•VOLUME 31, No. 3 SEPTEMBER 1954
FOURTH (FINAL) REPORT ON A TEST OF McDOUGALL'S
LAMARCKIAN EXPERIMENT ON THE
TRAINING OF RATS
BY THE LATE W. E. AGAR AND
F. H. DRUMMOND, O. W. TIEGS AND M. M. GUNSON
The Zoology Department, University of Melbourne
(Received n November 1953)
INTRODUCTION
This is the final report on the experiment, begun by us in 1932, and of which three
interim reports have already been published (1935, 1942, 1948). It was essentially
an examination of the well-known experiment of McDougall, purporting to have
demonstrated a Lamarckian effect in the inheritance of an induced light phobia
in rats.
Our experiment consisted in placing the rats into a tank of water from which
they emerged by the choice of one of two exits. Of these one was illuminated, the
other not; and a preference for the non-illuminated (dim) exit was induced in the
rats by electrifying the illuminated exit. With this apparently simple problem of
learning to avoid the lighted exit, the rats were daily confronted until they learnt
to solve it. The number of errors made by the rats was recorded and a sustained
diminution in the number of these errors in successive generations, measured
against a control series of generations, is the criterion for the operation of a
Lamarckian factor.
Over the thirty-two generations of McDougall's experiment, for which records
are available, there was such a progressive decline in the number of errors.
McDougall attributed this improvement in facility in learning to the inheritance of
the effects of ancestral training. At the time, this conclusion was justified to the
extent that no alternative explanation could be advanced to account for it. It was
this that led us, and Crew, to repeat the experiment. Crew (1936) found no
evidence of increased facility in learning during the eighteen generations of his
experiment.
TRAINING PROCEDURE
We will give a brief description of the methods we have used; fuller details are to
be found in our First Report. The apparatus was essentially as designed by
McDougall (1930). It consisted (Fig. 1) of a tank of water divided into three
parallel passages communicating with one another at the far curved end of the
tank. At the near end of each side-passage was a sloping wire ramp up which the
rats could scramble from the water. Behind a sheet of ground glass at the back of
each ramp was an electric lamp which shone down the passage and illuminated its
JEB . 31, 3 21
3o8 THE LATE W. E. AGAR AND OTHERS
communication with the central passage. The circuit was arranged so that one or
the other lamp could be lit alternately. Coupled with the lighting circuit was
a second circuit that electrified the ramp on the illuminated side; a current of
230 V., 1-2 mA. was used, with a duration of 3 sec.
A rat placed in the water at the near end of the central passage swam along it and
then had a choice of two escape routes. If it chose the bright ramp it escaped at the
expense of a 3 sec. electric shock. The rat had to learn to escape always by the dim
exit, irrespective of whether this was on the right side or the left. Facility in learning
was measured by the number of errors made, i.e. the number of escapes by the
bright exit, before it learnt to use the dim one always. A rat was held to have learnt
the task as soon as it made twelve consecutive correct runs.
Our routine procedure has been to wean the rats when 26 days old. To acquaint
Central passage
Fig. 1. Diagram of training tank.
them with the training apparatus, they were given, on the following day, six ' runs'
without illuminated ramps, after which normal training began. This consisted of
four runs per day for 5 days (the animals still being rather small) and thereafter
six per day until the task was learnt. A small proportion of rats, that had failed to
solve the problem by the 52nd day of training, were given 'special training'. They
were, almost without exception, rats that had developed the habit of going exclusively
to one ramp and were quite unable to solve the problem because of unawareness
of the alternative exit. 'Special training' consisted in forcing the animals,
usually against strong resistance, to take the correct pathway. Training of all rats
was continued after learning was complete, and until the time of mating, but was
limited to two runs per day.
McDougalVs Lamarckian experiment on the training of rats 309
CONTROLS
A fundamental weakness in McDougall's experiment was his failure to maintain
a control line of rats for comparison with his trained line. In our experiment we
instituted a proper control line, bred parallel with the trained line and under the
same conditions. All the rats were descendants of a single pair of Wistar origin.
