, or 30
trials rather than training his Ss to a criterion. He also used de-
layed correction and noncorrection rather than a correction procedure.
In a widely cited study, Dufort, Guttman, and Kimble (1954) found that
rats could quickly develop single error reversing. Their results were
very similar to those of House and Zeaman (1959) with retardates. Their
Ss were trained on spatial discrimination with food reinforcement. A
noncorrection procedure was used, and only four trials per day were given.
As in Buytendijk's study, Sa were trained to a criterion. Pubols (1957,
1962) has compared performance when rats are trained to a criterion with
performance of rats which are reversed after set numbers of trials. He
found that training to criterion is superior to reversing after a set
number of trials and that reversing after a large number of trials is
superior to reversing after a small number of trials. Gatling (1952)
studied rats on a series of visual discrimination reversals for food
reward. The Ss were trained to a performance criterion. The results
showed the usual mammalian pattern of a drastic increase in errors on
the first couple of reversals and a subsequent improvement to a level of
greater efficiency than that demonstrated on the original problem. How-
ever, unlike the situation in most studies using rats, the Ss were still
making about 40 errors per reversal after a dozen reversals. Gatling
did show that the pattern of elimination of errors was basically the
same throughout training. That is, errors were reduced at all stages of
intrareversal learning throughout all stages of reversal training. Estes
and Lauer (1957) reported the only recent rat study in which there was
not a clear-cut improvement across successive reversals. Their study
differed from most experiments in that their Sa only received one, two,
or four trials per day. Also, they only gave their Ss four reversals.
Lawrence and Mason (1955) studied reversal learning in rats with varying
numbers of relevant cues and noted that with three relevant cues a S is
apt to perform on the basis of position and is less efficient than a S
which is given only two relevant cues.
Birds.--In the first reversal study using birds, Reid (1958) failed
to find the interreversal improvement which is so characteristic of
mammals. He used pigeons on a color discrimination with food reward.
Bullock and Bitterman (1962) studied pigeons on both visual and spatial
discrimination and found progressive improvement. When the discrimina-
tion was spatial rather than visual, this improvement was faster.
Instead of a standard correction or noncorrection procedure they used a
"guidance" procedure in which an incorrect response is followed by the
presentation of the positive stimulus alone. A more recent study
(Bitterman, 1965) also showed that pigeons were capable of improving
across reversals when given visual discrimination. Warren, Brookshire,
Ball, and Reynolds (1960) studied chicks on spatial and confounded
problems for food reward and found that older chicks were more efficient
at reversing than were younger ones. Successive improvement by chickens
on multi-dimensional visual tasks has been reported (Bacon, Warren, &
Schein, 1962) in another study using food reward.
Reptiles and Amphibians.--An early study (Kirk & Bitterman, 1963)
tested turtles in a T maze with confounded cues for food reward. The Ss
given five trials per day and reversed after errorless days showed no
improvement, but Ss given ten trials per day with daily reversals showed
some improvement over the course of 70 days (but no statistical tests
were reported). Bitterman (1965) reported some work done with Holmes in
which turtles were tested on spatial end visual discrimination. His
conclusions (again without reporting statistical significance) were that
turtles improved on the spatial task but not on the visual task. The
data in his graph of the errors by the visual group could have been
generated by Se which were operating strictly on a position basis. Since
Se were reversed every four days, this could have easily happened with
position preference responses being reinforced about half of the time.
Recently this work has been published in greater detail with additional
experimentation (Holmes & Bitterman, 1966). It was learned that turtles
improved substantially on a visual task when they were trained to a
criterion before being reversed. Iguana lizards showed improvement in
a T maze with confounded cues (Alkov & Crawford, 1966). Davidson, (1966a,
1966b) studied alligators, but only three of his Sa learned the second
reversal. He used escape from drying heat for reinforcement, and could
only give his Sa a single trial per day.
One study (Seidman, 1949) cited frequently in early reversal learn-
ing studies, reported that terrapins were more efficient than newts at
reversing a direction habit in a T maze. A block, visible from the
choice point, was used to convert the incorrect arm into a cul-de-sac,
and Seidman's criterion for learning was only three consecutive correct
trials. Perhaps he showed that newts cannot see as well as terrapins,
but we still have no information regarding reversal learning by
Fish.--When fish (African mouthbreeders) were first tested in a
habit reversal paradigm (Wodinsky & Bitterman, 1957), they showed
improvement over the first six reversals and then a leveling off in
performance. The Ss were tested on a visual discrimination to a crite-
rion of 17 correct of 20 daily trials. A correction procedure, in which
a trial was terminated by a correct choice or by five incorrect choices,
produced greater improvement than did a noncorrection procedure. It was
noted that the improvement occurred only in the later stages of train-
ing on each reversal, and thus it was qualitatively different from the
improvement reported by Gatling (1952). Further studies inspired by
this qualitative difference were reported in a paper the next year
(Bitterman, Wodinsky, & Candland, 1958). However, no improvement was
found on either visual or spatial tasks using either correction or
guidance procedures and training to criteria. Further experimentation
with fish has failed to demonstrate improvement on confounded problems
(Warren, 1960; Bitterman, 1965), and Warren even found a progressive
increase in errors while testing paradise fish. Behrend, Domesick, and
Bitterman (1965) tried even more methodological variations and reported
typical negative results. However, they mentioned that two or three of
the individual Ss showed a pattern similar to that shown by typical
mammals in which there was an increase in errors on the first reversal,
followed by a decrease across reversals.
