• home
• about
• new
• essays
• AI
• book

Frank Visser, graduated as a psychologist of culture and religion, founded IntegralWorld in 1997. He worked as production manager for various publishing houses and as service manager for various internet companies and lives in Amsterdam. Books: Ken Wilber: Thought as Passion (SUNY, 2003), and The Corona Conspiracy: Combatting Disinformation about the Coronavirus (Kindle, 2020).

The Antechamber Is Not the Room

A critical review of John Abramson's reply on Rupert Sheldrake

Frank Visser / ChatGPT

John Abramson's new essay, “The Question That Can Be Answered,” is considerably more interesting than a conventional defense of Rupert Sheldrake. Rather than attempting to rescue morphic resonance by appealing to scientific revolutions, suppressed anomalies, or the supposed inadequacy of mechanism, Abramson accepts the central methodological challenge: a radical hypothesis has to earn its place by making predictions, surviving measurement and replication, and eventually providing an explanation. His strategy is therefore more modest and, for that very reason, more defensible. Before asking what morphic resonance is, or how it might work, he proposes asking whether there is a genuine statistical effect in the first place.

This is a clever move. It also brings Abramson's position considerably closer to mine than his rhetoric sometimes suggests.

From Metaphysics to Measurement

My essay “Rupert Sheldrake and the Revolt Against Mechanism” identified a missing middle between similarity and causal influence. Sheldrake observes that similar systems sometimes seem to behave similarly and proposes morphic resonance as the explanation. But the crucial step—the coupling that supposedly connects present systems with previous similar systems—remains unspecified. Abramson concedes this point explicitly: “I have no missing middle to offer him; nobody does.” Instead, he proposes what he calls an “antechamber” to that missing middle. Before worrying about the mechanism, he argues, we can ask whether there is any coupling to explain at all, once ordinary statistical dependence has been removed.

That is entirely reasonable. In fact, it is probably the right place to begin.

The trouble starts when Abramson moves from this modest methodological proposal toward a stronger interpretation of what a statistical residual would mean.

Nearly all of Sheldrake's empirical claims have the same basic structure. A quantity measured in one set of systems appears to change as similar systems have previously exhibited the same behavior. Rats supposedly learn mazes more readily after other rats have learned them; crystals supposedly crystallize more readily after other crystals have formed; and, in the latest example, Wordle players supposedly perform better as more people around the world have already solved the day's puzzle. Sheldrake interprets such correlations as evidence that systems are somehow “tuned” to the history of similar systems.

But there is an obvious alternative explanation. Systems that appear independent may not actually be independent. Information can spread between laboratories. Experimental methods can change. Populations can change. Seed crystals can contaminate subsequent experiments. Participants can learn from one another. The difficulty of a task can vary. And in the Wordle case, the answer itself can spread through social media and websites.

Abramson correctly observes that this is, at least initially, a statistical problem. Given repeated observations across systems and over time, statistical methods can estimate shared sources of variation. Hierarchical models, random-effects models and variance-component analysis can distinguish variation within systems from variation shared across systems. If an apparent collective effect disappears once the ordinary shared component has been accounted for, the mysterious effect has lost much of its mystery.

That is good statistical hygiene.

But it is important not to confuse statistical hygiene with causal explanation. A residual is simply what remains after a statistical model has accounted for what it was designed to account for. It is not automatically a new causal influence. A residual can arise from measurement error, omitted variables, model misspecification, nonlinear relationships, temporal dependence, selection effects, random fluctuation or an inadequate model of the data-generating process. The fact that 95 percent of a phenomenon has been explained does not make the remaining five percent a new force of nature.

Abramson knows this, and to his credit he repeatedly insists that his proposed method must be capable of returning a null. Nevertheless, his language occasionally goes further than his statistics warrant. He talks about establishing that “there is something in the room” once the mundane explanations have been subtracted. But a statistically unexplained remainder is not yet a “something” in the ontological sense. It is a signal that the current model has not exhausted the data.

What the KCBS Example Really Shows

This distinction becomes especially important when Abramson turns to his KCBS example from physics.

The example is rhetorically effective. Abramson takes shot-level data from a published contextuality experiment and finds what initially looks like an extraordinary excess of correlation. Such an anomaly could, at least metaphorically, look like the sort of collective effect that a morphic-resonance enthusiast might seize upon. Yet Abramson argues that the apparent excess is largely accounted for by shared instrumental drift. Runs recorded close together in time share a common component, and removing that component eliminates approximately ninety percent of the excess. Furthermore, the effect was not reproduced across three datasets from the same apparatus. What initially looks like an exotic collective phenomenon turns out to have a much more mundane explanation.

If Abramson's reanalysis withstands scrutiny, it is a useful demonstration. It shows exactly why an impressive correlation is not enough. It also illustrates the virtue of a method capable of destroying an anomaly rather than merely discovering one.

