Blog · Telephone Telepathy
Rupert Sheldrake's telephone-telepathy studies report hit rates above chance, including the often-cited result of about 40% correct guesses where 25% would be expected. That is interesting, but it does not establish telepathy. The real questions are whether the effect survives stronger controls, independent replication and a pre-registered analysis — which is exactly how Ashta proposes to test it.
Sheldrake's telephone-telepathy work starts with a common experience: the phone rings, and before looking, someone feels they know who is calling. The testable version is simple. A participant nominates several possible callers; on each trial one is selected at random; the participant must guess who is calling before answering.
In the best-known early paper, Sheldrake and Pamela Smart reported 571 trials with 63 participants and four possible callers. Chance expectation was 25%; the reported hit rate was 40%, with confidence limits from 36% to 45%. Later automated mobile-phone trials, with software selecting callers and recording guesses, reported 41.8% correct against a 33.3% chance baseline. Those results are not nothing — and they are also not enough to treat telephone telepathy as established science.
The strongest objections are methodological, not ideological. Were all trials genuinely blind? Could participants or callers infer anything from timing, failed calls, message routing or incomplete sessions? Were analyses planned in advance, or chosen after the data was seen? Were high-scoring participants selected after the fact? Sheldrake's own automated paper is careful on one point: it describes the mobile-phone studies as exploratory and notes most participants were unsupervised, leaving room for uncontrolled communication. That does not mean cheating occurred — it means the design does not fully exclude it.
Replication is mixed. An independent study by Schmidt and colleagues attempted the phone-call paradigm and did not find a significant overall effect in self-selected samples, though one participant later scored above chance in follow-up. If there is an effect, it may not be a broad human ability detectable in ordinary samples — or it may be what noisy data and repeated testing look like. For Ashta the right response is neither dismissal nor belief but measurement: a pre-registered, automated, timestamped, blinded test with locked analysis code, where the participant sees no caller ID, notification metadata or timing side-channel, and incomplete-trial rules are fixed before the study begins.
So "40% versus 25%" is a starting point, not a conclusion. It tells us one research programme has reported an anomaly worth auditing. Ashta's interest is narrower and stricter: can a person identify an unknown caller above chance under conditions that remove ordinary cues — and does it replicate? That is the protocol behind the Tuned forced-choice guessing game. If yes, we say so. If no, the null is still published. A clean null is not a failed story; it is the result.
