Third Report On A Test of Mcdougall’s Lamarckian Experiment on the Training of Rats

The nature of the experiment is well known. The rats are ‘trained’ (but actually have to discover for themselves) to choose the less brightly illuminated of two exits from a tank full of water, and the number of errors (choices of the bright exit) made before they take to using the dim exit exclusively, is recorded generation by generation. McDougall’s experiment extended over forty-four generations, but records of only the last thirty-two are available and relevant. Over these thirty-two generations McDougall records a marked and progressive (though of course not quite regular) decrease in the number of errors made in successive generations. Unfortunately, he made the irremediable mistake of failing to maintain a parallel control line for comparison with his trained line. Crew repeated the experiment, with some inessential modifications of training technique, and found no evidence of a decrease in the number of errors during the eighteen generations of his experiment.

As it is five years since we published our last report, we will repeat a brief description of the methods employed. Further details are to be found in our First Report.

The apparatus is essentially as figured by McDougall (1930). It consists of a tank, containing water, divided into three parallel passages by two partitions which stop short of the far, curved end of the tank. At that end, therefore, the passages communicate. From the near end of each side passage a gangway in the form of a wire ladder leads to a platform above the water-level. The rat is placed in the water at the near end of the central passage. In its search for a way to escape from the water it swims down the central passage, turns right or left into one of the side passages, and swims back along this to the gangway at the end of it and climbs out. Behind a sheet of ground glass at the back of each gangway is a low-power electric lamp, which shines through it down the whole length of the passage, illuminating also its communication with the central passage, so that the rat at its starting-point in the central passage can see which of the two side passages is illuminated. The rat is given six trials a day (except for the first 5 days of training, when, unlike McDougall, we give only four trials) with the left and right passages illuminated alternately. The gangways are connected with an alternating current in such a way that the current is thrown into whichever gangway is illuminated, but not into the other. The rat can therefore escape from the tank by either gangway, but if it chooses the illuminated (‘bright’) one it does so at the expense of a 3 sec. electric shock. The rat has to learn always to escape by the dim gangway, irrespective of whether this is on the right or left. Facility in learning the task is measured by the number of errors, that is to say, number of escapes by the bright gangway, made by the rat before it learns to avoid this and always chooses the dim one. Occasionally a rat makes contact with the electrified gangway, but instead of climbing it, turns back on receiving the shock, swims to the other exit and escapes by it. This very rarely happens except during the first few days of training. This is recorded as an error. The number of errors made is therefore the same as the number of shocks received. A rat is held to have learnt the task as soon as it has made twelve consecutive correct runs. Further details of the apparatus and training procedure are set out in our First Report, and have not been changed since. The various factors which influence the rate of learning are also discussed in that report, especially those which appear to be responsible for the great variation in the number of errors made by individual rats even of the same litter.

All the rats are descended from a single pair of albino rats of Wistar origin. The first generation obtained from this pair (which was not trained) was divided into two groups, one of which (five rats) was trained and became the ancestors of the trained line; the other group (four rats) was not trained and became the ancestors of the control line. The two lines have been bred parallel with each other, and under the same conditions. In each generation the required number of rats in the trained line are trained, and mated as parents of the next generation. In the control line some of the litters produced are not trained, but are kept as parents of the next generation. Other litters of this line are trained to provide controls to the samé generation of the trained line. These trained controls are, of course, not used for breeding. In this way each generation of the trained line is tested against an approximately equal number of trained controls, differing from the trained line only in the fact that their ancestors were not trained.

At the time of weaning each rat is given its individual mark by means of dye marks on different parts of the body, a different colour being used for each of the three groups, the trained line, untrained controls for breeding, and trained controls. These dye marks are renewed from time to time before they fade, so that track is kept of every rat throughout its lifetime. As an additional safeguard the three groups are permanently distinguished by ear clippings.

Our system of mating ensures that the number of errors made by any rat in the trained line does not influence its chance of becoming a parent. The rats are weaned and given their identification marks when 26 days old; on the next day they are given six preliminary runs in the tank without the differential illumination of the gangways, or the shock. Training proper begins on the next day when they are 28 days old. As described in our First Report, the few rats which have not learnt after 52 days of training (by which time they are 80 days old) are given ‘special training’. These slow learners are usually rats which, early in training, have formed the habit of using exclusively either the right- or left-hand exit passage, whether this is illuminated or not. Consequently after adopting this habit, they receive a shock on every alternate run. As the rats are given four trials a day for the first 5 days, and six a day thereafter, by the end of 52 days they have had 302 trials, and have received approximately 151 shocks. ‘Special training’ consists in forcing them for a few times to use the unfamiliar exit. After this they invariably learn to use the dim passage, either at once or after a very few days. Thus even the slowest learners have completed their task before the age at which any rat is mated. In fact, the reason for giving ‘special training’ to these rats is to ensure that slowness in learning shall not result in later age of mating (for they cannot be mated before they have learnt) and therefore in diminishing their chance of becoming parents of the next generation. The number of rats which receive this ‘special training’ can be seen from Table 2, where they are shown by the letter S.

