Is psi a legitimate topic of study?

Weird stuff happens from time to time – it’s a fact of life. When I was applying for a position as assistant professor in Exeter, the weeks before the interview the word ‘Exeter’ started to pop up in weird places in my life – newspaper articles, names of conference rooms I happened to present in, stuff like that. Or that time that my then four-year old, right before leaving home, started to talk about the street in front of his school being changed into a river – when we arrived fifteen minutes later we found the street to be flooded by a bursted water pipe.

Coincidence? Probably. Nevertheless, most people will have experienced such weird coincidences, and sometimes people interpret these experiences as ‘paranormal’. Paranormal experiences include clairvoyance, telepathy, precognition, and psychokinesis. Not surprisingly, these phenomena, collectively labelled as ‘psi’, are met with great interest by the general audience, but traditionally, psychologists have been very interested in them as well. William James, the founding father of psychology, studied paranormal phenomena, and the name-giver of our research intstitute, Gerard Heymans, credited as the first Dutch experimental psychologist, was also the founder of the Dutch Society for Psychical Researcy, for example.

However, psi research (or parapsychology) was met with increasing skepticism. The weird experiences we all have from time to time turned out to be very difficult to replicate in laboratory settings, and basically, ever since its inception, parapsychology is struggling to confirm psi even exists, let alone come up with a solid theoretical background. As a result, nowadays, parapsychology is typically considered to be (at best) fringe science, or even pseudoscience, and is actively ignored by main stream psychology.

A small number of researchers have continued parapsychological investigations, though. Rather than aiming for dramatic demonstrations of psi by mediums, for example, modern-day parapsychologists study far more subtle effects. One of the best-studied topics is that of presentiment (see eg. Bierman and Radin, 1997): an anomalous baseline response to a stimulus that yet has to appear. An emotional picture, for example, induces a strong physiological response. In presentiment, this response appears to be temporally mirrored in the baseline: emotional picture evoke a physical reaction before they are presented, and apparently this physiological response is symmetrical in time (i.e., if the typical physiological response occurs at t = 2.5s, the ‘presponse’ will occur at t = -2.5s). Presentiment has been quite widely studied, and the effects seem to be more stable and legit than your typical psi-effect.

Presentiment has drawn wide mainstream attention with the publication of Bem’s landmark 2011 paper ‘Feeling the future’ in JPSP (Bem, 2011). Bem claimed to show solid evidence for the existence of presentiment in a whole series of experiments, in over 1000 subjects. However, the paper was met with a storm of criticism. I am calling Bem’s paper a landmark paper because it can be pinpointed as the paper that started off the massive debate on questionable research practices in (social) psychology, even before the uncovering of the fraud cases of Stapel, Hauser, Smeesters, and (likely) Foerster. The main critique on Bem’s work was that Bem appeared to ‘shop’ in his results, and did not use appropriate statistics: basically, he was accused of practices such as ‘p-hacking’ (mildly massaging your data until you hit p <.05), and ‘HARKing’ (hypothesizing after the results are known). Moreover, replications of Bem’s experiments failed, and the refusal of JPSP to publish these failed replications only added to the uproar surrounding the field of social psychology (one set of null results has been published in PLOS One, though: here).

Bem, however, did not give up. He, and his coworkers, have published two meta-analyses on the presentiment effect. Armed with the statistical techniques used by his critics (such as p-curves and Bayesian analyes), Bem and colleagues analyzed 90 experiments and show that there ‘is something going on’ in their latest meta-analysis, to be found here.

What to make of this? Does Bem’s latest work indeed support the existence of psi? Let me first say that a) I consider myself to be an open-minded skeptic (more on that later), and b) that I am far from an statistics expert. Having reviewed the latest paper I find the case compelling at first sight. I have got two major concerns, though.

1. How big is the file drawer? Bem et al. estimate that the overall effect size they find for psi can only be negated by at least 520 studies finding no effect. At first sight, this seems a large number. But is it really that unlikely that there are 520 experiments yielding no result Bem et al. did not know about and therefore did not include? Well, it is not. The Bem-experiment has drawn wide attention, and is likely to have been replicated. In our own institute several teaching assistants (ie. not academic faculty!) have used the Bem case as example in second year research practicals. Such data is typically not archived, as it is not considered to be ‘research’, and therefore goes unnoticed. Tenured and tenure-track faculty are less likely to engage in direct replications for the ‘usual’ reaons: lack of time, concerns about tenure, or simply lack of interest. In other words, I do not think 520 experiments (worldwide!) is a completely unrealistic number. The file drawer on psi might be quite substantial.