The first generation obtained from this pair (which was not trained) was divided
into two groups, one of which was trained and became the ancestors of the trained
line (T). The other group was not trained and became the ancestors of the control
line (C). In each generation the required number of rats of the trained line was
trained and mated as parents of the next generation. In the control line some
litters were not trained but were kept as parents of the next generation; other
litters of this line were trained to provide controls to the same generation of the
trained line. These trained controls were, of course, not used for breeding. In this
way each generation of the trained line was tested against an approximately equal
number of controls, differing from the trained line only in the fact that their
ancestors were not trained.
In our Third Report we stated that genetic differences in colour pattern and
body size between the trained and control lines had appeared, and suggested that
mutations could have been responsible for the consistent superiority of the trained
line over the control between about generations 12-28. This raised the possibility
that further mutations, having a direct effect on the rate of learning, might occur.
If such mutations accumulated in the trained line their effect would simulate that
of Lamarckian inheritance. To meet this possibility we took the precaution of
maintaining from generation 41 onwards, two trained sublines (TA and TB) and
two control sublines (CA and CB). The offspring of generation 40 of the trained
line were divided into two groups. One became the ancestors of subline TA, the
other, the ancestors of subline TB. The controls were treated in a similar manner.
The two control sublines were thus joint controls to the two trained sublines; there
was no special relationship between TA and CA, or between TB and CB. It may
be stated at once that there was no evidence of divergence, in respect of facility in
learning, between the sublines during the ten generations, 41-50.
MATING AND MORTALITY
The minimum age at which the rats were mated was 85 days. By this time even the
rats which required ' special training' had learnt the task. In order that every rat,
whether it had learnt quickly or slowly, should have an equal chance of becoming
a parent, all the rats of a generation were mated at the same time.
The rats were mated without reference to their training scores. Except for
a short period, brother-sister matings were avoided as far as possible. Not all the
mated rats became parents of the next generation, for many of the matings proved
infertile, and others did not produce Utters till after the number of young required
had been obtained. This, of course, applied to both the trained and control
lines.
3io THE LATE W. E. AGAR AND OTHERS
Of the 4654 rats which started training forty-three died before they had learnt
the task. These forty-three rats have been excluded from our figures. Throughout
the whole of the experiment there were no injuries of any kind attributable to the
electric shock.
MEASURE OF PERFORMANCE
In previous reports we have discussed the problem of finding a satisfactory measure
of the performance of a group of rats as a whole. The use of the arithmetic mean
number of errors made by the rats is unsatisfactory owing to the extreme skewness
of the distribution (Second Report, Table 1) and, in any case, is invalidated by our
practice of giving 'special training' to the very slow learners. In this report, and
for the analyses of the results of the experiment, we have used the measure adopted
in our Third Report: ' The scores of the first thousand control rats were arranged
in order of magnitude, and the whole group divided into ten classes, each containing
as nearly as possible an equal number of rats, having regard to the fact that the
number of errors are necessarily whole numbers.'
The resulting distribution is shown in Table 1.
Table 1. The first 1000 controls classified according to the number of errors made
Class
Thus all the rats with training scores 0-5 errors inclusive are placed in class 1,
and so on. The arithmetic mean of the classes so obtained will be referred to as the
mean class of the group of rats concerned.
BODY SIZE IN THE TRAINED AND CONTROL LINES
In our Third Report we discussed a genetic difference in body size between the
trained line and the control. Weighings made in generations 25-28, and also in
generations 34-36, showed that the rats of the trained line were substantially
heavier than the controls. At 26 days old the difference in mean weights was
approximately 13 g. (Table 2).
Further series of weighings at 26 days old were made in generations 49 and 50.
These showed (Table 2) that there was no difference in weight between the two
trained sublines, but that the mean weights, by comparison with those of generations
previously weighed, had fallen by about 9 g. The mean weights were: females
(85) 436 g., males (70) 45-2 g. Only 6% of the rats in these two generations
weighed more than 50 g.; in generations 25-28 and 34-36, 60 % exceeded this
weight.
In the controls, the mean weights of the two sublines in generation 49 were much
the same. They conformed to those of earlier generations of controls and this was
also true of subline CA of generation 50. However, in subline CB of this generation
McDougalVs Lamarckian experiment on the training of rats 311
the mean weights were: females (23) 43-2 g., males (23) 43-8 g. These are practically
the same as those of the trained sublines.
