Verified by 
If it is possible that events can have causes occurring later than the time they happen, it would be possible that our perceptions be caused earlier than the events they are perceptions of.
~ Keith Seddon (1991)
Ein Wunder, wenn es nämnlich Wunder gäbe
Ein Wunder, wenn es nämnlich Wunder gäbe
~ Christian Dietrich Grabbe, Westfalian poet [Scherz, Satire, Ironie und tiefere Bedeutung]
In a series of posts (1, 2, 3, 4), I have expressed my belief that, appearances notwithstanding, Daryl Bem (2011) has not demonstrated the existence of precognition. I now clarify and expand my skepticism. Recall that Bem claims to have demonstrated retroactive causation. Consider his first experiment. Participants look at a computer screen, which showis two curtains, one to the left and one to the right. The participant points at one of the curtains, trying to predict where an image will appear. Then, a computer with a random choice generator selects the curtain behind which an image is placed. Then the curtains are lifted. Bem finds a small, but statistically significant, accuracy effect when the images are erotic, and no effect when they are not.
Consider five critical points:
[1] Bem strenuously argues that the computer's selections are as random as modern physics and technology allow. This is important; Bem needs this premise to be in place before he can argue that the predictability of a participant's choices from the computer's choices reveals a causal effect of the later events (the computer's choices) on the earlier events. Correlation is necessary for causation. Yet, Bem's argument is logically incoherent because he must claim that a correlation between the participant's behavior and the computer's behavior simultaneously does (retroactive causation) and does not exist (randomness). A self-contradictory argument cannot be empricially valid.
[2] Bem suggests that his discovery of retroactive causation adds to our knowledge of ordinary proactive causation. He suggests that now we know something that we did not know before. He does not claim that the old, conventional knowledge of regular, forward working causation has become invalid. To many, this might seem like a reasonable point, similar to an argument for reciprocal causation. Often causation goes both ways. Jack yells at Jill, Jills yells back at Jack, and Jacks yells even more. This is an example of a dynamical and escalating process, in which the behaviors of each agent are both causes and effects. Note, however, that these behaviors are ordered in time. They conform to the conventional idea that things happen in sequence. In contrast, Bem's view implies that not only does Jill yell back at Jack because he has yelled at her first, but also that Jack starts yelling at Jill because she will yell back at him once he has yelled at her. In this rendering the terms ‘cause,' ‘effect,' and ‘time' no longer have any comprehensible meaning. If an event is both the cause of a later effect and the effect of that later event, it is neither. The whole language of causation turns into gibberish.
[3] Experiments are conducted so that, if they are successful, we can do something. Random assignment to conditions is critical. If you have 100 patches of land, you can randomly select 50 of them and apply fertilizer. The other 50 patches are the controls. If wheat grows more abundantly on the fertilized patches, you have evidence that the fertilizer works. Now you can act. Fertilize patches you select nonrandomly on the assumption that you can make wheat grow where you want. This is intervention. Now back to Bem. You can make people look longer if you show them erotic pictures as opposed to pictures of rocks. This is conventional causation. But how might you be able to use Bem's results for retroactive causation to influence anything? You cannot travel into the future (as best as we know) to cause events to happen now, nor can you do anything now to influence the past, which has already happened.
I am pointing out this limitation to Bem's claim, although I can imagine an objection: There are many cause - effect relations in nature that are beyond human intervention. The moon, for example, circles the earth because of gravity, and there's nothing we can do about that. Nevertheless, this limitation is noteworthy because in all of Bem's proactive causation experiments the results can be used at human discretion, whereas they cannot be used in any of the retroactive causation experiments.
[4] Bem, to his credit, has encouraged other investigators to run replication studies. For all the reasons I have discussed in this blog, I expect most of these replications to fail. Another reason for this expectation is that even non-parapsychological effects tend to decrease over time. A simple explanation for this decline is statistical regression. Novel effects have an advantage of being published if they are large. This is so because novel effects must overcome the barriers of surprise and skepticism. Once published, statistically significant replications count as successes, but the effect size bar has been lowered. The field of parapsychology is familiar with this phenomenon. Over the last century, there have been repeated cycles of excitement and disillusionment. What is creepy is when the declining effect size is reified as an effect in its own right, which then becomes part of the phenomenon. Should this happen to Bem's work, psi enthusiasts may come to believe that retroactive causation occurred only once, under special circumstances, in a lab at Cornell University. With this mindset, one can hold on to the belief that an extraordinay, anomalous event occurred, if only once. Bem's finding would then be remembered as an existence proof, not as a stable, replicable feature of nature.
