Rivista di Biologia - Biology Forum 85 (3/4), 1992, 455-460
by Rupert Sheldrake

Rose's predictions about the outcome of this experiment were refuted by the empirical data. His aggressive tone and extravagant rhetoric conceal this simple fact. I will not attempt to answer his polemic, ranging from Nietzsche to ley-lines, but simply start by looking again at his predictions about the chicks: "No secular trends apparent; latencies to peck the illuminated bead after ten weeks are no different from those on week I, and the differencies between latencies for illuminated and chrome beads, if they occur, are also unchanged". In fact secular trends were very apparent, latencies to peck the illuminated bead after ten weeks were very different from those on week I, and the differences between latencies for illuminated and chrome beads were not unchanged. Rose and I discussed various interpretations of the data over a period of eighteen months. At the outset, he seemed certain that the hypothesis of formative causation would be disconfirmed. He had already publicly denounced it in the strongest terms. He appeared to have no doubt that when tested in his own laboratory, under his own supervision, in my absence, by an experimenter working blind, the data would reveal no trace whatever of morphic resonance. But it soon became clear that there had been an effect of the kind predicted by the hypothesis of for- mative causation. (I discuss below Rose's alternative interpretation of this effect in terms of "floors" and "ceilings"). After lengthy delays, Rose withdrew from our agreement to write a joint paper, and no longer wanted to publish the results. This is the background to my present paper Formative Causation and Rose's response to it.

Before discussing our interpretations, I briefly review what we have in common. A large decline in latency with both test and control stimuli occurred in the first week: in other words the chicks pecked at the stimuli much sooner. Rose and I agree that this effect was largely, if not entirely, a result of Ms Harrison learning how to test the chicks. This was the largest single effect in the whole experiment, and inevitably dominates the statistics in any overall analysis. We agreed to analyse separately the results from the second week onwards to see more clearly what happened in the rest of the experiment. We agree that the continued decline in latency in the control chicks (not predicted by either of us) is best explained as a result of Ms Harrison's increasing familiarity with the chicks and the test procedures. Both of us failed to forsee these obvious experimenter effects, and therefore we were both equally wrong in our prediction that latencies in the control group would show no secular trend over the course of the experiment.

Our essential disagreement concerns the trend in the differences between the latencies with yellow light (test) and chrome bead (control). Rose wrongly predicted that there would be no trend in these differences. There was in fact a trend (my Figs 4 and 6), consistent with a morphic resonance effect. As the experiment progressed, there was a cumulative increase in the number of previous chicks that had developed an aversion to pecking the yellow light because they had been made sick, and hence by morphic resonance subsequent chicks showed an increasing tendency to avoid it, even though they themselves had not been made sick. This increasing aversion to pecking the yellow light countervailed the tendency to peck sooner as a result of the experimenter's practice.

Both Rose and Bateson offer an alternative explanation of this same trend in terms of "floors" and "ceilings". If their arguments are confusing to the reader, this is not surprising-, Rose and Bateson each use the words "floor" and "ceiling" in the opposite sense. They claim that soon after the experiment began, the latency scores for the yellow light reached a minimum, which Rose calls a "floor" and Bateson calls "an obvious ceiling effect", and then stayed at this limit, unable to fall lower in spite of the experimenter's continued improvement. Meanwhile the latency scores for the chrome bead, which started off higher, continued to decline as a result of the experimenter's practice. This, they claim, explains the difference over time between the latencies for the test and control stimuli.

I acknowledge that "floor" and "ceiling" effects could arise in experiments of this type, and preclude the drawing of strong conclusions. In future research it will be important to use a design that can rule out this objection conclusively, for example by repeating the present experiment using control and test stimuli with similar mean latencies, starting off neither too high nor too low. (The preliminary research for such an experiment, comparing the latencies with different stimuli, would also enable the experimenter to get used to testing the chicks, and hence reduce the experimenter practice effect within the main experiment).

But can "floor" effects in fact explain the data in the present experiment? An inspection of the actual data (my Fig. 3; cf Rose's Fig. 2a shows that with the yellow light the supposed minimum latency was reached after the first 6-9 days; but far from remaining at this "t\oor" the latencies thereafter increased up to around the 30th day.

The "floor" explanation is even less convincing when considering the other relevant set of data, that obtained by testing control chicks after they had received an injection of saline solution. Here there was a clear tendency for latencies with the yellow light to increase while those for the chrome bead declined (my Fig. 5), with no sign of "floor" or "ceiling effects. Rose and Bateson simply ignore these results.

I should also point out that there is a serious systematic error in the data used by Rose, as shown in his Fig. 2a and 2b. These data exclude all the chicks that failed to peck the stimulus during the 30s training period (recorded as having the maximum latency of 30s). The data on these maximum-latency chicks had been recorded separately, and were mistakenly omitted from the data-file entered into the computer. brought this mistake to Rose's attention at an early stage in our discussions, and supplied him with the corrected data. But he is still using the erroneous data-file. Hence his statistical analysis is wrong, and so is his conclusion: "There are thus no significant differences between the secular trends in either yellow or chrome beads, even with the 'correction' of omitting the first days of the experiment. Thus the crucial prediction made by the hypothesis, which Sheldrake agreed before the experiment began, is also disconfirmed". Repeating Rose's analysis with the right data shows that the secular trends are in fact significantly different at the p=0.01 level (using the Williams test, as described by HOWELL (1987). The crucial prediction was confirmed, not disconfirmed.