Greenman & Duhring (1931) have recorded the weights of a large number of
rats from the Wistar Institute colony. Over a period of 4 years, eight groups of
males and females were weighed. At 25 days old, the mean weights of the groups
varied from 343 to 48-6 g. If 2-5 g. is added to the figures of Greenman & Duhring
to allow for the fact that they refer to rats 1 day younger than ours, the total mean
weights become: females (423) 43-2 g., males (455) 43-8 g.
Thus, at the end of our experiment there was no evidence of the genetic difference
in body size which previously had distinguished the rats of the trained line from the
controls and also from Wistar Institute stocks.
Table 2. Mean weight in grams, with standard errors, of rats at 26 days.
The figures in brackets are the number of rats weighed
Generation
GENERAL RESULTS OF EXPERIMENT
Data covering generations 1-36 are given in our earlier reports. Tables 3 and 4 of
the present report give the data for generations 37-50 which conclude the experiment.
The results are summarized in two graphs (Figs. 2, 3). Fig. 2 shows the
annual performances over the 20 years of the experiment; Fig. 3 gives the performances
of successive generations. In the latter figure, we have arranged the
generations as nearly as possible in groups of 4 (see Table 4), in order to minimize
chance fluctuations.*
The general result is that periods of progressively decreasing scores have
alternated with periods of progressively increasing scores, and in this the controls
have participated. Thus there was a fairly regular decrease in the number of errors
during the first sixteen generations, a slight increase in the following four and then
a further decrease until the twenty-eighth. Over the next eight generations there
was a marked increase in the number of errors, and high scores were maintained
until the 40th generation after which a further decrease occurred. Thus in spite
of the great improvement during the first half of the experiment the scores of the
• The last group necessarily contains only two generations (49 and 50) and in the first group of
controls only generations 2-4 are, of course, included.
Table 3. The number of errors made {shocks received) by each rat, the median
number of errors, and the mean class, in each of generations 37-50
(T, trained line; C, control line. S indicates that the rat qualified for special training. In the trained line the rats which became parents of the next generation are in heavy type.)
Genera- tions
No. of
rats Median Mean class No. of errors made by each rat
THE LATE W. E. AGAR AND OTHERS
last six generations were of the same order as those of generations 13-16. Throughout
the whole experiment the parallelism between the performances of the trained
and control lines was remarkable; over the last few generations the controls were
generally even superior to the trained line.
Table 4. Summary of the results of the fifty generations in
groups of four generations
Genera tions
1933 '34 '35 '36 f37 '38 '39 '40 '41 '42 '43 '44 '45 '46 '47 '48 '49 '50 '51
Fig. 2. Continuous line, line T; broken line, line C. The mean classes are those of all rats which
began training in the year referred to. The first point, 1933, includes some rats which began
training late in 1932; the last point, 1951, includes some rats which began training early in 1952.
It is unfortunate that McDougall did not publish full details of the performances
of his rats; his reports give only the arithmetic mean of the scores in each generation
and the scores of the best and worst rats. But although an accurate comparison of
the rate, and extent, of changes in learning in the two experiments cannot be made,
it is clear that the improvement which characterized our first twenty-eight generations
closely parallels that of McDougall's thirty-two generations and it seems
probable that the same factor, or factors, operated in the two experiments. But
McDougaWs Lamarckian experiment on the training of rats 315
McDougall's claim that the improvement was due to Lamarckian inheritance is
plainly invalidated first, by the performance of our control line and secondly, by
the fact that, in our experiment, the improvement was not maintained in later
generations.
DISCUSSION OF RESULTS
What is the explanation of the observed changes in the rate of learning? Selection
can be ruled out. McDougall found that improvement was continued when he
deliberately practised adverse selection and, in our experiment, it would be difficult
to explain on any selection hypothesis, why, with a standardized system of training
and mating, the direction of change in rate of learning should be periodically
reversed. The parallel performance of the trained and control lines suggests that
J I
1-4 5-8 9-12 13-16 17-20 21-24 25-28 29-32 33-36 37-40 41-44 45-48 49-50
Generations
Fig. 3. This figure shows Table 4, column Mean Class, in the form of a graph. Continuous line,
line T; broken line, line C.
the changes were related to factors, not necessarily having any genetic basis, which
influenced the rate of learning.
