An excerpt from Chapter 11 of A New Science of Life

In mechanistic biology, a sharp distinction is drawn between innate and learned behaviour: the former is assumed to be 'genetically programmed' or 'coded' in the DNA, while the latter is supposed to result from physical and chemical changes in the nervous system. There is no conceivable way in which such changes could specifically modify the DNA, as the Lamarckian theory would require; it is therefore considered impossible for learned behaviour acquired by an animal to be inherited by its offspring (excluding, of course, 'cultural inheritance', whereby the offspring learn patterns of behaviour from their parents or other adults).

By contrast, according to the hypothesis of formative causation, there is no difference in kind between innate and learned behaviour: both depend on motor fields given by morphic resonance (Section 10. 1). This hypothesis therefore admits a possible transmission of learned behaviour from one animal to another, and leads to testable predictions which differ not only from those of the orthodox theory of inheritance, but also from those of the Lamarckian theory, and from inheritance through epigenetic modifications of gene expression.

Consider the following experiment. Animals of an inbred strain are placed under conditions in which they learn to respond to a given stimulus in a characteristic way. They are then made to repeat this pattern of behaviour many times. Ex hypothesi, the new behavioural field will be reinforced by morphic resonance, which will not only cause the behaviour of the trained animals to become increasingly habitual, but will also affect, although less specifically, any similar animal exposed to a similar stimulus: the larger the number of animals in the past that have learned the task, the easier it should be for subsequent similar animals to learn it. Therefore in an experiment of this type it should be possible to observe a progressive increase in the rate of learning not only in animals descended from trained ancestors, but also in genetically similar animals descended from untrained ancestors. This prediction differs from that of the Lamarckian theory, according to which only the descendants of trained animals should learn quicker. And on the conventional theory, there should be no increase in the rate of learning of the descendants of untrained or trained animals.

To summarize: an increased rate of learning in successive generations of both trained and untrained lines would support the hypothesis of formative causation; an increase only in trained lines, the Lamarckian theory; and an increase in neither, the orthodox theory.

Tests of this type have in fact already been performed. The results support the hypothesis of formative causation.

The original experiment was started by William McDougall at Harvard in 1920, in the hope of providing a thorough test of the possibility of Lamarckian inheritance. The experimental animals were white rats, of the Wistar strain, that had been carefully inbred under laboratory conditions for many generations. Their task was to learn to escape from a specially constructed tank of water by swimming to one of two gangways that led out of the water. The 'wrong' gangway was brightly illuminated, while the 'right' gangway was not. If the rat left by the illuminated gangway it received an electric shock. The two gangways were illuminated alternately, one on one occasion, the other on the next. The number of errors made by a rat before it learned to leave the tank by the non-illuminated gangway gave a measure of its rate of learning:

Some of the rats required as many as 330 immersions, involving approximately half that number of shocks, before they learnt to avoid the bright gangway. The process of learning was in all cases one which suddenly reached a critical point. For a long time the animal would show clear evidence of aversion for the bright gangway, frequently hesitating before it, turning back from it, or taking it with a desperate rush; but, not having grasped the simple relation of constant correlation between bright light and shock, he would continue to take the bright route as often or nearly as often as the other. Then, at last, would come a point in his training at which he would, if he found himself facing the bright light, definitely and decisively turn about, seek the other passage, and quietly climb out by the dim gangway. After attaining this point, no animal made the error of again taking the bright gangway, or only in very rare instances.[i]

In each generation, the rats from which the next generation were to be bred were selected at random before their rate of learning was measured, although mating took place only after they were tested. This procedure was adopted to avoid any possibility of conscious or unconscious selection in favour of quicker-learning rats.

This experiment was continued for 32 generations and took 15 years to complete. In accordance with the Lamarckian theory, there was a marked tendency for rats in successive generations to learn more quickly. This is indicated by the average number of errors made by rats in the first eight generations, which was over 56, compared with 41, 29 and 20 in the second, third and fourth groups of eight generations, respectively. [ii] The difference was apparent not only in the quantitative results, but also in the actual behaviour of the rats, which became more cautious and tentative in the later generations. [iii]

McDougall anticipated the criticism that in spite of his random selection of parents in each generation, some sort of selection in favour of quicker-learning rats could nevertheless have crept in. In order to test this possibility, he started a new experiment, with a different batch of rats, in which parents were indeed selected on the basis of their learning score. In one series, only quick learners were bred from in each generation, and in the other series only slow learners. As expected, the progeny of the quick learners tended to learn relatively quickly, while the progeny of the slow learners learned relatively slowly. However, even in the latter series, the performance of the later generations improved very markedly, in spite of repeated selection in favour of slow learning (Fig. 29).