The rats are mated without any reference to their training scores; we have avoided brother-sister matings as far as possible, except during a short period of the experiment to be referred to later. Not all the rats mated become parents of the next generation, for many of the matings prove infertile, and others do not produce litters till after the number of young required has been obtained. This, of course, applies to both the trained and control lines. It is our practice to reduce litters of more than six young to that number within a few days of birth, occasionally leaving as many as eight in second litters of large mothers.

Of the total 2848 rats which have started training, twenty-one died before achieving the criterion of learning to escape by the dim gangway. These twenty-one rats are excluded from our figures. Throughout the whole of this experiment there have been no deaths, or injuries of any sort, attributable to the electric shock.

In both our previous reports we have discussed the problem of finding a satisfactory measure of the performance of a group of rats as a whole. McDougall’s method of measuring this by the arithmetical mean number of errors made by the rats is unsatisfactory owing to the extreme skewness of the distribution (as shown by our Second Report, Table 1). Moreover, such a measure is further invalidated in our case by our practice of giving ‘special training’ to rats which have not learnt after 52 days of training. In our Second Report we adopted three measures, the median score, and the percentages of rats learning with less than ten and more than 100 errors. We have now reverted to the type of measure we used in our First Report, but based on much larger numbers. 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.

Thus all 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.

Table 2, which is a continuation of Table 2, Second Report,^* with the addition of the column Mean class, gives the training scores of every rat since our last report. Table 3 gives the figures for the whole experiment to date, in groups of four generations. As, however, no rats were trained in the first generation of controls, the first group contains only generations 2, 3 and 4 of line C. The mean class and percentages are the weighted figures, i.e. each group of four generations is taken as a single unit.

In our Second Report we drew attention to periodic changes in facility of learning (as measured by the number of errors made) which ran roughly parallel in the trained and control lines. That this tendency has continued is shown by Table 3, and in graphic form by Figs. 1 and 2. The greater smoothness of the graph in Fig. 2 is due to the large number of scores on which each point is based. For example, the violent fluctuation of line T in 1936 (Fig. 1) is based on only twenty-five rats.

It will be seen that during the first fifteen or sixteen generations, trained in 1932-9, there was a progressive decrease in the number of errors in both lines. Then, in spite of considerable fluctuations, both lines remained at this low level of errors, or even with a slight further decline till about generations 28-30, trained in 1944-5. About this time the number of errors began to rise again in both lines, this trend being continued to the end of the period under review (April 1947).

A glance at Table 3, column Mean class, and at Figs. 1 and 2, shows how the validity of McDougall’s conclusions suffers from his failure-to maintain a control line. Had we not done so, and if our experiment had stopped at about generation 28—or even at generation 32 or 33—our results could have been used as evidence of the operation of a Lamarckian factor. But the parallel behaviour of the two lines throughout, as well as the later rise in the number of errors, excludes this interpretation.

We have no explanation to offer for these long-scale changes. As they are exhibited in a parallel way by both lines, between which there has of course been no cross-breeding since they were established from the same pair of parents in 1932, it is hardly possible that the factors concerned are genetic. Of possible environmental factors the one which immediately suggests itself is variation in the severity of the punishment for taking the bright gangway, which as McDougall showed (1930) influences the number of errors made. This must depend on the strength and duration of the electric shock. Our methods for keeping these constant are described in our First Report (p. 194); they have not been varied since. During the training of generation 35 the electrical resistance which had been in use from the beginning of the experiment was tested and found not to have altered to a perceptible degree, and the circuit between the resistance, the gangways and the water in the tank, which is closed by the rat placing its forepaws on the wire ladder with its body in the water, was found to be in perfect order. Moreover, in spite of their increasing persistence in taking the bright gangway during the last few generations, the rats’ reaction to the shock on the electrified gangway is as vigorous as ever. The lamps used for illuminating the gangways are changed every 6 months as a matter of routine. We therefore have no reason to believe that the slow changes in rate of learning observed during the course of the experiment can be attributed to changes in training technique or apparatus. The diet and general husbandry has not been significantly varied since generation 17 (Second Report, p. 163). In fact, we have no explanation of these changes to offer. The fact that the fluctuations run parallel in the two lines excludes, however, not only the Lamarckian factor, but genetic factors in general.

There is, however, another fact brought out by Table 3 and Figs. 1 and 2. Although both lines participated in the decline in the number of errors during the first sixteen generations, line T reached a lower level than line C, and maintained its superiority until about generation 32, after which the mean classes of the four-generation groups are practically equal. The superiority of line T over line C during the sixteen or twenty generations was mainly, though not entirely, due to the excess of very quick learners in line T (Table 3, percentage in class 1 ; the violent fluctuations in the percentage are partly explicable by the comparatively small number of rats concerned).