Invertebrates.--The results have been quite difficult to interpret
at the invertebrate level. Thompson (1957) tested sow bugs on a spatial
problem for eight reversals. There was no significant improvement, but
the data clearly indicated a trend toward improvement. Significant
findings were probably very unlikely because of a small number of So and
very strong position preferences. The Bermuda land crab was tested
(Datta, Milstein, & Bitterman, 1960) on a confounded problem, with escape
from fresh water used for reinforcement. Ss were reversed either daily
or every four days, and no improvement was found. On the other hand,
Mackintosh (1964) has reported clear-cut improvement across a dozen
reversals in octopuses. They were performing at the level of the original
problem when it became necessary to terminate the study because the Ss
became unhealthy. Datta (1962) has found a decline in errors by earth-
worms given five trials per day and reversed every four days. However,
control So quickly reached the efficiency of reversed SE when they were
given reversals. Therefore, Datta concluded that the improvement was not
due to learning to reverse but was due to a greater familiarity with and
adaptation to the maze. Longo (1964) reports similar results and con-
clusions using cockroaches. He used control Ss, Ss reversed daily, and
Ss reversed every four days. The errors for the Ss reversed daily de-
clined significantly, but those with reversals every four days did not.
After 44 days, the four-day reversal and control Ss were switched to
daily reversals and reached the level of the daily reversal group in
six days. Therefore, Longo concluded that the improvement by the daily
group was a result of better adjustment to the maze situation. Crawford
and Henton (1965) reported that when tarantulas were given three re-
versals, the third was the easiest and the second was the hardest.
In summary, mammals usually, but not always, made more errors on the
first few reversals than they did on their original problems and then
improved across successive reversals to asymptotic performances which
were superior to the performances on the original problems. Frequently
mammals approached single error reversing, but sometimes they did not
do so even after extended training. Birds usually showed a pattern like
that shown by mammals, but they have not attained single error reversal
learning. Interreversal improvement has been demonstrated in reptiles,
but it has been found less frequently than in higher forms. Fish have
failed to improve across reversals in all but one of the studies using
them as Ss, and that was the first study in which relatively crude
techniques were used. The data collected using invertebrates as Se
were rather contradictory, and interpretations of interreversal improve-
ment by invertebrates are open to controversy.
Methodological and Interpretive Considerations
If animals form learning sets for habit reversals, these sets should
increase the efficiency of the animal in solving a problem. In the
phylogenetic review, it was frequently mentioned that many animals sig-
nificantly improved across a series of reversals. However, little was
said about comparisons between the performance on the original discrimina-
tion and the performance at the end of reversal training. Although the
graphs presented by several authors indicated that there may have been
some gains over the original levels of efficiency, there have been no
reports of submammalians reaching a level of proficiency that signifi-
cantly surpassed their ability to solve the first problem. It is very
hard to be convinced that an animal has formed a set to reverse if all
it does is eliminate negative transfer effects. However, when a mammal
can consistently reverse after single errors when it had made three
or four errors on the first problem, it seems reasonably safe to assume
that it has formed such a set.
Bitterman (1965) has stressed the differences which are often
evident when results from visual discrimination are compared with those
from spatial discrimination, and Mackintosh (1965) has discussed the
role which attention may play in the overtraining reversal effect. If
a S is performing on the basis of position when visual cues are relevant,
it will receive about fifty per cent reinforcement for its preferred
position response. Therefore, if the S receives a series of reversals
after some set number of trials, it could easily keep performing on the
basis of its position preference. The resulting fifty per cent rein-
forcement should make its position response fairly resistant to extinc-
tion, and it would perform at a level of fifty per cent errors to the
relevant cue with no improvement across reversals. Therefore, failure
to improve may not always result from an inability to improve, but
rather from the S's lack of attention to the relevant cues. Training
to a performance criterion assures that the S attends to the relevant
cue and effectively eliminates this problem; this training to criterion
has been shown to be more effective in producing interreversal improve-
ment in rats.
One factor which may greatly influence the situation Just described
is the consequence of an error. If the S is poorly motivated, the con-
sequences resulting from an error are minimal. Reid's pigeons (1958)
apparently were poorly motivated as were the crabs studied by Datta,
Milstein, and Bitterman (1960). Bullock and Bitterman (1962) made the
suggestion that Reid's birds lacked proper motivation, but the lack of
improvement in their birds with a guidance procedure with a time out of
zero seconds seems to be quite parallel. They stated that this lack of
improvement was a function of delay of reinforcement, but perhaps it is
more useful to think of it in terms of elimination of the consequences of
an error. This could offer a possible explanation for the reason that
sometimes invertebrates seem to improve while fish do not. The fish
have all been tested with food reward, and may not have been "punished"
sufficiently for their errors. Typically, invertebrates have been
studied in situations where they were escaping aversive stimuli.