I now turn to other points raised by Rose in his response:

  1. Rose shows in his Fig. la and lb data for the behaviour of chicks that have been made averse to the yellow light by an injection of lithium chloride. There is no problem here. We agree that a strong aversion was in fact established.
  2. Rose and I agreed from the outset that the crucial measurements were of latencies to peck, rather than the number of pecks the chicks made after they first pecked and before the end of the 30s test period. Once the chicks have overcome any initial aversion to pecking the stimulus, the subsequent number of pecks depends on several factors, including the remaining time available in the test period (chicks that first peck at 29s obviously have less scope for further pecks than those that peck at Is) as well as the general level of activity of the chicks, which is influenced by the weather during incubation, in particular the atmospheric pressure, as BATESON (1974) has shown. (Variations in general activity probably also affect the latency, but may well affect peck nember to a greater extent). Rose introduced peck numbers in his Fig. lb and 2b but failed to point out how unreliable they may be as an index of aversion. In any case, the data in his Fig. 2b and the statistical analysis based upon them are erroneous for the reasons given above.
  3. Rose quoted Bateson's remarks to add authority to his own position. But Rose and Bateson differ considerably in their approach. Bateson employed a method of statistical analysis involving the median latencies in 7-day blocks. Bafore we received Bateson's comments, Rose and I agreed that in analysing this experiment, means were in general more informative than medians. We also agreed that the linear regressions provided a simple and appropriate way of analysing secular trends. Bateson's method is much blunter. Instead of looking for overall trends, it involves the comparison of median latencies for the control and test stimuli in each separate 7-day block.
    Despite quoting Bateson's remarks approvingly. Rose still uses means (albeit incorrect ones) in his analysis of the data (his Figs 2a and 2b) and analyses the trends by means of regressions. Thus in terms of statistical methodology, Rose and I are in closer agreement with each other than we are with Bateson. Indeed, I think one of the best ways of ana- lysing the data is to use Rose's own method, routinely followed in his own laboratory, namely to use as a measure of aversion the proportion of the chicks in any given sample that do not peck at the stimulus with- in the first 10s. This is the method I used in the analysis of data shown in my Figs 2, 3 and 4.
  4. If morphic resonance takes place, then the response of chicks to stimuli routinely used in similar experiments, such as chrome beads and coloured LEDs, should be influenced by what happened in previous ex- periments. Such stimuli would not be tabulae rasae. In order to mi- nimize background morphic resonance, and also to reduce any possible interference from concurrent aversion experiments with chicks, Rose and I agreed to use a yellow LED, a stimulus which has not, so far as we know, been used in previous research with chicks. Red and green LEDs are commonly employed. The chromium-plated bead used as the control stimulus has been used in previous aversion experiments, sometimes as a test stimulus and sometimes as a control.

In general, as Bateson points out, the chicks pecked sooner at the yellow than at the chrome bead. He suggests that this was because "the yellow light caught the attention of the chicks to a greater extent than the chrome bead". He does not explain why this should be so. Is the yellow light more attractive because it is a light, yellow, or both? The first of these possibilities can be ruled out, because LEDs in other colours are relatively unattractive. In a subsidiary experiment designed by Rose and myself and carried out by Ms Harrison, we compared the mean latencies to peck at the chrome bead, the yellow LED, and a green LED of the type regularly used by Rose's group in aversion experiments (e.g. BARBER et al., 1989). The mean latencies were 4.1 (n=27) for the yellow LED, 8.9s (n=56) for the chrome bead and 18.0s (n=29) for the green LED. Maybe the chicks' strong aversion to pecking the green LED was, at least in part, a result of morphic resonance from chicks exposed to green LEDs in previous aversion experiments.

In this context, Bateson's final remark does not make much sense to me: "the evidence that, overall, the chicks pecked more quickly at the yellow bead than at the chrome bead runs counter to the prediction from the morphic resonance hypothesis, and [Sheldrake's] analysis obscures this fact". The yellow light was chosen because it was a fresh stimulus, not associated with a background morphic resonance from averse chicks in previous experiments, unlike the chrome bead. The fact that the yellow light was more attractive from the outset does not run counter to any prediction from the morphic resonance hypothesis; if anything, it tends to support it. But perhaps Bateson means that the yellow bead, in spite of being more attractive to start with, should by morphic resonance have rapidly become less attractive than the chrome bead; if so, he is ignoring the background morphic resonance from previous chicks averse to the chrome bead. Or perhaps Bateson means something else. Who is being obscure?

On the basis of confused reasoning, erroneous data and selective use of evidence, Rose announces that the hypothesis of formative causation is "disconfirmed". This categorical statement is a good example of his polemical style, familiar to me from our debate over the nature of memory in The Guardian, in the course of which he seriously misrepresented the experimental evidence. Rose has had many years of experience in the realm of political controversy. His technique is to try and endow his own belief with a tone of objective authority. At the same time he tries to discredit opposing opinions by playing on prejudices. In the present case, he gratuitously attempts to associate the hypothesis of formative causation with creationism, pseudoscience, parapsychology, crop circles, deep ecology, homeopathy, anti-rationalism and whatever else seems likely to arouse negative responses in readers who share his general beliefs. I imagine Rose would disapprove of such tactics on moral grounds if they were used by his political opponents. The results of this experiment do not disconfirm the hypothesis of formative causation, as Rose claims. They are consistent with it.

London, September 1992

For References and Figures, see Sheldrake's first article An Experimental Test of the Hypothesis of Formative Causation