It seems impossible to account for this prolonged difference between the two lines except by a temporary genetic difference, on which is superimposed non-genetic factors producing their parallel fluctuations. Clearly it is not due to a Lamarckian factor, for if this were concerned the divergence between the lines should increase as the number of generations of trained ancestry in line T increases. That this is not the case is shown by the column ratio T/C in Table 3.

Our conclusion is therefore that a genetic difference between the two lines in regard to learning facility appeared about generation 12, and disappeared, or greatly diminished, between generations 28-36. As we shall now describe, we have discovered two other independent genetic differences between the two lines which must be accounted for in some other way than by the accumulation of trained ancestry in the one line and not in the other. We are indeed faced with a difficulty inherent in all experiments requiring the independent maintenance through many generations of a control line for comparison with an experimental line. Whatever care is taken to ensure that the two lines are genetically identical to begin with, mutations will in time destroy this identity.

At about the 20th generation the fact had gradually impressed itself upon us that the rats of line T were almost consistently much larger and more vigorous than those of line C. The superior vigour of line T is manifested, among other things, by a much greater liveliness and eagerness when the lid of the cage is opened. This, however, may perhaps be accounted for by the greater tameness of the rats of this line owing to their handling when given their daily swim in the bath. This exercise, which continues from the age of 27 days till mating at 90 + days is also the only difference in the conditions under which the two lines live.

Systematic weighings were made in generations 25-28, the rats being weighed when 26, 75 and 125 days old. Line C rats, weighed at 75 and 125 days, were of course the untrained controls used for breeding. For the weights at 26 days it must be remembered that litters of more than six (which constitute the great majority of all litters) are reduced to that number within a few days of birth, except in a very few cases of second or later litters from large mothers which are left with seven or eight young to rear.

The mean weights with their standard errors are given in Table 4. If will be seen that line T rats are much larger at all three ages than those of line C. At 26 days the mean weight for line T exceeds that of line C by 31 % of the latter.

Although it was obvious that this size difference was persisting, further weighings were made in generations 34-36. Fifty males and fifty females of both lines were weighed when 26 days old, with the result: T ♀ ♀ 53’0 ±0’7 g-, T ♂ ♂ 54-8 + 0-7 g., C ♀ ♀ 38-0 + 0-7 g., C ♂ ♂ 40-3+0-7 g. In addition, eighty-five rats were weighed when 75 days old; the size differences between the two lines were again shown to be fully maintained—indeed, as at 26 days, they were slightly greater than in generations 25-28.

There remains to be discussed the possibility that this difference between the two lines is not genetic, but phenotypic only. It might be that the larger size of line T rats at maturity is due to their daily exercise in the tank, and that this accounts for the larger size of their offspring at weaning. The following experiments show, however, that the difference is inherited.

It is our custom to terminate the training of line C rats as soon as they have achieved the twelve consecutive runs to the dim gangway, so we have not many rats of that line which are still in training at the age of 75 days. In generations 34-35, however, owing to the slow rate of learning in these generations, to which we have already drawn attention, an unusually large number were still in training at that age. The mean weights of four females and five males still in training at that age were 136 and 187 g. respectively. Conversely, we reared a few rats of line T without training; at 75 days the mean weight of fourteen females was 175 g., and of eleven males the exceptionally high figure 247 g. This shows that giving the rats of line C a daily swim in the bath, and depriving rats of line T of this exercise does not diminish the difference in weight between the two lines at the age of 75 days. Moreover, a few (16) offspring from these untrained rats of line T were weighed, and found to be above the mean weight of the line.

The fact that the size difference between the lines is genetic was confirmed by making a number of crosses between the two lines.^* Ten such litters were obtained from the cross C ♀ × T ♂J, yielding 28 ♀ ♀ and 28 ♂ ♂. The mean weight of the females at 26 days was 44-1 g.,- and of the males 45-2 g. These weights are intermediate between the weights of the pure lines. We have only three litters from the reciprocal cross T × C; at 26 days the mean weight of the nine female offspring was 51-9 g., and of the nine males, 53-1 g. This is actually heavier than the weights of the pure line T as shown in Table 4, but not significantly different from the weights of that line in generations 34− 36, the.contemporaries of the heterozygous litters in question.

These results are consistent with dominance of the heavier weight combined with an inability of the smaller line C mothers to provide the intra-uterine, and possibly lactation, conditions to satisfy the full growth capacity of their heterozygous young which is so much greater than that of the usual offspring. In this respect these crosses resemble those between the horse and the ass, between species of Cavia, and between large and small races of rabbits, in all of which the mother has greater influence than the father on the body size of the offspring (Castle, 1941).

In order to elucidate further the genetics of the size difference between the two lines several of the young from the cross C♀ × T ♂ were reared and mated together. Thirteen F₂ litters were thus obtained. At 26 days the mean weight of the thirtyseven females of these litters was 49-7 g., and of the thirty-nine males 51-7 g. These weights are only slightly less than those of the pure line T in generations 34-36 with which they were nearly contemporary.