Another obstacle to one who is trying to determine whether animals
can form sets to reverse is the fact that Ss are typically reversed at
the beginning of a new day. This creates a special problem with lower
forms which may have trouble remembering from one day to the next. It
is quite conceivable that a S may gradually learn something about how
to perform in the situation, e.g., to repeat a correct response, without
being able to remember which way it went the day before. Thus, if a S
is being reversed on a daily basis, it may reach the point where the
first trial each day is performed on a chance basis, and the rest of
the trials that day are performed correctly. This could result in
interreversal improvement without anything being learned about reversing.
This could possibly explain how the raccoon studied by Warren and Warren
(1962) learned four of its 20 reversals without any errors. Another ex-
planation is that being reversed each day could serve as another cue for
the S to reverse.
The effects of using correction, noncorrection, or guidance pro-
cedures are hard to separate from other factors; for example, a correction
procedure is necessary in an escape situation. Perhaps further experi-
mentation may show that these procedures have come effects above and
beyond the effects due to interactions between them and other aspects
of the procedure.
Host of the comparative studies of habit reversals have been per-
formed by M. E. Bitterman, J. M. Warren, and their students. In making
comparisons, Bitterman looks for different functional relationships at
different levels, and Warren looks for qualitative differences or for
quantitative differences in which there is little or no overlap be-
The following study was performed to try to determine whether a
reptile, in this case Caiman sclerops, could improve across reversals in
a way comparable to that of mammals. Caimans were chosen to supplement
the rather sparse information regarding reptilian habit reversal learning
because they have not been studied previously; they have a slight amount
of neocortical tissue, and they are hardy animals which can be maintained
in good health for extended periods. They are of interest phylogenetical-
ly in that they are more closely related to the ancestral stock of birds
than to that of mammals. The results will be compared with mammalian
data to determine whether the functional relationships are the same at
the two levels and whether the caimans can attain performance which falls
in the range of efficiency usually reported in studies using mammals. In
order to make these comparisons in some way meaningful, a situation was
designed which, hopefully, produced optimal opportunity for inter-
reversal improvement. Therefore, Ss were first tested to see whether
they normally attended to spatial or visual cues; the chosen cue was
then made relevant for a study of reversal learning. Shock escape was
chosen so that a sufficiently high level of motivation could be achieved
with definitely noxious consequences following an error. Ss were
trained to one of two criteria on each problem and reversals took place
during the day's testing so that the first trial of a day would not
serve as a cue to reverse.
EXPERIMENT 1--CUE DOMINANCE
By training a S on a confounded task with both visual and spatial
cues relevant and then giving a test trial on which the cues give con-
tradictory information it should be easy to determine which cue is
normally used by the S. A group of caimans was tested in this way to
determine whether they would consistently choose one cue in a situation
in which both types of cues were available and relevant.
Subjects and Housing.--Eight spectacled caimans (Caiman sclerops)
purchased commercially about 18 in. in length served as Sa. They were
housed in a 2- by 4-ft. tank in which the water temperature was main-
tained at about 90* F. Each morning the water was changed, and a sun
lamp was turned on for about 9 hr. The diet consisted of ground chicken
parts (including bones) with cod liver oil added periodically. Identi-
fication of individual So was made possible by the use of varicolored
Apparatus.--The apparatus was a modified T maze with galvanized
metal sides and a floor made of 1 in. wide stainless steel plates
running diagonally and spaced 1/8 in. apart. It is illustrated in Fig. 1,
and the dimensions are given in Fig. 2. The goal box was a specially
built pan which was 18 in. long and held up to 7 in. of water. A metal
guillotine door converted the incorrect arm into a cul-de-sac. The
wall of the arms opposite the stem was made either black on the left and
Fig. 1. Maze with interchangeable black-white panel and
with goal box at left. The meter was to monitor
the current across two grid plates.
--- 35 1/2"
Fig. 2. Top view of maze and major dimensions.
All walls were 14 in. high, and the
maze was elevated 7 in. above the
white on the right, or vice versa, by the use of interchangeable sheet
metal panels. A variable voltage transformer delivered a shock through
a scrambler to the floor and sides of the maze. A short-circuit current
of 6 to 8 ma. was used, depending upon the behavioral reactions of the
individual Ss. Observations were made by using a mirror mounted at an
angle over the maze.
Procedure.--On each trial the S was removed from the home tank by
hand and dropped onto the start area of the grid facing the choice point.
When a S entered the goal pan, it was lifted to the home tank, and the
S was allowed to swim out. This prevented handling (which was obviously
noxious) in the goal area. The goal pan was then refilled with water
from the home tank for the next trial. An intertrial interval of 15 min.
was used to prevent cumulative effects of shock. An error was defined
as an entrance into the wrong arm as far as the S's hind legs, and the
criterion for learning used was eight consecutive correct trials. Four
Ss were trained with right and white positive, and four were trained to
go to black on the left. After a S reached criterion, it received a
single test trial with the positions of the black and white reversed.
A goal pan was at each arm of the T during the test trial.
Results and Discussion
Of the eight Ss, seven reached criterion in 8, 9, 10, 10, 11, 12, and
16 trials, and the eighth was discarded when it had shown no indications
of learning after 30 training trials. On their test trials, all seven Ss
made a turn to the side to which they had been trained. Even if the eighth
S had made a choice on the basis of brightness, the results would have been
significant (g<.05) in favor of the use of spatial cues.
Thus, it is readily apparent that caimans tend to use spatial cues
rather than visual cues in a maze situation such as this. This lends
some support to the usually untested assumption that most animals attend
more to spatial than to visual cues.
