My purpose in this essay is to address some claims Massimo has made over the years about parapsychology (the scientific discipline that studies claims of extrasensory perception, or ESP, psychokinesis, and survival of consciousness after bodily death), and to show why I think the scientific evidence for ESP (e.g. telepathy, clairvoyance, and precognition) is more plentiful than he seems to believe.
On many past occasions, I have heard Massimo publicly claim that ESP has been refuted, such as in a Skeptiko podcast interview last year in which he said “... research on the paranormal has been done for almost a century. We have done plenty of experiments, say on telepathy or clairvoyance or things like that, and we know it doesn’t work.” And in his recent book, Nonsense on Stilts: How to Tell Science from Bunk, Massimo even implies that parapsychology is a pseudoscience on par with astrology.
It would be reasonable to expect, especially from someone as learned as Massimo, that these bold claims about research on telepathy and clairvoyance, and the status of parapsychology as a discipline, were derived from a thorough assessment of the parapsychology literature (a literature which includes informed skeptical criticisms of parapsychology experiments). However, in my assessment of the parapsychology literature, I have been unable to find an evidenced basis for Massimo’s claims. Not only that, my study of the literature has turned up evidence that strongly supports a conclusion contrary to Massimo’s. Here’s why.
In parapsychology, the three research paradigms considered to provide some of the best evidence for ESP are (a) the Maimonides and subsequent dream telepathy/clairvoyance/precognition experiments, (b) the SRI, SAIC, and PEAR remote viewing experiments, and (c) the Ganzfeld experiments. Here I’ll limit myself to discussing (c) only, and refer the interested reader to this recent anthology which overviews the evidence from (a) and (b).
Within parapsychology, the Ganzfeld experiments have probably been the most widely used to test for the possibility of telepathy, clairvoyance, and to some extent precognition. For the unfamiliar reader, a concise account of the Ganzfeld procedure can be read here. The main point I want to make about the Ganzfeld experiments is that, since 1985, there have been 8 independent, published meta-analyses of Ganzfeld experiments; and with the exception of the 1999 meta-analysis by Julie Milton and Richard Wiseman, which was shown by statistician Jessica Utts and acknowledged by Wiseman (personal correspondence, July 2011) to have used a flawed estimate of the overall effect size and p-value of the combined results, all of them have shown statistically highly significant effects with a replication rate well above what’s expected by chance. The literature also shows rather convincingly, in my view, that the leading Ganzfeld critic, Ray Hyman, has been unable to account for these highly significant effects by prosaic means like publication bias, optional stopping, inadequate randomization of targets, sensory leakage, cheating, decline effect, etc. On this last point, I recommend reading Bem and Honorton’s 1994 paper, Bem’s reply to Hyman in 1994, and Storm and co.’s reply to Hyman in 2010.
As an example of the strength of the statistical evidence, let’s look at the most recent Ganzfeld meta-analysis by parapsychologist Patrizio Tressoldi, who applies a frequentist and Bayesian statistical analysis to 108 Ganzfeld experiments from 1974–2008. All these experiments were screened for adequate methodological quality and have an overall hit rate of 31.5% in 4,196 trials, instead of the 25% hit rate expected by chance. Moreover, using the conservative file-drawer estimate of Darlington/Hayes, the lower bound on the number of unreported experiments needed to nullify this overall hit rate is 357, which is considered implausible by Darlington/Hayes’ criterion.
For the frequentist analysis, Tressoldi applied two standard meta-analytic models, namely, a ‘fixed-effects’ model (which assumes a constant true effect size across all experiments) and a ‘random-effects’ model (which assumes a variable true effect size across all experiments). Whereas a standard deviation from the mean of only ~1.6 is needed for the results of a meta-analysis to achieve statistical significance, the fixed-effects model yields an overall effect that’s significant by more than 19 standard deviations from the mean effect of zero, while the random-effects model yields an overall effect more than 6 standard deviations from the mean. The corresponding odds against chance for the fixed-effects model is off the charts, and for the more conservative random-effects model is greater than a billion to 1.