There is no evidence of the segregation within these litters which is to be expected if the size difference between the lines is due to a single or few genes. Certainly there is no clear differentiation into two or three classes, nor any appearance of rats of the small size characteristic of line C. In the pure lines an occasional pathologically underdeveloped runt is found at weaning time, and rejected. One such runt occurred in the F₂ litters, weighing 26 g., its next smallest litter mate weighing 53 g. This rat has been omitted from the group. Except for it, not one of the rats is as light even as the mean weight of line C. Nineteen of them are heavier than the mean weight of line T, generations 34-36.

To obtain a rough quantitative comparison of the variability of these thirteen F₂ litters with that of the pure lines, the average deviation of each litter was found (without distinction between males and females), and also of fifteen litters in each of the pure lines. The mean average deviation of the F₂ litters is 2·30, of the line T litters 2·55 and of the line C litters 1·89. Expressed as percentages of mean weight, 100AD/M, these figures become: F₂, 4·56; line T, 4·81; line C, 4·90. The fact that the variability of the F₂ litters is no greater than that of the pure lines^* is surprising on the ordinary genetical theory of the inheritance of quantitative characteristics. If it is assumed, nevertheless, that the size difference, being inheritable, is the expression of gene differences, one is forced to the conclusion that many genes are concerned, which is again surprising in view of the short time since the two lines were derived from the common ancestral pair.

On ordinary genetical theory there are two ways in which this genetic difference could have arisen. Either the original pair, from which both lines are descended, was heterogeneous for genes affecting size, and by chances of segregation genes favouring larger size have become concentrated in line T, and genes for smaller size in line C; or else mutations have taken place in one or both lines. It seems almost certain that segregation from the original pair must be excluded, for no difference between the two lines was noticed in the earlier generations ; later, as shown by the figures quoted, the difference was far too great and constant to escape notice, not only by the eye but in the handling of the rats. Moreover, we happen to have some records of weights at 26 days old in generation 5, made for another purpose. For some reason, probably less efficient husbandry, the weights of the whole colony were much lower than in the later generations. The mean weight in line T (10 ♀ ♀ and 12 ♂ ♂) was 32·2 g., and in line C (21♀ ♀ and 23 ♂ ♂) was 30·3 g., giving a difference of 1.9 ± 0.9 g. in favour of T. This difference is not statistically significant, and in any case is of quite a different order of magnitude from the difference in later generations.

Comparison with the weights of Wistar albino rats given by Greenman & Duhring (1931), and by King (1915), suggests that the difference between our two lines has been caused by changes in line T, rather than in line C, during the period that has elapsed since their common origin. Greenman & Duhring record a large number of weighings spread over 4 years. Eight groups of males and females were weighed at 25 days old. The mean weights of the groups varied from 34·3 to 48 6 g. Thus even under the expert care available for the Wistar Institute colony there is great variation in weight at different times, which the authors show cannot be attributed, except in slight degree, to seasonal variation. The total mean weights (our calculation) for the 423♀ ♀ is 40·7 g., and for the 455 ♂ ♂, 41·3 g-To these figures about 2·5 g. should be added to make them comparable with our weighings at 26 days. Thus the Wistar Institute figures are intermediate between ours for lines T and C, but nearer line C. This applies also to the weights at 75 and 125 days. But our rats are unselected, the measurements applying to the whole colony, while those of Greenman & Duhring refer to a selected and specially cared for group, which are stated to be superior to the general run of the colony. It appears therefore that the Wistar rat in its own home is very similar in weight to our line C. The superiority of line T over the standard Wistar rat is also shown by the maximum weights attained. King obtained two males of over 400 g. (414 g. at 455 days and 437 g. at 485 days). In what appears to be about the same number of rats in generations 25-28 we had seven males in line T of over 400 g. Two of the largest of these, which were kept to enable them to reach maximum size, eventually exceeded 500 g. (503 g. at 350 days and 507 g. at 270 days).

An important question is—does difference in size directly affect rate of learning? If larger animals tend to learn more quickly than smaller ones, this would account for the slightly lower average level of training scores exhibited by line T for so many generatidns, though that would leave unexplained the more recent approximation of scores in the two lines in spite of the maintenance, and even increase, of the greater size of the rats in line T. So far as can be judged from variations in weight within each line, larger rats do not learn more quickly than smaller ones, but if anything the reverse. In the six groups furnished by the two lines in generations 26-28, the mean weight of quick learners (classes 1 and 2) was less than the mean weight of their whole generation in five cases. Not much stress can be laid on this, as the difference is of doubtful statistical significance. So far as the evidence goes, however, it is against the view that the smaller size of the control rats is responsible for their higher average training scores, and is in accord with the impression recorded by McDougall, Crew and ourselves (First Report) that weaker rats tend to make fewer errors than stronger rats.

However, there is little justification for applying to size differences of genetic origin conclusions drawn from the performance of rats of different weights within each line, which are certainly mainly phenotypic.

Since our rats are albinos, this was only revealed by accident when rats from our stock were used for another experiment involving crossing them with piebald rats of the type known as ‘hooded’. This revealed at once that lines T and C were carrying different allelomorphs of the gene concerned.