Incidentally, three of these Ss were then trained on a black-white
discrimination with their positive color unchanged and with the position
cues made irreleval.~. iwo of the Ss reached criterion in 51 and 57
trials, and the third had shown some indications of learning after 60
trials. Two of these Ss took over 40 trials to abandon an almost absolute
preference for the position to which they had been trained on the con-
founded problem. Therefore, caimans can learn a visual discrimination,
but it is a much more difficult problem for them than is a spatial
EXPERIMENT 2--REVERSAL LEARNING
After it had been determined that position was the primary cue used
by caimans in this situation, a study of habit reversal learning was
made using position as the relevant cue.
Subjects and Housing.--Sixteen spectacled caimans (Caiman sclerops)
ranging from 16 to 19 in. in length served as Ss. They were housed and
fed as in Experiment 1, except that prior to use as Ss they were housed
in a 2- by 5- ft. colony tank. On the days when data were being
collected from an individual S, it was housed in a 2- by 4- ft. tank
with about five other Ss.
Apparatus.--The apparatus was that used in Experiment 1, except
that the black and white sheet metal panels were not used. Thus, the
entire inside of the maze was unfinished, galvanized metal.
Procedure.--The procedure on any given trial was identical to that
in Experiment 1. However, each S received 20 trials per day throughout
training on the original problem and on 20 reversals. As before, half
of the Ss were trained to go to the right on the first problem and half
were trained to go to the left.
The Ss were divided into two groups of eight Sa, matched on the
basis of length. One group was reversed after having reached a criterion
of six consecutive correct trials, and the criterion for learning for
the other group was twelve consecutive correct. The next problem was
0 1 2 3 4 5
Fig. 3. Median number of errors per reversal for
the first five reversals. (Reversal 0
represents the original problem.)
0 1- 3- 5- 7- 9- 11- 13- 15- 17- 19-
2 4 6 8 10 12 14 16 18 20
Fig. 4. Mean number of errors per reversal averaged across
pairs of reversals.
begun on the trial after criterion was reached, rather than beginning
each new problem on the first trial of a new day as is typically done
in reversal learning experiments.
The median number of errors for each of the first five reversals is
given in Fig. 3. (Medians were used here because the averages were based
upon only eight scores.) It is of interest that the pattern is for an
increase in errors on the first reversal, followed by a consistent drop
in errors thereafter. It is interesting that the average S with a
criterion of six consecutive correct (group C-6) performs perfectly on
the fifth reversal. The group with a criterion of twelve consecutive
correct (group C-12) also shows impressive improvement. However, the
study was not terminated at this point, and the picture changed somewhat
with additional testing. The mean number of errors for each pair of
reversals is given in Fig. 4. Pairs of reversals are used for this and
for statistical purposes to eliminate any effects which position
preferences could have introduced. It may be noted that the original
decline in number of errors in each group was followed by an increase in
errors and then another decrease. Also, both groups are almost identical
for the last eight reversals. Numbers of errors and trials to criterion
for individual Ss on each reversal are given in the Appendix.
An analysis of variance of errors per pair of reversals for group
0-6 is summarized in Table 1. As is indicated in the table, there was
a significant reversals effect. Orthogonal tests for trend showed that
the cubic trend component was significant as was the quartic. An analysis
of trials to criterion data yielded almost identical results except
that the quartic component in the trend analysis was not significant.
Analysis of Variance of Errors
to Criterion for C-6
Source SS df MS F
Between Se 141.2 7
Within Ss [678.6] 
Reversals (pairs) (234.6) (9) 26.06 3.73**
Linear component .01 1 .01 .001
Quadratic component 3.29 1 3.29 .47
Cubic component 139.37 1 139.37 19.99**
Quartic component 28.27 1 28.27 4.05*
Quintic component .04 1 .04 .005
Residual 444.0 63 6.98
As was evident in Fig. 4, the final level of proficiency was no better
than on the original problem.
An analysis of variance of error per pair of reversals for group
C-12 is summarized in Table 2. As in group C-6, there was a significant
reversals effect. However, in this group the linear trend component
was significant as was the cubic component. There were no significant
differences in an analysis of trials to criterion data because of great
variability (a single eixot after eight or nine correct trials has a
drastic effect on number of trials to criterion). A comparison between
the original problem and the last pair of reversals showed that the
final level of efficiency surpassed the performance on the original
discrimination problem (t-1.67, df-22, y<.06). (If the last four rever-
sals are compared with the first problem, t becomes 2.37, df-38, and the
level of significance becomes .02.)
Analysis of Variance of Errors to Criterion for C=12
Source SS df MS F
Between Ss 243.9 7
Within Ss [1362.4] 
Reversals (pairs) (363.8) (9) 40.42 2.55*
Linear component 131.85 1 131.85 8.31**
Quadratic component 26.72 1 26.72 1.68
Cubic component 98.48 1 98.48 6.21*
Quartic component 53.65 1 53.65 3.38
Quintic component 25.64 1 25.64 1.62
Residual 998.6 63 15.85
The within-problem learning of group C012 is illustrated in Fig. 5.
The mean errors per trial are plotted for five-trial blocks, as averaged
across blocks of four reversals. It may be noted that the shapes of
the curves are very similar. Data for group 0C6 are plotted by blocks
of three trials in Fig. 6.