In our First Report we listed six such factors : (1) the severity of the punishment,
(2) vigour, (3) intelligence (ability to learn by experience), (4) the strength of the
right or left habit, (5) 'venturesomeness', (6) chance factors not causally related
to the learning process at all. Subsequent analysis of the data has shown that yet
another factor needs consideration and that the performance of the rats was influenced
by the season of the year in which they were trained. The separate or
additive effects of these seven factors could explain the great variation in the performances
of individual rats and the differences between particular generations,
but only the first two, i.e. severity of punishment and vigour, seem to offer any
basis for an explanation of trends of improvement or decline extending over a
number of successive generations.
(a) Seasonal effect
When the rats of the whole fifty generations of the two lines are grouped according
to the month in which they commenced their training and the mean class
is calculated for each of the 12 months, it is found that, starting in February, the
316 THE LATE W. E. AGAR AND OTHERS
means increase regularly to reach a maximum value in July and then decrease
regularly to a minimum value in November (Fig. 4, Table 5).
In order to find whether this seasonal factor had operated throughout the
experiment, similar analyses were made after dividing the experiment into four
5-year periods. In the early months of the year the results were erratic, but from
March-April onwards generally conformed to the pattern of the previous analysis
and showed consistently that rats which commenced their training during the
winter months of June, July and August were at a disadvantage by comparison with
those which commenced training in November and December.*
This effect may have been due, in part, to our failure to maintain a constant
temperature in the colony room and in the water in the tank. The only precautions
taken were, that in winter, the room was heated and warmed water was used in
the tank. In neither case, however, was the temperature raised to summer levels.
Jan. Feb. Mar. Apr. May June July Aug. Sept. Oct. Nov. Dec.
Fig. 4. The mean classes are those of the rats which began training in the month and period referred
to. , 1932-52; , :932-7J • 1938-42; , 1943-7;--, 1948-52.
The lower water temperature during the winter may have had a slightly adverse
effect on the performance of the rats, for Wever (1932) and Hack (1933) have found
that the rat's incentive to escape from immersion varies with water temperature,
being greatest at low temperatures. At lower temperatures (10-200
C.) the rats
after being placed in the water swam strongly and directly to the landing platform.
At temperatures from 30 to 400
C, Wever states that 'many of the animals did not
head for the goal immediately but spent some time in casual exploration'. Hack
describes the rats at 37-5° C. as showing an 'inquisitive attitude' with frequent
reversals and re-entries into the blind alleys of the maze. This latter type of behaviour,
we believe, facilitated learning in our experiment, but we are disinclined
to attribute seasonal variation in performance to differences in water temperature;
for the greatest range in temperature would not have exceeded 12-220
C. and the
• The main divergences from the pattern are the low values of the class mean in the May group, 1948-52, and the June group, 1932-37. In both groups the number of rats was small—considerably
less than half the average number for the other months.
Table 5. The rats of lines T and C combined classtjied accordi7lg to the month in which thq, began their training
318 THE LATE W. E. AGAR AND OTHERS
maximal temperatures would have occurred in summer, not in November, which
was the period of minimum scores. But in any case, seasonal effects cannot possibly
account for trends, lasting over a period of some years, and it is primarily with
these that we are here concerned.
As will be shown later, we do attach great importance to the effects on learning,
of differences in behaviour in the tank, but we believe that these differences were
largely the expression of variations in the health and vigour of the rats.
(b) Severity of punishment
In his preliminary experiments, McDougall found that the strength of shock had
considerable effect on the rate of learning. Thus if the severity of punishment tended
to wax and wane over long periods the performance of the rats would show corresponding
trends. We are satisfied that this effect did not operate in our experiment.
The importance of a standardized punishment was recognized from the beginning
and the precautions taken to ensure this (see First Report) have been maintained.
The electrical installations, as shown by periodical tests, have remained in perfect
order.