In addition, if they exist, pre-cognitive effects should also emerge in main-stream research. If a stimulus can exert retro-active influence, this should also be apparent in paradigms that are not explicitly designed to measure psi. For example, in a typical priming experiment, one could look at effects of primes (or targets) presented in trials after the critical stimulus. If precognitive effects do exist, one should be able to find effects in such datasets as well. Now, I myself have several datasets in such weird things are going on, although the most parsimonious account is still a main-stream one, obviously. However, ideally, to make a strong case for psi, main-stream research should be taken into account as well.

2. The theoretical background. The Achilles-heel of psi-research, though, is the lack of a clear theoretical framework. Just that you find a difference between two conditions in an experiment does not mean that much: laws of probability dictate you will find a difference in 1 out of every 20 experiments with p<.05. What matters if your difference fits with a given theory: were you able to predict you would find a difference? Many critics argue that parapsychology lacks a theoretical basis, and that given that psi is incompatible with the laws of physics, research into psi is by definition pseudoscience.

And that is where many critics are wrong.

Psi is not necessarily incompatible with physics. Bem et al. drag in quantum physics, which is something more scholars appear to do when stuff gets complicated (I consider it to be a bad habit), but there is no need for quantum physics to find a loophole in physics that allows psi. Psi involves the flow of information backward in time. Perhaps surprisingly, this is not impossible according to physics. Most physical laws are in fact time-symmetric. For example, in the formula x(t) = v * t, which gives the position x of a point mass at moment t given its speed v, there is nothing against entering a negative number for t. The only physical law which imposes a direction of the flow of time is the second law of thermodynamics, which postulates that the entropy of a closed system will increase over time. Basically it means that the sugar cube you put in your coffee will dissolve, but not spontaneously re-appear after it dissolved, and that your coffee will get colder over time, but not hotter (unless you heat it).

This is such a fundamental property of the universe that it dictates a direction of time on all other physical processes. However, it is also not a law, but a statistical property of the universe. Basically, potential high-entropy states of a system are far more abundant than potential low-entropy states. Consider the sugar cube in your coffee: the organized, low-entropy state in which sugar molecules are when packed into a cube is quite specific. All molecules of the sugar cube are packed into a confined spatial location (the cube), which limits the number of possible states. However, molecules of a fully dissolved cube can be anywhere in your mug, yielding a far larger number of possible states. Yet, it is not impossible that a sugar cube will rematerialize in your cup of coffee after dissolving (there is no physical law prohibiting the sugar molecules in your cup of coffee of reassembling themselves into a sugar cube), it is just impossibly unlikely. And to date, we do not know how the universe got itself into this mess – why was entropy at the beginning of everything so low?

That aside – technically, time reversals, or better, the universe adopting a state where effect appears to precede cause, are not impossible, just extremely unlikely. But does it mean that presentiment can be real? Well, there is a second problem. Even if we accept that information in extraordinary circumstances can travel backwards in time, presentiment potentially creates time paradoxes. The best-known example is the grandfather-paradox. Suppose you have a time-travelling machine, and travel back in time to when you grandfather was a child. And you shoot him. Apart from that being quite cruel in the typical circumstance where you have a loving grandpa, it creates a paradox: if your grandfather dies before he fathered one of your parents, how can you exist and travel back in time to shoot him?

The same thing applies to presentiment. Technically, it is possible to build a presentiment detector if the effect is real: Bierman and Radin (1997), for example, report an anomalous increase in galvanic skin response before presentation of an unpredictable emotional stimulus compared with the baseline response to neutral stimuli. If you sample a participants GSR to stimuli for a sufficient number of trials, it is well possible to estimate a participant’s typical baseline to neutral stimuli and estimate whether the baseline response on a given trial is typical of a subsequent neutral or subsequent emotional stimulus. That would allow you to predict the future: by a quick analysis of anticipatory GSR activity you should be able to guess (not perfectly, though) whether the upcoming stimulus will be emotional or neutral. And this is where the paradox arises: if your analysis is quick enough, you can quickly switch of the monitor, blindfold the participant, or even change the upcoming stimulus before the stimulus is presented and thus prevent the participant from seeing the stimulus that would trigger the emotional reaction. Effectively, this is the same as shooting your grandfather in the grandfather-paradox: if you erase or change the event that triggered the presentiment response in the first place, how can it trigger presentiment?