For the Bayesian analysis (which I know Massimo believes is more reliable and valid than the classical approach), Tressoldi follows Rouder and co. in considering two hypotheses. The first is the null hypothesis that the true effect size is zero for all experiments, and the second is the ESP hypothesis that the true effect size is constant and positive across all experiments. He then finds that the ratio of the prior probability of the ESP hypothesis to the prior probability of the null hypothesis, each conditioned on the combined Ganzfeld data, yields a ‘Bayes factor’ of 18,886,051 (or the number of times the latter probability divides into the former probability). So, for a skeptical person who gives prior odds of, say, 1,000,000:1 against ESP, they should update their beliefs by a factor of 18,886,051 in favor of ESP. In other words, if we divide 1,000,000:1 by 18,886,051 we obtain posterior odds of about 0.053:1, or equivalently, 19:1 in favor of ESP. Interestingly, according to Utts, Ray Hyman told her personally that he would put prior odds of about 1,000:1 against ESP being real. For the Ganzfeld Bayes factor calculated by Tressoldi, this would mean that Hyman’s posterior odds should be 18,868:1 in favor of ESP.
And Tressoldi’s Bayesian analysis is not the only one. In the Ganzfeld meta-analysis by Utts and co., a Bayesian approach is also used. Their approach differs from Tressoldi’s in that they assume the true ESP effect size to be variable across experiments, which is the more common assumption in Bayesian statistics. They consider three priors labeled “psi-skeptic,” “open-minded,” and “psi-believer,” each corresponding to a guess about the most likely median Ganzfeld hit-rate. Then they examine the extent to which the combined empirical results of 56 procedurally standard Ganzfeld experiments shift each prior median hit rate to posterior median hit rates above chance. What they find is that the shift depends strongly on how wide one’s subjectively decided uncertainty is around each prior median.
In sum, I don’t see how to escape the conclusion that a classical statistical analysis of the Ganzfeld data gives strong evidence for ESP, judging by the standards of evidence commonly accepted by the social and behavioral sciences. And using Bayesian analysis, we’ve seen that one approach shows overwhelmingly strong evidence for ESP, while another shows that the strength of the evidence depends strongly on the uncertainty of one’s prior belief about the possibility of ESP.
Given this evidence from the Ganzfeld alone, it is difficult for me to see on what basis Massimo claims that “... we have done plenty of experiments, say on telepathy or clairvoyance or things like that, and we know it doesn’t work.” And to the best of my knowledge, he has never tried to justify his assertion.
Correspondingly, I don’t see an evidenced basis for Massimo’s characterization of parapsychology as ‘pseudoscience.’ In Nonsense, Massimo says, “lack of progress, i.e., lack of cumulative results over time, is one of the distinctive features of pseudoscience.” But with the example of the Ganzfeld, it seems indisputable to me that there are cumulative results in parapsychology, and that those results provide evidential support for the ESP hypothesis.
Finally, it is noteworthy that the prominent CSI Fellow, psychologist, and self-described parapsychologist, Richard Wiseman, has previously stated that "I agree that by the standards of any other area of science that remote viewing is proven, but begs the question: do we need higher standards of evidence when we study the paranormal? I think we do.”
Wiseman later clarified his comment: “That’s a slight misquote because I was using the term in more of a general sense of ESP. That is, I was not talking about remote viewing per se, but rather Ganzfeld, etc. as well. I think that they do meet the usual standards for a normal claim but are not convincing enough for an extraordinary claim.”
Even granting Wiseman’s insistence on higher standards of evidence for extraordinary claims, I wonder what Massimo thinks of Wiseman’s assertion that the scientific evidence for ESP is decisively in favor of it by the standards of “normal” scientific claims. Surely Massimo would agree that if Wiseman’s assertion is true, then this is characteristic not of pseudoscience but of science.