Figure 29 The average number of errors in successive generations of rats selected in each generation for slowness of learning. (Data from McDougall, 1938).

These experiments were done carefully, and critics were unable to dismiss the results on the ground of flaws in technique. But they did draw attention to a weakness in the experimental design: McDougall had failed to test systematically the change in the rate of learning of rats whose parents had not been trained.

One of these critics, F.A.E. Crew, of Edinburgh University, repeated McDougall's experiment with rats derived from the same inbred strain, using a tank of similar design. He included a parallel line of 'untrained' rats, some of which were tested in each generation for their rate of learning, while others, which were not tested, served as the parents of the next. Over the 18 generations of this experiment, Crew found no systematic change in the rate of learning either in the trained or in the untrained line. [iv] At first, this seemed to cast serious doubt on McDougall's findings. However, Crew's results were not directly comparable in three important respects. First, the rats found it much easier to learn the task in his experiment than in the earlier generations of McDougall's. So pronounced was this effect that a considerable number of rats in both trained and untrained lines 'learned' the task immediately without receiving a single shock! The average scores of Crew's rats right from the beginning were similar to those of McDougall's after more than 30 generations of training. Neither Crew nor McDougall was able to provide a satisfactory explanation of this discrepancy. But, as McDougall pointed out, since the purpose of the investigation was to bring to light any effect of training on subsequent generations, an experiment in which some rats received no training at all and many others received very little would not be qualified to demonstrate this effect. [v] Second, Crew's results showed large and apparently random fluctuations from generation to generation, far larger than the fluctuations in McDougall's results, which could well have obscured any tendency to improve in the scores of later generations. Third, Crew adopted a policy of very intensive inbreeding, crossing only brothers with their sisters in each generation. He had not expected this to have adverse effects, since the rats came from an inbred stock to start with.

Yet the history of my stock reads like an experiment in inbreeding. There is a broad base of family lines and a narrow apex of two remaining lines. The reproductive rate falls and line after line becomes extinct. [vi]

Even in the surviving lines, a considerable number of animals were born with such extreme abnormalities that they had to be discarded. The harmful effects of this severe inbreeding could well have masked any tendency for the rate of learning to improve. Altogether, these defects in Crew's experiment mean that the results can only be regarded as inconclusive; and in fact he himself was of the opinion that the question remained open. [vii]

Fortunately, this was not the end of the story. W. E. Agar and his colleagues at Melbourne University carried out the experiment again, using methods that did not suffer from the disadvantages of Crew's. Over a period of 20 years, they measured the rates of learning of trained and untrained lines for 50 successive generations. In agreement with McDougall, they found that there was a marked tendency for rats of the trained line to learn more quickly in subsequent generations. But exactly the same tendency was also found in the untrained line. [viii]

It might be wondered why McDougall did not also observe a similar effect in his own untrained lines. The answer is that he did. Although he tested control rats from the original untrained stock only occasionally, he noticed 'the disturbing fact that the groups of controls derived from this stock in the years 1926, 1927, 1930 and1932 show a diminution in the average number of errors from 1927 to 1932'. He thought this result was probably fortuitous, but added:

It is just possible that the falling off in the average number of errors from 1927 to 1932 represents a real change of constitution of the whole stock, an improvement of it (with respect to this particular faculty) whose nature I am unable to suggest. [ix]

With the publication of the final report by Agar's group in 1954 the prolonged controversy over 'McDougall's Lamarckian Experiment' came to an end. The similar improvement in both trained and untrained lines ruled out a Lamarckian interpretation. McDougall's conclusion was refuted, and that seemed to be the end of the matter. On the other hand, his results were confirmed.