As the result of the work of Castle & Philips (1914), Castle (1916) and others, it is established that this gene has three allelomorphs : H, which gives a self-coloured rat, dominant to the other two allelomorphs h′ and h ; h′ gives the pattern known as Irish, which in the homozygous condition is a rat fully pigmented except for a white spot, or thin white streak, on the belly. This is dominant to the allelomorph, h, which gives the hooded rat.

In Castle’s well-known selection experiment on the pigmented area in hooded rats (hh) two rats from the same father appeared in his roth generation which exhibited the colour pattern now known to be characteristic of the heterozygous Irish, h′h; this represents a considerably higher grade, that is to say, with a larger pigmented area than any which had previously appeared in Castle’s stock. Experiments showed that these two rats differed from the rest of the stock, not in the modifying genes responsible for the ordinary variations of the extent of the hood in hh rats, but in the gene itself, which had mutated from h to h′.

Curtis and Dunning (1937) also obtained two independent mutations of the gene h to its dominant h′ in unrelated stocks.

There are therefore three records of this mutation. Apparently we have to add a fourth.

The strain of hooded rats used fór crossing with our albino stock varied but little from Castle’s grade o, the typical hooded pattern. The members of the strain actually used for the crosses ranged from grade to grade 0. The colour of the pigmented area is black.

In generation 19, two males of our stock, one from line T and one from line C were mated in succession to the same hooded female. The C male gave a typical hooded F_ly all about grade 0, while the F_t from the T male were all about grade ⁠, which is typical of the heterozygous Irish pattern. This unexpected evidence of a genetic difference between the two lines was followed up in generations 24-26 in which fifteen C rats and thirteen T rats were tested by mating with hooded rats. The fifteen C rats produced 125 F₁ young, ranging from grades to +I. The thirteen T rats produced 136 F₁ young, all between grades +4 and +5. Matings between the F_r rats derived from line T, the colour pattern of which indicated the genetic composition h’h, yielded approximately the expected proportion with the pattern of the homozygous Irish, h′h′.

This constant difference between the rats of the two lines, picked at random from four generations, is sufficient to show that line T is homozygous for h′ and line C for h. Since the difference was first discovered in generation 19, we are not in a position to say definitely whether it arose as a mutation in one line, or whether both allelomorphs were present in the original pair from which both lines owe their origin. The standard Wistar rat is however known to be homozygous for h, and as our stock has had no outside blood introduced into it while it has been in our hands, and we are assured that this applies also to the many generations which elapsed from the time it was obtained from the Wistar Institute and the time it came into our possession, it is highly probable that a mutation from h to h′ occurred in an early generation of our line T.

Since lines T and C carry different allelomorphs of the pattern gene H, it seemed possible that the difference in body size might be an expression of the same gene difference. To test this, F₁ heterozygotes from crosses between the two lines were mated to hooded rats to introduce the pigment factor, and the offspring of these matings were bred together, giving pigmented hh, h′h and h’h′ rats (as well, of course, as albinos of unknown genotype). A total of 156 of the pigmented rats were weighed when 26 days old, showing no significant difference between the three genotypes.

The h and h′ allelomorphs are therefore not responsible for the difference of body size between the two lines. These now differ by at least two genes (for hooded and weight), or more likely by many if, as appears probable, the difference of weight is an expression of many genes.

For our immediate problem the important fact is that since at least generation 19, lines T and C have not been genetically identical for factors which cannot be related to the effects of training. Since our rats are albinos, mutation in colour or pattern genes would not have been detected but for the accident that some of the rats were used for another experiment which involved crossing them with a pigmented form. We have no way of telling what other somatically invisible mutations may have taken place, nor whether these may directly or indirectly influence the rate of learning.

As we have shown, between about generations 12-28 the mean training scores of line T were lower than in line C—almost consistently so when individual generations are compared, and quite consistently for the larger groups of four generations. Later this difference disappeared. In view of the other genetic differences which have developed between the two lines, it seems reasonable to attribute this variation in training scores, maintained over many generations, also to mutations.

Line C is not available for testing parent-offspring correlation, since the parents are not trained. Line T was tested for this correlation by correlating the mean classes of litters with the mean classes of their mid-parents—that is to say, the mean class of the two parents. This was carried out for the fifteen generations 18-32, beginning therefore with the first generation reared in the new cages and on the new diet as described in our Second Report. The offspring furnished by these fifteen generations comprise 121 litters and 696 rats.

The mean class of the mid-parents is 4·79 ± 0·17, and of the litters 4·19 ± 0·13. The parent-offspring correlation is + 0·064 ± 0·091. Therefore no genetic differences in learning rate between members of the line T are disclosed.