The percentage of correct choices on the first trial of each day as
a function of the number of consecutive correct trials at the end of the
previous day is shown in Fig. 7. As indicated in Fig. 7, Ss which
finished a day with two or fewer correct trials performed at a chance
level at the beginning of the next day. However, binomial tests indi-
cated that a significant majority of those Ss which finished the day with
three or more correct made a correct choice on the first trial of the
following day (for 3-5 correct, z-2.12, 2-.03; for 6-11 correct, z-2.97,
Reversals 1-4 a H
Reversals 5-8 o o
Reversals 9-12 x----
n \\Reversals 13-16 *- -*
S\ \ Reversals 17-20 +
1 2 3 4 5
Fig. 5. Within-problem learning averaged across sets
of four consecutive reversals for group C=12.
Reversals 9-12 o----o
a \\\ Reversals 17-20 +---
1 2 3 4 5
Fig. 6. Within-problem learning averaged across sets
of four reversals for group C=6.
0-2 3-5 6-8 9-11
Correct Trials at End of Previous Day
Fig. 7. Per cent of first trials of each day correct
as a function of the number of consecutive
correct trials at the end of the previous
The results clearly answered several questions, but they raised
still others. The data presented with Fig. 7 made it quite clear that
the caiman is capable of remembering from day to day which way it had
learned to turn the day before. Since many of the Ss were averaging
a reversal each day near the end of training, they had to remember
not only which way they were turning the day before, but they had to
remember which way they were turning at the end of the day rather than
at the beginning. Knowing that the Ss were able to remember from day
to day relieves one of concern about a possible disruption of reversal
learning as a result of the daily break in testing.
From a phylogenetic standpoint, the pattern of within-problem
learning is more like that of the rat (Gatling, 1952) than that re-
ported for the fish (Wodinsky & Bitterman, 1957). The similarity of
the curves presented in Fig. 5 indicates that errors were eliminated in
the same way across the series of reversals. Improvement took part
at all stages of learning within each problem.
There are some very difficult questions to answer regarding inter-
reversal improvement. Why did one group improve while the other group
did not, and why did the Ss in each group improve initially, regress,
and then improve again? Perhaps an attempt to design an ideal experi-
mental situation resulted in a problem which was too easy, at least
for the Ss with the less stringent criterion. In fact, the median
number of errors for group C-6 on the original discrimination problem
was 1.5. This does not leave much room for improvement. The increase
in errors on the first reversal was expected and typical, as was the
rapid decline over the next few reversals. However, when it appeared
that the Ss had completely mastered the situation, some of them re-
gressed severely. The pattern of regression was not uniform enough to
be easily analyzed. About 5 Ss appeared to have developed position
habits which had not been evident in early training. The shock level
which was used was moderately severe, and an error sometimes led to
trials of lengthy duration, which caused some temporary impairment of
locomotion. It is possible that trauma suffered on one of these trials
could have caused an avoidance of one side and, thus, created the
position preferences noted late in training. Several Ss appeared to
anticipate the reversals and would make errors just before they would
have reached criterion; for example, the scores by trials for one S on
one problem were - + + + + + + + ++ + + + + + + +. Perhaps some
of the Ss had learned a temporal discrimination and were trying to
eliminate all errors. Performing on a position basis, while fairly
effective, produced periodic errors, so some of the Ss may have de-
serted this to try other "hypotheses." Since any system of responding
other than a win-stay, lose-shift strategy based upon position would
result in more punishment, a S which had deserted this system would
eventually return to it. This return could account for the final de-
cline in errors seen in both groups. The author has studied data
collected by Jack Sandler in which monkeys in a shock avoidance situ-
ation will sometimes cease to respond for periods of time for no
apparent reason; perhaps this phenomenon is related to the regression
shown by the caimans.
While part of the preceding discussion applies to both groups, it
is certainly not a complete explanation for group 0-12. The more
rigorous criterion made their original problem much harder than that
for group 0.6, so that there was ample room for improvement. Indeed,
their performance at the end of the series of reversals is markedly
similar to group 0=6's performance. One very important difference,
however, is that in the trend analysis there was a highly significant
linear component for group 0-12. This, coupled with the fact that
there was a significant difference between their performance on the
last few reversals and on the original problem, makes their overall
performance quite similar to that usually found in mammals. True, they
did not, as a group, attain single error reversing, but neither did
the mammals in a number of studies.
Behrend, et al. (1965) have mentioned the great individual
differences among their fish. The abilities of caimans in this situ-
ation also varied greatly. One S in group 0C6 is especially noteworthy.
It learned the original problem and all 20 reversals with a total of 36
errors, of which five were on the first reversal. This S, which
learned 13 problems with single errors and had one errorless reversal,
is certainly as efficient as most mammals when it comes to performing
in this reversal learning situation. Given the data from this experi-
ment, Warren would probably have to concede that there is no basis
for stating categorically that mammals have a greater ability to form
habit reversal sets than do reptiles. The data agree very well
quantitatively with typical data from experiments using birds.