(c) Vigour
McDougall, Crew and ourselves have all noted that less vigorous rats tended to
learn more quickly. Crew attributed this to their receiving more severe punishment.
We, however, agree with McDougall that it was mainly a result of their slower and
more hesitant progress through the water which gave them more time to perceive
the situation.
McDougall, while admitting the relationship, was inclined to minimize its effect.
He wrote (1938, p. 374): 'very great vigour and liveliness is a little unfavourable
to quick learning and a somewhat diminished or less-than-average vigour is probably
slightly favourable'. Crew, having noted that poorly developed, feeble rats learnt
quickly, suggested that ' as a general rule, the more vigorous the rat the higher the
score may be expected to be'. In our First Report we produced evidence to support
this view. Fifteen rats of the 3rd-5th generations which, before the 10th day of
training and before learning, had been noted on the training cards as 'weak' or
' undersized' had an average score well below the general average of these generations.
There was also evidence from generation 17 which was severely debilitated
by a mite infestation. After this had been controlled and the diet changed there
was an extraordinary improvement in the health of the colony and a striking change
for the worse in training performances; forty-five rats trained prior to the eradication
of the mites had a class mean of 3 29; thirty-five rats born afterwards had
a class mean of 6-40. But in the following generations, although the rats remained
in a good condition, there was a return to low training scores and we were then
inclined to accept McDougall's conclusion that variation in health and vigour
could not explain the fluctuations in the rate of learning.
Further experience led us to revise this opinion. During the course of the
experiment, the general health and fertility of the rats varied considerably. At
irregular intervals the colony went into a decline extending over several generations
McDougalVs Lamarckian experiment on the training of rats 319
and then, for no apparent reason, regained its health and vigour. McDougall
evidently had the same experience for he refers to 'waves of decline of vigour'.
Greenman & Duhring (1931), working with a selected group of rats at the Wistar
Institute, report large weight fluctuations over a succession of generations so that
changes in the general physique of the albino rat may occur, even in colonies
maintained in the most favourable environment. The occurrence of such changes
in our colony, coupled with the facts set out above, justified a detailed analysis of
possible relationships between health and training scores.
As the need had not been foreseen, proper records of the health of the rats were
not kept and the data available are therefore few. We have, however, fairly complete
fertility records, and as fertility is correlated with health these have provided us
with an indirect measure of health. Since all the members of a single generation
were mated at the one time, they were all therefore given an equal chance of
reproducing; and it is justifiable to use, as an index of fertility, the number of
fertile rats in each generation, expressed as a percentage of the number mated.*
The rats were weaned at 26 days and commenced their training on the 28th day.
The great majority (about 80%) learnt during the first fortnight of training, i.e.
between the ages of 31 and 42 days, when their general health would be determined,
in large measure, by the nursing capacity and therefore the general health of the
mother.
Thus if rate of learning were influenced by the health of the rats one would
expect a correlation between fertility, used as a measure of health, of one generation,
and the training scores of the next.
The first analysis was made in respect of generations 1-40, i.e. for the period
prior to the splitting into sublines. The coefficient of correlation, for the trained
line was +040 and for the control line +042. Both values are significant at the
1 % level, and fertility is therefore shown to be positively correlated with high
training scores.
The analysis was then extended to include the whole fifty generations, treating
the trained rats and the controls each as a single population. For the trained line
there was again a positive correlation significant at the 1 % level. For the control
line the correlation was positive but was not significant.
When the control sublines of generations 41-50 were analysed separately, it was
found that there was no significant correlation between fertility and training scores
in either subline, but that in subline B, such correlation as occurred, was negative.
This latter fact was not wholly unexpected for until the 48th generation it had been
obvious at the time of training that subline B was atypical in that, while fertility
was fairly high, the rats were undersized and poor in health. Their inferiority was
indicated by their small litters. In every generation from 41 to 48, the average litter
• In the trained line our records show, as fertile, only those rats whose offspring were taken into training. When fertility was high, some first litters were discarded without their parentage being
recorded and the fertility index would thus give an underestimate of the true level of fertility. This situation did not often arise, and, in fact, the value of the index for the trained line was not signi- ficantly different from its value for the controls, where all first litters were required, either for breeding or training.