To test this further, a comparison was made between the offspring of the quickest and slowest learning parents. Within the fifteen generations there are eight matings in which both parents were quick (class 1 or 2). There are nine matings of slow parents (both class 7 or over). The total number of offspring is fifty-one in both groups. The mean class of offspring from the quick parents is 4·29 ± 0·37 and from the slow parents 4·20 ± 0·41. This test therefore also fails to disclose any genetic differences, even between the quickest and slowest rats within line T.

Crew (1936), on the other hand, found a parent-offspring correlation of +0-3, making it ‘certain that genetic factors are largely concerned in determining the score that a given individual shall make’. It must be noted, however, that, granted the existence of genetic factors affecting rate of learning, Crew’s method of breeding favoured the production of individuals and pedigrees tending to homozygosity for such factors. The total population which yielded the correlation coefficient was composed of six lines, maintained without interbreeding from the second generation. Only two of these lines, however, survived to the end of the experiment (eighteen generations). These two lines provided 586 (line A) and 564 (line B) of the total 1449 rats. Crew states that the average score of line A rats is significantly lower than of line B. Moreover, within each line he practised mating quick learners with quick, and slow with slow.

Our method of random mating does not tend to segregate factors for quick or slow learning into individuals or pedigrees.

Between generations 28 and 33 we made a succession of brother-sister matings, but without regard to the individual scores. If significant genetic differences exist, this should result in splitting the line into.sublines of different average training scores. As no evidence of such differentiation was obtained, we reverted from generation 34 onwards to our old method of mating, which excludes brother-sister matings as far as possible.

The conclusion therefore is that our line T is genetically homogeneous for genes affecting rate of learning, or else that if genetic differences are present, their influence on the training score is so small compared with that of non-genetic factors as to fail to produce a significant parent-offspring correlation.

In his Third Report (1933, in collaboration with Rhine) and his Fourth Report (1938), McDougall introduced another test of the effects of ancestral training. This is the initial preference of the rats for the bright (B) or dim (D) gangway, before any experience of the shock on B. His practice was to give his rats, on the day preceding the beginning of training proper, six runs in the tank with the light alternating between left and right in the usual way, but with no shock on B. He found that the earlier generations of his trained lines, and his batches of control rats, showed a slight preference for B on this day (zero day). In the Third Report he produces striking evidence of a change in this respect. The lines concerned are (1) the principal line, TR, on which the conclusions from his experiment are mainly based; (2) a line selected for slow learning, W C (WH) in which the slowest half of each generation was used for breeding; (3) a line, WC (BH₂) similarly selected for quick learning.

In the 16th generation in both selected lines there was a change from the previous slight average excess of choices of B on zero day to a large excess of choices of D (250 D-140 B in one line, and Í27 D-77 B in the other) (1938, Tables VII, X, XI).

A similar, but very slight, change occurred in generation 32 of the main line (i.e. TR 32); this was much more marked in TR 33. His Fourth Report shows that the altered behaviour was maintained, on the whole, in the later generations of all three lines. McDougall italicizes his opinion that ‘this preference (for D, before experience of the shock on B) has thus become one of the leading evidences which I adduce in support of the Lamarckian interpretation’ (1938, p. 325). Because of the importance that is likely to be attributed to this conclusion, we have made a detailed analysis of the evidence for it. We believe that this analysis demonstrates the invalidity of his reasons for invoking the Lamarckian factor, or indeed any genetic factor at all.

There can be no doubt of the reality of the change from an average preference for B on zero day in the earlier generations to one for D in the later generations (1938, Tables VII, X, XI). The question is—was it due to a genetic change, whether Lamarckian or not, or may it with equal reason be attributed to some external factor? There would be strong grounds for postulating such a factor operating similarly on all three lines if the change in the three lines took place synchronously. If, however, it occurred at different times in the three lines, especially if these times bore some relation to the number of generations during which the lines had been in training, there might be some justification for McDougall’s Lamarckian interpretation. It is a striking fact, however, that when the change appeared one line had been trained for some twelve years, the other two for six.

We are left therefore with the question—did the change appear synchronously in the three lines or not? In discussing, and rejecting, the possibility that the change might be the result of some external factor, McDougall stresses that the change was not synchronous in the three lines. This, however, depends on his location of the change in TR 32 and not in TR 33, and also in generation 16, not in generation 15, of the two selected lines WC (see his dates of training these generations, 1933, p. 230). If, however, we locate the changes in TR 33 and generation 15 of the two selected lines, then the change is as nearly synchronous as different dates of training permit. A decision on this point is essential to McDougall’s argument, and depends on whether the figures for the generations in question are to be considered fluctuations in the earlier series of generations with its average preference for B, or in the later series with its average preference for D.