Some may argue that the significant cubic trend components indicate
a different functional relationship between reversals and efficiency
in caimans than that found in mammals. One answer would be to ask
what would have happened if the mammals had been tested beyond the
time when they reached their asymptotic performance. After all, the
curves in Fig. 3 look like ideal mammalian results. Further ex-
perimentation is necessary to clarify this point, but the linear trend
in group (=12 indicates that they probably did improve in the same way
as higher forms.
On several occasions, individual Ss made unusually great numbers
of errors before learning a problem. With only eight Ss per group, this
raised the question of whether the means were excessively influenced by
the behavior of one or two individual Ss. However, when the median numbers
of errors per trial were plotted, the general shape of the carve was the
same as the shape of the curve when the means were plotted.
While these data were being collected, one study was published
(Setterington & Bishop, 1967) which may alter some thinking regard-
ing phylogenetic trends in ability to form reversal sets. In this
experiment, seven African mouthbreeder fish did improve progressively
across a series of 80 reversals. They also improved at all stages of
training within problems. Setterington and Bishop used what they called
an "unlimited correction" method (actually a noncorrection procedure
with a variable number of trials per day), daily reversals with 20
"trials" per day, and a spatial discrimination problem with the targets
almost twice as far apart as those used in previous studies. Greater
separation of goal boxes has been shown to improve delay of response
performance by chimpanzees (Carpenter & Nissen, 1934), so it is not
unlikely that separating the targets improved the discriminability of
the situation for the fish. The unlimited correction would also make
it possible for an incorrect response to be repeated indefinitely,
thereby increasing the consequences of an error by delaying a reinforce-
ment for a considerable length of time.
Perhaps comparative psychologists are unable to see the forest for
the trees when they study habit reversal learning. Maybe we will
eventually learn that there is not some point at which animals higher
on the phylogenetic scale have the ability to improve on successive
reversals and those below that point do not. There may not even be
a point which will discriminate between those which can reverse after
single errors and those which cannot. What we might eventually learn
is that those animals which are at the top of the phylogenetic scale
have greater plasticity and can benefit from reversal training in a
variety of situations, while it is necessary to set up an ideal experi-
mental situation to demonstrate improvement across reversals in lower
forms. Questions such as these can only be answered by testing a
greater variety of organisms in a greater variety of situations.
There are some insurmountable problems for one who is interested
in the evolutionary aspects of comparative psychology. Foremost among
these problems is the fact that one cannot study the behavior of extinct
forms of life. Although both mammals and birds evolved from reptiles,
the mammal-like reptiles are now extinct, and there are no living reptiles
which are very closely related to mammals.
Successive habit reversal tasks have shown some promise as a tool
with which to study the differences in learning abilities at various
phylogenetic levels. There is some evidence that there is a trend
toward greater ability to benefit from reversal training at progressive-
ly higher levels, at least within vertebrates. A study was conducted
using spectacled caimans as Ss in order to add to the rather sparse
information available regarding reversal learning in reptiles.
Since there are sometimes differences between the results when the
Ss are given a spatial discrimination as compared with those when the
Ss are given a visual discrimination, Experiment 1 was conducted to
determine whether caimans would normally attend to visual or to spatial
cues. Eight caimans about 18 in. long were trained to escape shock in
a T maze in which both visual and spatial cues were relevant. After a
S made a correct choice on eight consecutive trials, a single test
trial was given with the positions of the visual cues reversed. There-
fore, the visual and spatial cues were giving contradictory information
on the test trial. Seven Ss quickly learned the problem, and all of
them performed on Lhe basis of the spatial cue on the test trial. It
was thus clear that caimans attended to spatial cues in this type of
Experiment 2 was a habit reversal study using the dominant cue of
position as the basis for the two-choice discrimination problems.
Sixteen Sa were divided into two matched groups and trained on a total
of 21 problems, with one side correct on odd-numbered problems and the
other side correct on even-numbered problems. Eight of the Sa were
reversed every time they made six consecutive correct choices, and the
other eight were reversed after 12 consecutive correct.
There was a substantial increase in errors on the first reversal,
as compared with the original problem, followed by a rapid decline in
errors over the next few reversals. However, as testing continued,
there was an increase in errors followed by a second decrease. The
group with a criterion of six correct (C-6) showed no overall improvement
across reversals. This was probably because the original problem was
so easy that there was almost no room for improvement. The group with
the criterion of 12 correct (C-12) did show significant overall improve-
ment and made significantly fewer errors on the last few reversals than
it had made on the original discrimination. The pattern shown by this
group was very much like that usually displayed by birds and mammals,
and the number of errors per reversal was within the range of errors
made by mammals. One of the individual Ss in group C=6 learned 13 of
the 21 problems with a single error. It was certainly as efficient as
most mammals on this type of task. It was concluded that under some
circumstances reptiles can perform with an efficiency equal to that of
many mammals on a successive habit reversal problem. An attempt was
made to explain possible causes for the increase in errors midway in
testing, but further experimentation is necessary before a definitive
explanation can be made.
Alkov, R. A., & Crawford, F. T. Successive reversal training in the
lizard, Iguana iguana. Paper presented at the meeting of the
American Psychological Association, New York, September, 1966.
Bacon, H. R., Warren, J. M., & Schein, M. W. Non-spatial reversal
learning in chickens. Anim. Behav., 1962, 10, 239-243.
Behrend, E. R., Domesick, V. B., & Bitterman, M. E. Habit reversal
in the fish. J. comp. physiol. Psychol., 1965, 60, 407-411.