320 THE LATE W. E. AGAR AND OTHERS
size in subline B was smaller than in subline A. For the eight generations combined
the average for B was 7-3, for A 93.
While there are grounds for regarding control subline B as atypical, the occurrence
of a negative correlation between fertility and scores is disconcerting. It does
not, however, discredit the highly significant correlation established for the first
forty generations of the control line and for the whole fifty generations of the
trained line.
The possibility that there might be a correlation between fertility and rate of
learning was considered by both Crew and McDougall. Crew pointed out that if
there were a positive correlation between fertility and quickness, i.e. the reverse of
the one established above, it could explain the progressive improvement of
McDougall's rats. McDougall had been fully aware of the significance of the point
made by Crew, but such evidence as he collected, supports our conclusion. In his
water-maze experiments, several attempts to isolate a superior stock, by breeding
from selected quick learners, failed on each occasion because of the infertility of
the superior rats. He had the same experience in his first selection experiment with
quick learners in the tank.
We have established a positive correlation between the fertility of one generation
and the scores of the next but it is difficult to believe that there could be any direct
causal relationship between them. They must have been connected by a third
factor. The basic premise of the foregoing analysis is that this third factor was the
general health and vigour of the rats. If it be accepted that healthy vigorous rats
are more fertile than less vigorous ones, then the above correlation indicates that
the more vigorous the rat the higher its expected training score.
Tryon (1929, 1932) has suggested that the reverse relationship may hold for
maze-learning. By breeding rats selectively, according to their ability on a maze,
he developed a strain of 'brights' and a strain of 'dulls'. Over a number of
generations he found that both lines showed a progressive improvement in learning
ability. Tryon was inclined to attribute this improvement to increased vigour of
the rats. Were this so, it would not necessarily conflict with our conclusion on the
influence of vigour on learning in the tank. Krechevsky (1932) has analysed the
performances of Tryon's two strains, and has concluded that the difference between
them was related specifically to maze learning. The 'brights' learned quickly
because they used 'spatial hypotheses'. The 'dulls' used 'visual hypotheses'.
Drew (1939) has pointed out that this, while penalizing them on a maze, would have
favoured them in a Ught-o^scrirriination test, and that Tryon's 'dulls' would
probably have been quick learners in the tank. The two learning situations were so
different that vigour could have been favourable in one, and a handicap in the other.
The training scores of individual rats were necessarily influenced by a variety of
factors and all of these, no doubt, played some part in causing the changes in the
average rate of learning which occurred during the course of the experiment.
We believe, however, that the major changes, that is the changes involving a progressive
improvement or decline extending over several generations, were due
primarily to changes in the general level of health in the rat colony. The cause of
McDougalVs Lamarckian experiment on the training of rats 321
these fluctuations in health of the colony is quite unknown. It may be that they
result from infection, and that recovery from a decline involves a selection process,
in which the enfeebled strain is eliminated through diminished fertility.
In retrospect it seems that we have been less successful than Crew in standardizing
the factors that cause variation in the rate of learning. This has, however, had the
positive advantage that it has enabled us to obtain the effect of a progressive improvement
in learning rate that McDougall found. McDougall attributed it to the
operation of Lamarckian inheritance. Our own results forbid this interpretation
for the effect is not sustained, and is displayed in equal measure by the controls.
SUMMARY
This is the final report of an experiment of 20 years' duration, in which we have
repeated, in its essentials, the well-known experiment of William McDougall purporting
to reveal a Lamarckian inheritance of the effects of training on rats. The
test is one involving light discrimination, and McDougall recorded a steady improvement
in the rate of learning on a succession of 32 generations; but he omitted
to check the results against a properly conducted control.
Our experiment confirms McDougall to the extent that we too have obtained
long duration trends of improvement in learning-rate (Figs. 2, 3); but we find that
the effect is not sustained, and that it is, moreover, shown also by a control experiment,
using animals of untrained ancestry. This forbids a Lamarckian interpretation.
Statistical analysis of the data indicates that the ' condition' of the rat markedly
affects its speed of learning, and that progressive changes in learning-rate, over
a succession of generations, are in reality correlated with the health of the laboratory
colony, which is subject to periods of decline and recovery.