On these grounds there appears to be no justification for McDougall’s location of the change in TR 32 rather than in TR 33. The zero-day choices in TR 32 (42 rats) were D 137-B 115, or 54·4% D. We can find^* from 1938, Table VII, that the total zero-day choices for the eighteen generations before TR 32 for which records are given (TR 13-31, 345 rats) are D 1005-B 1065, or 48·6% D; for the twelve generations after TR 32 (TR 33-44, 313 rats) the choices are D 1141-B 737, or 60·8% D. The question therefore is: Is the 54·4% D of TR 32 to be considered as a chance fluctuation under conditions giving the earlier 48·6% D, or a chance fluctuation under conditions giving the later 60·8 % D ? McDougall implies that he accepts the latter, for this is necessary if the change to increased preference for D is to be located in that generation. The figure 54-4 % is however almost exactly midway between the earlier 48-6% D and the later 6o-8% D. On this criterion therefore there is no justification for considering it, as McDougall’s view necessitates, a member of the later rather than of the earlier series.

Moreover, if we examine in detail the generations before and after TR 32, we find that that generation falls within the range of variation of both series. Two of the eighteen generations of the earlier series (with twenty-three and fourteen rats respectively) show a greater preference for D than do the forty-two rats of TR 32, and two of the later series (twenty-nine and thirty rats) show a smaller preference.

We conclude therefore that McDougall’s own figures show no justification for classing TR 32 with the later rather than with the earlier series of generations, and therefore for locating the change in that generation rather than in TR 33. TR 32 falls within the range of variation of both series and is intermediate between the average choices of D and B in the two series.

We must now examine McDougall’s justification for locating the change in generation 16, rather than in generation 15, of the two selected lines, for to establish the synchrony of the change in the three lines we must disagree, not only with McDougall’s location of the change in TR 32, but also with his location of it in generation 16 of the two selected lines.

In these two lines, the generations before 15 are common to both, the line WC (WH), selected for slow learning, being split in generation 14 into two sublines ; selection for slow learning was continued in the old WC (WH), and reversed to selection for quick learning in the other, now designated WC (BH₂).

Generation 16 (1938, Tables X, XI) certainly shows a pronounced change from the former average preference for B, giving D 250-B 140 in the one line, and D 127-B 77 in the other. The preceding generation 15, however, also shows, although less markedly, a decline in the former preference for B, the two lines together giving an equality of choices (D 168-B 168).

In deciding whether the change in these lines is to be located’in generation 16 (as McDougall’s argument requires) or in generation 15 (which would make it synchronous with TR 33) we have again to decide whether generation 15 falls most naturally within the range of variation of the earlier generations before the change took place, or of the nine later generations after it had taken place. Examination of these later generations, 16-24, in the two lines shows that generation 15 falls within the range of variation of these later generations, for in both lines three out of the nine generations show a considerable preference for B, and therefore a much greater deviation from the average preference for B exhibited by the group of nine generations as a whole than does generation 15 with its equality of choices.

To sum up our discussion of the grounds on which McDougall denies synchrony to the change in the three lines, on which denial he bases his exclusion of an external factor and attributes it to the Lamarckian factor: there is no doubt that from TR 33 and generation 16 of the two selected lines, onwards, all three lines show a marked average preference for D, compared with an average preference for B in the earlier generations. If we accept McDougall’s contention that the change began in TR 32, WC (WH) 16 and WC (BH₂) 16, then it began nearly a year earlier in TR than in the two selected lines, thus excluding its explanation by the action of an external factor operating in common on the three lines. But if we locate the change in TR 33 and in generation 15 of the two selected lines, as we contend it is at least as reasonable to do, then all three lines show it in the first generation whose training began later than September 1932.

In further discussion, and rejection, of the possibility that the change was due to an external-factor, McDougall refers to the fact that in 1932, in which this new, decided preference for D began to be manifested, it became necessary to build new tanks to replace the old pair which had been in use for some ten years (1938, p. 325). The new tanks proved to offer to the rats a less difficult task than did the old ones, less difficult by approximately one third. The illumination of the gangways and of the walls of the passage leading to them was rather brighter than in the old tanks, and the curved end of the tank (facing the’rats at their starting point) was painted in vertical stripes of black and white (1933, p. 218). Unfortunately we are not told which of the generations which began their training in 1932 was the first to be trained in the new tanks. McDougall, however, presents evidence that the change in the zero-day preferences for D or B, which took place about the time of the introduction of the new tanks ‘cannot be attributed to the change from the old to the new tanks, save in some very slight degree’ (1938, p. 327). The evidence consists of several batches of control rats with untrained ancestry, comprising a total of 106 trained in the old, and 103 in the new, tanks. The former gave zero-day choices D 302-B 334 (47-5% D), and the latter D 313-B 305, or 50-6% D.

The fact that these control rats in the new tanks show only such a slight preference for D is taken by McDougall as proof that the change from the old to the new tanks cannot be responsible for the much larger increase in preference for D exhibited by the three trained lines in the new tanks. But again this conclusion depends upon the exclusion of the small preference for D shown by the control rats in the new tanks as a chance fluctuation from the larger preference for D to be expected in the new tanks if these were responsible for the change in the trained lines. But, as we have seen, six out of the eighteen generations of the two selected lines in the new tanks show not only a smaller preference for D than these controls, but actually a preference for B. For the third time therefore a group of rats whose performance McDougall relies on for denying the operation of an external factor (this time, specifically, the new tanks) is found to fall within the range of variation of the groups from which it is necessary to exclude it if McDougall’s contention is to be substantiated.