Bitterman, M. E. Phyletic differences in learning. Amer. Psychologist,
1965, 20, 396-410.
Bitterman, M. E., Wodinsky, J., & Candland, D. K. Some comparative
psychology. Amer. J. Psychol., 1958, 71, 94-110.
Bullock, D. H., & Bitterman, M. E. Habit reversal in the pigeon. J.
comp. physiol. Psychol., 1962, 55, 958-962.
Buytendijk, F. J. J. Uber das Umlernen. Arch. neerl. Physiol., 1930,
15, 283-310. Cited by A. J. North, 1950b.
Carpenter, C. R., & Nissen, H. W. An experimental analysis of some
spatial variables in delayed reactions of chimpanzees. Psychol.
Bull., 1934, 31, 689.
Crawford, F. T., & Henton, W. H. Position response and response shift
in the tarantual with differential visual stimuli. Paper presented
at the meeting of the American Psychological Association, Chicago,
Cronholm, J. N., Warren, J. M., & Harn, K. Distribution of training and
reversal learning by cats. j. genet. Psychol., 1960, 96, 105-113.
Datta, L. G. Learning in the earthworm, Lumbricus terrestris. Amer. J.
Psychol., 1962, 75, 531-553.
Datta, L. G., Milstein, S., & Bitterman, M. E. Habit reversal in the
crab. J. comp. physiol. Psychol., 1960, 53, 275-278.
Davidson, R. S. Laboratory maintenance and learning of Alligator missis-
sippiensis. Psychol. Rep., 1966, 19, 595-601. a
Davidson, R. S. Operant stimulus control applied to maze behavior:
heat escape conditioning and discrimination reversal in Alligator
mississippiensis. J. exp. Anal. Behav., 1966, 9, 671-676. b
Dufort, R. H., Guttman, N., & Kimble, G. A. One-trial discrimination
reversal in the white rat. J. comp. physiol. Psychol., 1954, 47,
Estes, W. K., & Lauer, D. W. Conditions of invariance and modifiability
in simple reversal learning. J. com. physiol. Psychol., 1957, 50,
Gatling, F. The effect of repeated stimulus reversals on learning in
the rat. J. comp. physiol. Psychol., 1952, 45, 347-351.
Harlow, H. F. The formation of learning sets. Psychol. Rev., 1949, 56,
Holmes, P. A., & Bitterman, M. E. Spatial and visual habit reversal in
the turtle. J. comp. physiol. Psychol., 1966, 62, 328-331.
House, B. J., & Zeaman, D. Position discrimination and reversals in low-
grade retardates. J. comp. physiol. Psychol., 1959, 52, 564-565.
Kirk, K. L., & Bitterman, M. E. Habit reversal in the turtle. Quart. J.
exp. Psychol., 1963, 15, 52-57.
Lawrence, D. H., & Mason, W. A. Systematic behavior during discrimination
reversal and change of dimensions. J. comp. physiol. Psychol., 1955,
Longo, N. Probability-learning and habit-reversal in the cockroach. Amer.
J. Psychol., 1964, 77, 29-41.
Mackintosh, N. J. Discrimination learning in the octopus. Anim. Behav.
Supplement 1: learning and associated phenomena in invertebrates,
Mackintosh, N. J. Selective attention in animal discrimination learning.
Psychol. Bull., 1965, 64, 124-150.
North, A. J. Improvement in successive discrimination reversals. J.
comp. physiol. Psychol., 1950, 43, 442-460. a
North, A. J. Performance during an extended series of discrimination
reversals. j. comp. physiol. Psychol., 1950, 43, 461-470. b
Pubols, B. H. Successive discrimination reversal learning in the white
rat: a comparison of two procedures. J. comp. physiol. Psychol.,
1957, 50, 319-322.
Pubols, B. H. Serial reversal learning as a function of the number of
trials per reversal. J. comp. physiol. Psychol., 1962, 55, 66-68.
Reid, R. L. Discrimination-reversal learning in pigeons. J. comp.
physiol. Psychol., 1958, 51, 716-720.
Schusterman, R. J. Transfer effects of successive discrimination-
reversal training in chimpanzees. Science, 1962, 137, 422-423.
Seidman, E. Relative ability of the newt and the terrapin to reverse
a direction habit. J. comp. physiol. Psychol., 1949, 42, 320-327.
Setterington, R. G., & Bishop, H. E. Habit reversal improvement in the
fish. Psychon. Sci., 1967, 7, 41-42.
Thompson, R. Successive reversal of a position habit in an invertebrate.
Science, 1957, 126, 163-167.
Warren, J. M. Reversal learning by paradise fish (Macropodus opercularis).
comp. physiol. Psychol., 1960, 53, 376-378.
Warren, J. M. The comparative psychology of learning. In P. R. Farnsworth,
O. McNemar, & Q. McNemar (Eds.), Annual review of psychology. rclo
Alto: Annual Reviews, 1965, Pp. 95-118. a
Warren, J. M. Primate learning in comparative perspective. In A. M.
Schrier, H. F. Harlow, & F. Stollnitz (Eds.), Behavior of nonhuman
primates. Vol. 1, New York: Academic Press, 1965, Pp. 249-281. b
Warren, J. M., & Warren, H. B. Reversal learning by horse and raccoon.