But there is a limit to what the example establishes. The KCBS experiment involves measurements performed on the same physical apparatus, where common-mode instrumental drift is an obvious candidate for producing correlations. Wordle is an entirely different statistical environment. It involves millions of heterogeneous human participants distributed across countries, time zones, cultures, devices, levels of expertise and information networks. Demonstrating that one particular physical correlation can be decomposed into ordinary instrumental drift does not establish that the same procedure can identify all of the relevant mundane dependencies in a global human population.

The physics example therefore demonstrates an important methodological principle, but not yet a general-purpose “morphic resonance detector.”

Wordle and the Confounding Problem

The Wordle discussion is nevertheless the strongest part of Abramson's essay.

The recent Wordle studies by Georgia Black, Bethany Butzer and Rupert Sheldrake provide a particularly revealing test case because the apparent positive result is entangled with several other variables. The hypothesis is straightforward: as more people solve a particular Wordle puzzle, later solvers should find it easier. The second study reported a significant morning-to-evening improvement, while the first and third studies did not produce the same result.

Abramson makes an important observation about what “morning versus evening” actually means in the second study. It does not mean that the same population of people was tested in the morning and then again in the evening. The later sample is a different and larger slice of the global population. Consequently, several things change simultaneously. The number of previous solvers increases, which is the supposed “morphic dose.” The geographical and demographic composition of the player population changes. And the amount of information circulating about that day's answer increases.

Those variables are not independent. They all increase as the day progresses.

This is a serious problem for the original design. If more people have already played, there has also been more time for the answer to circulate. A person solving Wordle in the evening may have been exposed to the answer, a hint, a discussion or even an outright spoiler. Thus the variable that Sheldrake interprets as the morphic influence is also a variable that has perfectly ordinary causal pathways into performance.

Abramson makes another interesting observation. The reported improvement in Study 2 was concentrated in the first and second guesses, while later guesses moved in the opposite direction. That is potentially important because answer leakage would be expected to have a particularly strong effect on the earliest guesses. Someone who has encountered the answer can simply enter it immediately. This makes the first-guess effect particularly vulnerable to an information-leakage interpretation.

That is a good criticism of the study.

But it does not establish that leakage is the explanation. It establishes that the design does not adequately distinguish leakage from resonance. That is a more modest—and scientifically more appropriate—conclusion.

Can Guess Distribution Separate Leakage from Resonance?

Abramson therefore proposes looking inside the guess distribution rather than merely comparing pooled morning and evening scores. If morphic resonance genuinely makes the puzzle easier, he argues, one might expect a broader improvement across attempts. If answer leakage is responsible, the effect should be disproportionately concentrated in the earliest guesses.

This is a useful empirical prediction.

But it is not yet a prediction that follows uniquely from Sheldrake's theory. That is the problem.

Why exactly should morphic resonance improve the entire solving process rather than increase the probability of suddenly recognizing the correct answer? Sheldrake's theory is not sufficiently developed to tell us. Perhaps resonance would make a solver marginally better at every stage. Perhaps it would produce a sudden insight. Perhaps it would affect different cognitive processes differently.

Abramson therefore needs to derive his proposed signature from the actual theory rather than simply stipulate it as the expected signature of resonance.

Otherwise we have replaced one ambiguity with another.

The deeper problem is that morphic resonance has never been specified with enough quantitative precision to tell us what its empirical fingerprint should look like. That is not a minor technical omission. It is central to the question of whether the theory is genuinely testable.

The Burden of Proof Quietly Shifts

There is an important rhetorical maneuver in Abramson's essay.

He begins with the reasonable question: can we distinguish genuine resonance from ordinary statistical dependence?

But what counts as “ordinary statistical dependence”?

That category can become extremely broad.

Suppose we control for time of day, geography, sample composition, answer leakage, puzzle difficulty, player experience, social-media exposure, temporal autocorrelation, selection effects and measurement error. Suppose a residual remains.

What exactly have we demonstrated?

We have demonstrated that the statistical model did not account for the residual.

We have not demonstrated that all ordinary causal explanations have been eliminated.

This is the distinction between statistical modeling and causal identification. The former can tell us that an association remains after specified adjustments. The latter requires a much stronger argument that the relevant alternative causal pathways have actually been blocked or accounted for.

Abramson's proposed analysis is therefore useful as a way of generating evidence, but it cannot by itself determine whether the residual is paranormal, biological, psychological, social or simply an artifact of an incomplete model.

And that brings us directly back to your original “missing middle.”

Correlation Is Not Morphic Resonance

This is where I think the original essay has the stronger philosophical position.

Abramson writes as though Sheldrake's fundamental empirical claim were that similar systems exhibit correlations across time. But that is not really morphic resonance. That is the observable pattern Abramson wants to investigate.

Sheldrake's hypothesis is considerably stronger. Previous similar systems are supposed to exert a causal influence on present similar systems through morphic fields or morphic resonance. The statistical correlation is therefore not the theory itself. It is supposed to be the observable footprint of the theory.