These results seemed completely inexplicable; they made no sense in terms of any current ideas, and they were never followed up. But they make very good sense in the light of the hypothesis of formative causation. Of course they cannot in themselves prove the hypothesis; it is always possible to suggest other explanations, for example that the successive generations of rats became increasingly intelligent for an unknown reason unconnected with their training. [x]

In future experiments, the most unambiguous way of testing for the effects of morphic resonance would be to cause large numbers of rats (or any other animals) to learn a new task in one location; and then see if there was an increase in the rate at which similar rats learned to carry out the same task at another location hundreds of miles away. The initial rate of learning at both locations should be more or less the same. Then, according to the hypothesis of formative causation, the rate of learning should increase progressively at the location when large numbers are trained; and a similar increase should also be detectable in the rats at the second location, even though very few rats had been trained there. Obviously, precautions would need to be taken to avoid any possible conscious or unconscious bias on the part of the experimenters. One way would be for experimenters at the second location to test the rate of learning of rats in several different tasks, at regular intervals, say monthly. Then at the first location, the particular task in which thousands of rats would be trained would be chosen at random from this set. Moreover, the time at which the training began would also be selected at random; it might, for example, be four months after the regular tests began at the second location. The experimenters at the second location would not be told either which task had been selected, or when the training had begun at the first location. If, under these conditions, a marked increase in the rate of learning in the selected task were detected at the second location after the training had begun at the first, then this result would provide strong evidence in favour of the hypothesis of formative causation.

An effect of this type might well have occurred when Crew and Agar's group repeated McDougall's work. In both cases, their rats started off learning the task considerably quicker than McDougall's when he first began his experiment. [xi]

If the experiment proposed above were actually performed, and if it gave positive results, it would not be fully reproducible by its very nature: in attempts to repeat it, the rats would be influenced by morphic resonance from the rats in the original experiment. To demonstrate the same effect again and again, it would be necessary to change either the task or the species used in each experiment.


AGAR, W.E., DRUMMOND, F.H., AND TIEGS, O.W. (1942) Second report on a test of McDougall's Lamarckian experiment on the training of rats. Journal of Experimental Biology 19, 158-67.

AGAR, W.E., DRUMMOND, F.H., TIEGS, O.W., and GUNSON, M.M. (1954) Fourth (final) report on a test of McDougall's Lamarckian experiment on the training of rats. Journal of Experimental Biology 31, 307-21.

CREW, F.A.E. (1936) A repetition of McDougall's Lamarckian experiment. Journal of Genetics 33, 61-101.

McDOUGALL, W. (1927) An experiment for the testing of the hypothesis of Larmarck. British Journal of Psychology 17, 267-304.

McDOUGALL, W. (1930) Second report on a Lamarckian experiment. British Journal of Psychology 20, 201-18.

McDOUGALL, W. (1938) Fourth report on a Lamarckian experiment. British Journal of Psychology 28, 321-45.

[i] McDougall (1927), p. 282. Return

[ii] McDougall (1938). Return

[iii] McDougall (1930). Return

[iv] Crew (1936). Return

[v] McDougall (1938). Return

[vi] Crew (1936), p. 75. Return

[vii] Tinbergen (1951), p. 201. Return

[viii] Agar, Drummond, Tiegs and Gunson (1954). Return

[ix] Rhine and McDougall (1933), p. 223. Return

[x] A number of possible explanations were suggested at the time these experiments were being carried out; they are discussed in McDougall's papers, to which the interested reader should refer. None of these explanations turned out to be plausible on closer examination. Agar et al. (1954) noticed that fluctuations in the rates of learning were associated with changes, extending over several generations, in the health and vigour of the rats. McDougall had already noted a similar effect. A statistical analysis showed that there was indeed a low but significant (at the 1% level of probability) correlation between vigour (measured in terms of fertility) and learning rates in the 'trained' line, but not in the 'untrained' line. However, if only the first forty generations were considered, the coefficients of correlation were somewhat higher: 0.40 in the 'trained' line, and 0.42 in the 'untrained'. But while this correlation may help to account for the fluctuations in the results, it cannot plausibly explain the overall trend. According to standard statistical theory, the proportion of the variation 'explained' by a correlated variable is given by the square of the correlation coefficient, in this case (0.4)2 = 0.16. In other words, variations in vigour account for only 16% of the changes in the rate of learning. Return

[xi] McDougall estimated that the average number of errors in his first generation was over 165. In Crew's experiment this figure was 24, and in Agar's, 72; see the discussions in Crew (1936), and in Agar et al. (1942). If Agar's group had used rats of identical parentage and followed the same procedures as Crew, their initial score might have been expected to be even lower than his. However, owing to the different parentage of their rats, and to differences in their testing procedure, the results are not fully comparable. Nevertheless the greater facility of learning in these later experiments is suggestive. Return