J. genet. Psychol., 1962, 100, 215-220.
Warren, J. M., & Warren, H. B. Spatial reversal learning by kittens.
j. comp. physiol. Psychol., 1967, 63, 523-525.
Warren, J. M., Brookshire, K. H., Ball, G. G., & Reynolds, D. V. Reversal
learning by white leghorn chicks. J. comp. physiol. Psychol., 1960,
Wodinsky, J., & Bitterman, M. E. Discrimination-reversal in the fish.
Amer. J. Psychol., 1957, 70, 569-576.
Performance of Individual Ss in Group 0-12 on each Reversal
(Errors/Trials to Criterion)
S-2 S-3 S-4 S-5 S-6 S-7
8/28 21/69 2/16 4/21 6/34 6/28
10/33 19/39 2/15 3/17 7/39 8/26
3/20 8/33 6/36 5/19 5/23 7/30
5/23 3/20 8/29 2/14 5/18 5/29
3/26 5/30 5/24 5/31 3/19 5/24
4/29 7/25 3/34 1/13 3/21 7/36
4/19 4/26 6/33 4/23 4/26 3/24
2/19 3/21 3/21 1/13 2/23 6/40
3/28 5/26 2/14 3/22 1/13 3/26
7/43 3/26 5/31 4/20 1/13 5/28
4/27 1/13 6/27 3/20 3/19 4/26
6/33 8/38 2/14 2/14 4/25 6/30
4/25 5/27 7/32 5/31 5/29 2/14
23/82 6/22 8/28 3/16 1/13 2/16
4/18 2/18 7/38 5/24 5/20 4/20
3/20 4/24 1/13 4/25 2/18 3/19
3/20 3/17 6/43 5/30 4/20 3/17
9/30 2/22 4/25 1/13 1/13 3/24
3/15 4/25 2/16 5/25 3/27 1/13
13/67 3/24 2/16 1/13 3/20 1/13
6/29 2/15 4/28 4/19 4/20 2/20
Performance of Individual Ss in Group 0C6 on each Reversal
(Errors/Trials to Criterion)
S-1 S-2 S-3 S-4 S-5 S-6 S-7 3-8
R-0 2/10 4/13 6/16 0/6 1/7 2/8 0/6 1/11
R-1 2/9 8/14 2/8 10/20 5/17 4/12 7/15 5/16
R-2 2/9 3/10 4/12 2/8 1/7 8/27 1/7 6/16
R-3 2/12 3/13 3/15 1/7 1/7 5/19 3/11 2/13
R-4 0/6 0/6 7/16 2/11 2/11 2/9 1/7 2/8
R-5 2/9 1/7 1/7 3/10 1/7 2/8 1/7 1/8
R-6 3/13 2/10 3/9 3/10 2/8 8/25 2/10 2/8
R-7 3/12 5/17 2/12 1/7 1/7 3/10 4/15 3/10
R-8 6/16 2/12 3/13 0/6 4/14 5/16 2/9 1/7
R-9 5/13 4/12 4/16 7/24 0/6 5/15 4/11 2/8
R-10 1/7 2/13 5/21 3/11 1/7 7/30 4/10 1/7
R-11 4/17 4/16 3/13 2/13 1/7 1/10 4/14 4/15
R-12 3/16 3/10 2/10 5/17 3/12 4/18 4/22 3/11
R-13 10/18 2/13 7/19 5/19 1/7 9/26 12/35 6/17
R-14 3/12 5/15 3/16 2/9 5/13 4/13 4/15 3/12
R-15 8/25 3/14 4/15 2/9 1/7 1/7 11/23 6/13
R-16 4/17 3/12 2/8 1/7 1/7 3/14 2/10 0/6
R-17 4/22 4/15 4/14 4/20 1/7 1/7 9/20 5/21
R-18 1/7 2/11 2/12 3/11 2/12 5/23 2/12 2/11
R-19 4/10 3/14 3/13 3/15 1/7 7/15 2/11 2/8
R-20 2/13 3/13 3/12 1/9 1/7 2/14 1/7 1/9
The author, John Taylor Williams, Jr., was born in St. Charles,
Missouri on January 31, 1941. He was graduated from Sunset High School
in Dallas, Texas in 1959. After having attended Southern Methodist
University for two years, he transferred to Arlington State College
where he received his Bachelor of Arts degree in 1963. He enrolled
as a psychology student in the Graduate School of the University of
Florida in 1963 and received his Master of Arts degree in 1965. Since
that time he has been working toward the degree of Doctor of Philosophy.
During his enrollment at the University of Florida, he has spent a
total of eighteen months at the Veterans Administration Hospital in
Augusta, Georgia as a psychology trainee and as a research technician.
John Taylor Williams, Jr., is married to the former Linda Carol
Gratt. He is a member of the University of Florida chapter of Psi Chi,
Phi Kappa Theta, and the Southeastern Psychological Association.
This dissertation was prepared under the direction of the chairman
of the candidate's supervisory committee and has been approved by all
members of that committee. It was submitted to the Dean of the College
of Arts and Sciences and to the Graduate Council, and was approved as
partial fulfillment of the requirements for the degree of Doctor of
August 12, 1967
Dean, College of Arts and Sciences
Dean, Graduate School