Stephen Hawking has argued that the universe ‘actively resists’ time paradoxes. Bierman (2001, and personal communications) supposes that this is one of the reasons why psi effects are so elusive: as soon as they become informative, and are able to create time paradoxes, they cease to exist. I am not sure if this is the case or not. However, I find the time paradox argument against presentiment compelling – a lot more compelling than the incorrect argument that psi does not fit with the laws of physics.

In their meta-analysis, Bem et al. cite a Bayesian analysis of Bem’s 2011 paper by Wagenmakers et al. In Bayesian statistics, one evaluates the likelihood of evidence gained by an experiment, or set of experiments, relative to the likelihood of the data given a prior hypothesis. Wagenmakers et al. set their prior to 10^20, meaning that they would be convinced of the reality of presentiment if the outcome of an experiment was 10^20 as likely to be found under the hypothesis that psi exists than under the hypothesis that it does not. Bem et al. report a Bayes-factor of 10^9 in favour of the existence of psi. Impressive – but not impressive enough for Wagenmakers et al. Bem et al. argue that a prior of 10^20 is unrealistically and unreasonably high. Well, is it?

Wagenmakers et al. base their prior on the fact that psi is incompatible with the laws of physics. With that argument I do not agree (see above) per se, but I do think the time paradox argument is convincing enough to warrant a very high prior. 10^20 seems not unrealistic or unreasonable in that sense to me

So to conclude: I am open to the possibility of psi. There are no universal laws against it, it is just extremely unlikely. Bem’s meta-analysis is convincing, but to me not convincing enough. It would be far more convincing if Bem et al. could make more specific theoretical predictions about when and how presentiment might occur. They do offer some interesting points – in particular, presentiment seems to be most prominent in ‘fast thinking’ scenarios. Now, that is actually what I would predict as well. If it exists, presentiment is most likely not a ‘conscious’ phenomenon. Personally, I have never consciously experienced a time-symmetric emotion. However, most of us will know the experience of a ‘gut feeling’ (“I knew something bad was gonna happen!”) My own research has shown that such ‘gut feelings’ are easier picked up when you’re not consciously trying to ‘listen’ to them. So, in that respect, if presentiment exists, it might indeed be easier to find it in experiments that require ‘fast thinking’.

One final remark: some critics argue that psi should not be studied at all, and ridicule any serious treatment of the topic. I have even seen some tweets arguing that psi is an excellence benchmark for novel statistical methods – if you find evidence for psi using your statistical method, your method is wrong!

This stance seems profoundly unscientific to me, in particular coming from people who claim to have strong dislike for pseudoscience. You simply cannot and should not dismiss results purely on the basis that they do not fit your beliefs. I’d say we have some bad experiences with that in the Middle Ages.

Arguably, the theoretical case for psi is not strong, but it is strong enough to take seriously. As a researcher, I study consciousness, and to be frank, we haven’t got a clue how consciousness works. The hard problem of consciousness is still far from solved, and there are profound philosophical problems with the materialist theories of consciousness. Psi (if it exists) may offer new insights into the nature of consciousness. For me, this is the main reason to regard the field with interest and not dismiss is right out of hand. In that respect, I think Bem’s work is valuable, interesting, and deserves recognition from main stream science, and I’d rather have skeptics trying to shoot at the data and analyses than actively ignoring psi altogether. Therefore, I am happy to see that many people have taken a critical look at the data; in the end, this can only give us a clearer picture of what is going on.

Showing 1 Reviews

  • Thumb 203 203
    D Samuel Schwarzkopf
    Originality of work
    Quality of writing

    This is a nice commentary. I already discussed in some of the points you make here in response to your blog post ( in which you presented as similar argument as in this letter.

    For some time I have been meaning to reply to Mossbridge et al.'s response to me ( I will do this here and also include a discussion of the points you make in the present letter.

    For now I just want to point out that, like in your blog post, you seem to be conflating presentiment and precognition. Many commentators, especially in social media coverage of these papers, also conflate these concepts. I personally regard both of them as equally implausible. However, since the authors of the psi literature appear to draw a clear distinction between the two, I think this is important to do the same when discussing them.


This article and its reviews are distributed under the terms of the Creative Commons Attribution 4.0 International License, which permits unrestricted use, distribution, and redistribution in any medium, provided that the original author and source are credited.