We may also draw attention to another fact adverse to McDougall’s interpretation of the cause of the change to a preference for D after transference to the new tanks. If, as he maintains, it is the Lamarckian effect of previous generations of training, it might surely be expected that with continuance of the training later generations would show a further increase of preference for D. This can be investigated from McDougall’s records. In his Fourth Report he gives the figures for thirteen generations of TR, and nine of each of the two selected lines, after the change from an average preference for B, to an average preference for D. Leaving out the middle generation of each line to get equal halves, we can compare the first six with the last six generations of TR, and the first four with the last four generations in each of the two selected lines. The first halves, consisting of generations TR 32-37, WC (WH) 16-19 and WC (BH₂) 16-19, a total of 447 rats, show 56-5% choices of D. The second halves, TR 39-44, WC (WH) 21-24 and WC (BH₂) 21-24 (371 rats) show 57^-o% choices of D. The combined lines show therefore no significant increase of preference for D with further accumulation of trained ancestry.

Summing up the whole of the foregoing analysis of the evidence, we cannot accept McDougall’s conclusion that the change from an average zero-day preference for B in the earlier generations to an average preference for D in the later generations, was a Lamarckian effect ; or, indeed, that it was a genetic change of any kind. The evidence is equally compatible with, or even more favourable to, the conclusion that it made its appearance in the first generations of the three trained lines which began their training later than September 1932. This points to the operation of an external factor, which continued to operate with equal force throughout the later generations, rather than to a Lamarckian factor which produced its effect suddenly and almost simultaneously on the three trained lines, in spite of the fact that one of them had been in training for about twice as many years as the other two, and which failed to increase its effect after many more generations of training. While it is not necessary for our argument to identify the external factor responsible for the change, we do not think that McDougall has produced any valid grounds for his contention that the factor in question could not be the transference from the old to the new tanks.

We cannot present figures of the initial preferences for D or B quite comparable to McDougall’s, owing to the fact that on the first day (McDougall’s zero day, before training proper begins) we, like McDougall, give our rats six runs in the tank to give them a preliminary acquaintance with it, but on this day we use neither the light nor shock, whereas McDougall used the alternating light but no shock. On zero day, therefore, McDougall’s rats could take the bright gangway without receiving the shock. Our rats experience the shock on the first occasion (usually on the first day of training proper) that they choose that exit. As, however, only a very few of the exceptionally quick learners are avoiding B on the later runs of the first day as a result of experience of the shock on B on an earlier run on that day, the ratio of the choices of B and D on our first day of training are practically equivalent to McDougall’s figures for zero day.

Table 5 gives the ratios of the choices of B and D in the four runs of the first day of training proper, dividing the total experiment into four successive quarters. This table shows a general parallelism between the changes in per cent choices of D on the first day of training, and the changes in training scores as shown in Table 3, column Mean class. For (1) in both, the changes run parallel in the two lines, T and C, (2) in the early part of the experiment the choices of D are low, followed by a long period when the choices of D are much higher; in the last quarter of the experiment there is a return to the original low percentage of choices of D. It will be noted again that if we had had no control line, and had stopped, the experiment at about generation 27, the increasing preference for D on the first day of training could have been interpreted as a Lamarckian effect.

It will be noted that McDougall’s figures and our own both show an inverse correlation between initial choices of D and training score. Generations with a higher initial choice of D learn with fewer errors than generations showing a lower initial choice of D. This is not a spurious correlation, due to the fact that the number of runs to B (errors) on the first day is included in the final score, for the extreme differences in the per cent choices of D in Table 5 represent much less than a difference of one choice of D in the four runs ; this is a negligible contribution to the differences in mean training scores of different generations. This suggests the influence of a common factor on the initial preference for B or D, and the total number of errors made before learning.

This experiment, to test McDougall’s conclusion that the effects of training are inherited, has now been carried on for thirty-six generations, involving the training of 2827 rats. The present position of the problem raised by McDougall may be summarized as follows:

Neither our own experiment, nor that of Crew, shows any evidence of increasing facility in learning attributable to trained ancestry. McDougall’s claim that the progressive decline in the number of errors which he found in successive generations of trained rats is an example of Lamarckian inheritance cannot be maintained in face of the facts (a) that he did not keep a control line, (b) that we have found a progressive decline in our trained line similar to McDougall’s, but this was paralleled by the control line ; moreover, after about twenty-eight generations, the number of errors progressively increased again in both lines.

McDougall’s further argument from the change from a zero-day preference for the bright gangway in earlier generations to a preference for the dim gangway in later generations is invalid; it is shown, from his own figures, to be capable of a different explanation.

The discovery of genetic differences in colour pattern and body size between our trained and control lines, presumably due to mutations, emphasizes the difficulty of interpreting genetic differences in facility of learning, even if they should occur, as’ due to the Lamarckian factor.

The experiment is being continued.