This distinction matters because one can establish an anomalous correlation without establishing morphic resonance.

Suppose Abramson performs his proposed analysis and discovers that Wordle performance really does improve as the number of previous solvers increases, even after controlling for geographical composition, information leakage, puzzle difficulty and other identifiable confounds. That would certainly be interesting. It would justify further investigation. It might even establish a genuine anomaly.

But it would not establish morphic resonance.

The conclusion would instead be something like: “Here is a reproducible statistical regularity for which our current causal models do not provide an adequate explanation.”

That is a scientifically interesting result. It is not yet evidence for morphic fields.

This is why Abramson's distinction between an “antechamber” and the actual room is useful. His statistical procedure might get us into the antechamber. It cannot, by itself, take us into the room.

The Missing Quantitative Theory

Indeed, Abramson eventually admits exactly this. Even a positive result, he says, would not provide the “missing middle.” It would merely establish that something remains after ordinary explanations have been removed.

But this concession substantially changes the significance of his essay. He is not actually providing evidence for Sheldrake's theory. He is proposing a research program for determining whether Sheldrake's theory deserves further investigation.

That is a much more defensible claim.

There is also a deeper issue that Abramson does not sufficiently confront: morphic resonance still lacks precise quantitative predictions.

What effect size should we expect? What mathematical relationship should exist between the number of previous similar systems and the performance of subsequent systems? Should the effect be linear, logarithmic or power-law? Should it saturate? Should temporal distance matter? Should geographical distance matter? How should “similarity” itself be quantified? How much prior activity should produce how much subsequent change?

Without answers to such questions, even a sophisticated statistical analysis can become somewhat open-ended. The hypothesis predicts that there should be an effect, but the size, functional form and conditions of that effect remain poorly specified.

This is precisely where a theory has to move beyond suggestive language. “Resonance” sounds like a mechanism, but until it generates quantitative predictions it remains largely a label for the very phenomenon it is supposed to explain.

The “Subtle Effect” Escape Hatch

The danger is the familiar escape hatch of the “subtle effect.”

If an experiment finds the predicted effect, resonance is supported. If the effect disappears, it was perhaps too subtle to detect. If one study succeeds and another fails, perhaps conscious cognition interfered with the resonance. If the effect occurs under one set of conditions but not another, those conditions can retrospectively be declared unsuitable.

This is where pre-registration becomes essential. But pre-registration can only solve the problem if the hypothesis itself makes sufficiently precise predictions beforehand. A preregistered analysis of an underspecified hypothesis can prevent statistical p-hacking, but it cannot magically make the underlying theory falsifiable.

Abramson's proposal is therefore only as good as the theoretical predictions that go into the preregistration.

And that is precisely the part of the morphic-resonance program that remains underdeveloped.

Where Abramson Is Right

Nevertheless, I think Abramson has identified a worthwhile improvement in the debate. He is right that one need not demand a complete mechanism before establishing an empirical regularity. Science often proceeds from phenomenon to regularity, from regularity to causal structure, and only then to mechanism. There is nothing illegitimate about asking whether an unexplained statistical pattern exists before demanding an explanation for it.

I would therefore slightly modify the challenge posed in my original essay.

Rather than insisting that Sheldrake immediately provide the mechanism, the more productive demand is this: first establish a robust, preregistered and independently replicated effect that survives serious attempts to identify ordinary causal pathways. Then we can worry about the mechanism.

But that effect must itself be diagnostic.

And this is where Sheldrake remains in difficulty.

The Antechamber Is Still Not the Room

Abramson's most revealing sentence comes near the end of his essay, when he says that a positive result would establish that “there is something in the room” without establishing how the systems are coupled.

That is a perfectly reasonable research position.

But it is a long way from demonstrating morphic resonance.

In fact, the disagreement between Abramson and me now appears narrower than his essay suggests. I am saying: show me a robust phenomenon that genuinely demands a new explanation. Abramson is saying: let us first construct a statistical instrument capable of determining whether such a phenomenon exists.

Those positions are entirely compatible.

The real disagreement would begin only after that instrument produced a positive result.

At that point the burden would return squarely to Sheldrake. Why should the unexplained residual be called morphic resonance rather than some as-yet-unidentified ordinary causal process? What distinctive prediction does morphic resonance make that its competitors do not? Under what conditions should it operate, how strongly, and how could we prove it wrong?

Those questions cannot be answered by variance decomposition alone.

Abramson has built an antechamber. That is useful. It may even be exactly the right place to start. But an antechamber is not the room, and finding an unexplained statistical residual is not the same thing as discovering a new causal principle of nature.

The scientific challenge to morphic resonance therefore remains essentially where it was: first show us a phenomenon that survives rigorous attempts to explain it conventionally; then show us why morphic resonance, rather than merely our ignorance, is the best explanation of what remains.