[5] Conventional wisdom makes a distinction between the impossible and the improbable. Mathematicians and philosophers debate how this distinction can be justified, if it can be justified at all. If inviolable laws exist, then by definition, events that violate them are impossible. The dead do not return to life and pigs don't fly. In the past, and in some quarters down to the present day, a third category of event is allowed: Miracles. A miracle is a singular, unrepeatable event that violates a law of nature without nullifiying it. The law persists and people can accept the miracle as real at the same time.
Bem can be read as suggesting that he worked a miracle. He managed to show that the future affects the past. Even if no one replicates his results, the results are what they are (and advocates of psi will so argue for years to come). The findings are miraculous because they violate the ingrained notion that time goes by and does not return without simultaneously nullifying this notion.
To overcome the logical conundrum of how an impossible event can happen, one can turn to the improbable. This was David Hume's approach. In a famous essay, Hume (1748/2000) asked what it would take to establish belief in a miracle. He suggested that the falsity of the evidence would have to be more surprising than the existence of the miracle. In his words:
"When anyone tells me, that he saw a dead man restored to life, I immediately consider with myself, whether it be more probable, that this person should either deceive or be deceived, or that the fact, which he relates, should have really happened. I weigh the one miracle against the other; and according to the superiority, which I discover, I pronounce my decision, and always reject the greater miracle. If the falsehood of the testimony would be more miraculous, than the event which he relates; then, and not till then, can he pretend to command my belief or opinion."
Hume's argument can be expressed in Bayesian terms. The question is, what does it take for the probability that an event was a miracle given the evidence, p(M|E), to be greater than the probability that the event was not a miracle given the evidence, p(~M|E). According to Bayes's Theorem,
Assuming that the probability of the evidence given that a miracle has occurred, p(E|M), is 1, and obtaining the probability of the evidence under the no-miracle null hypothesis, p(E|~M), as the conventional p value from significance testing, we conclude that p(M|E) > p(~M|E) if p(E|~M) < p(M)/p(~M). The evidence for the miracle must be less probable under the null hypothesis that no miracle has occurred than the prior probability that a miracle has occurred divided by the probability of no miracle (McGrew, 2011).
Hume's recommendation to believe the lesser miracle is mathematically only an approximation. As miracles must, by definition, be improbable, the division by the probability of no miracle does not do much damage. To make the approximation visible, assume that p(E|~M) = .10. Now p(M) only needs to be greater than .091 to be supported by the evidence.
The appeal of the Hume-Bayes rule is that it quantifies the mandate that extraordinary claims require extraordinary evidence. If advocates and skeptics could agree on the prior probability of the claim being true, they could run significance tests with a p value criterion that is set before study and that reflects the riskyness of the hypothesis. Doing so, they would be using Bayesian statistics not to replace significance testing, but to guide it.
Realistically, however, one cannot expect advocates and skeptics to agree on the prior probabilities. Disagreement regarding the priors is what makes them advocates and skeptics in the first place. The skeptics have an advantage. They can set a very low prior probability for the event in question, thereby making it difficult for the reseracher to meet the threshold of sufficient evidence. The advocates face a psychic conflict (so to speak). They will seek to relax the prior to give evidence a chance, while at the same time trying to maintain a strict (low) prior for the sake of being able to argue that the phenomenon sought is indeed anomalous.
Perhaps I have protested too much. After 5 posts and 5 points within this last post alone, I wonder about overkill. Once the logical incoherence of precognition/backwards causation is established, the issue can be regarded as settled. The empiricist critiques by Humians and Bayesians are then not only redundant, but they entail a tacit admission that procognition is possible in principle, and that the empiricists would believe in it if there only were sufficient evidence.
Bem, D. J. (2011). Feeling the future: Experimental evidence for anomalous retroactive influences on cognition and affect. Journal of Personality and Social Psychology, 100, 407-435. doi:10.1037/a0021524
Hume, D. (1748/2000). An enquiry concerning human understanding. (Ed. T. L. Beauchamp). New York, NY: Oxford University.
McGrew, T. (2011). Miracles. The Stanford Encyclopedia of Philosophy (E. N. Zalta (ed.), URL = <http://plato.stanford.edu/archives/fall2011/entries/miracles/>
Seddon, K. (1991). Precognition and backwards causation. Philosophy Now. http://www.philosophynow.org/issue2/Precognition_and_Backwards_Causation
