Paranormal Claims curriculum

Popper's test is whether a claim can conflict with possible observations: it must name, ahead of time, a result that would show it wrong. A match with the data counts only when the prediction was risky and could have failed. The flip side matters too: critics of Popper point out that many claims labeled pseudoscience are testable and have simply been tested and failed.

An escape clause is a condition added after a failure to rescue the claim, such as “a skeptic in the room blocked it.” Each one restores the fit, but a claim that can be saved from every outcome stops predicting anything. Popper called such rescues ad hoc; a test arranged so it can only confirm, never refute, is a recognized warning sign.

Before a claimed ability can be tested it has to become something you can count. In ganzfeld studies, “sensing a hidden picture” becomes a hit: choosing the real target from a set of one target plus three decoys. Fixing the measure first also fixes what chance alone would produce, so the result has a known baseline to beat.

A prediction goes on record before the outcome, so it can fail; a postdiction is an interpretation fitted to events afterward. Critics have long noted that a hypothesis can be adjusted to fit almost any data after the fact. Philosophers still argue over whether and why a successful prediction deserves extra weight, but preregistration exists to keep the two apart.

Many philosophers doubt that any single feature marks pseudoscience, so lists of warning signs are common, and some authors read them as family resemblances rather than strict rules. Typical entries are deference to a special authority, experiments that others cannot repeat, examples picked because they fit, refusal to test what could be tested, and quietly ignoring results that refute the claim.

One positive result says this test, run this way, produced data that would be unusual under chance. It does not say why: a small p-value can come from a large random error or a broken assumption as well as a real effect. Hume's rule applies, proportion belief to the evidence, so one result earns a cautious claim and independent replications earn more.

Someone can truly see a figure in the doorway or hear a voice, and that report deserves to be taken seriously as a report. What caused it is a separate question; people around the world appraise similar unusual experiences in very different ways. An old distinction from debates over miracles captures it: the event is a matter for testimony, its cause is not. Tests target the explanation.

Whoever makes the claim owes the evidence for it. An opponent's inability to offer a better explanation does not make the claim true, and inferring truth because nothing shows it false is a recognized fallacy. Insisting that doubters must disprove a claim is what argument theorists call misallocating the burden of proof.

Hume's maxim: testimony for a breach of natural law succeeds only if its being false would be more remarkable than the event, because the law rests on uniform experience. Scholars still dispute how to read the maxim, but its core is a weighing, with the evidence for a claim set against all the evidence it contradicts.

Bayes' theorem makes the starting point explicit: how believable a hypothesis is after the data depends on how believable it was before and how well it predicted the data. Two people can agree on what a study shows yet conclude differently because their priors differ. On one influential model, in a field where few tested ideas are true, more of the published positives will be false.

Evidence counts for more when it is varied: agreement among different methods is harder to explain away than one method giving the same answer again. Repeating a study checks chance, artifacts and fraud, but it can repeat a shared flaw too. Replications that change the setting and measures test whether the idea holds up in new ways.

A null result is informative only if the study could have detected the effect. A small, noisy study that finds nothing mainly shows that the data cannot separate the competing hypotheses, and even large effects can be drowned in noise. A search counts against an effect when its confidence interval excludes effect sizes big enough to matter.

“We can't explain it” describes what is known so far; it points to no particular cause, paranormal or otherwise. A US intelligence review of unidentified aerial reports said most stayed unidentified probably because of limited data, and Hume argued that odd events, examined closely, generally turn out to have hidden causes.

Before a new mechanism is on the table, the ordinary ones have to be ruled out one by one: chance, misperception, memory error, instrument faults and trickery. A US intelligence review of unexplained aerial reports set up the same kind of list: any case it could resolve was expected to land in a bin such as birds, balloons and debris, atmospheric effects, or developmental or foreign craft, with a catch-all bin last.

A vivid story shows that something happened once to someone, but not how often it happens, or how often the opposite does. Even genuine personal experiences cannot show what caused them, since a placebo effect or other factors may be responsible. US guidance on health advertising treats collections of such experiences, even surveys of them, as never enough on their own to back a claim.

Accounts reach an investigator because someone found them striking enough to pass on, so a file of remarkable cases is a sample chosen for being remarkable. Treating it as typical is the handpicked-example error. Reports also pile up where people are told to watch and report, which a US intelligence review flagged as a possible collection bias.

Dim light, long distance and a brief look all limit what a witness can take in, and no one can control those conditions after the fact. In one large study, people watched a mock crime for either 6 or 34 seconds, and memory for the culprit was weak in every age group under both conditions, so even a longer look can leave thin information.

Certainty is not a simple readout of accuracy. Some researchers argue that high confidence, recorded at once under fair, double-blind procedures, does track accuracy well; other studies find that link breaks down when viewing conditions are poor. And vivid memories of striking events tend to keep their high confidence even as their details drift over the years.

The wording of a question shapes the answer. A question that assumes a detail invites the witness to accept it, and some people go along with such suggestions more readily than others. Open-ended prompts let the witness supply the content instead, and one research team notes that questions offering closed alternatives should be avoided in real interviews.

Each retelling is a fresh reconstruction, not a replay, and witnesses often describe an event many times before testifying. Talking it over adds a further risk, because people can take on details from a fellow witness they believe knows better. Studies of whether retrieval itself raises later suggestibility are mixed, but a first account written down right away at least predates that influence.

Honest witnesses can be wrong without lying. In tests of staged slate-writing reported by 1900, sincere sitters wrote up marvels that never happened. Credentials do not settle it either: US guidance on health advertising says an endorser's honest opinion is not enough to show a product works, and counts a practitioner's observations of patients as anecdote.

Without a comparison group, improvement in the people tested could come from a placebo effect, natural change over time or simple practice at the task. A control group, ideally given a sham version of the procedure, goes through the same conditions, so a difference between groups can be pinned on the treatment rather than on those other causes.

If people choose their own group, or an investigator chooses for them, the groups can differ in ways that drive the result. Letting a random process assign people helps prevent that selection bias and keep the groups alike in age, health and other traits. Large differences between the groups undermine any comparison drawn from them.

Blinding the participant keeps expectations from shaping what they report; blinding the researcher keeps knowledge of the condition from leaking into handling and scoring. Double-blinding does both, greatly reducing the chance that anyone, knowingly or not, biases the result. Where full blinding is impossible, the standard is to blind as far as possible.

An experimenter who knows what result is expected can shift participants' responses without meaning to. These effects are clearest when both experimenter and participant know the condition. That is why, in ganzfeld work, the assistant who deals with the participant is kept blind to the target until the participant's judging is done.

A target picked by hand, a poorly shuffled set or a reused picture can give a guesser real patterns to exploit. Critics of one ganzfeld meta-analysis argued that studies with manual randomization showed larger effects than computerized ones, a reading its authors disputed. Protocols also have the guesser choose from an unhandled copy of the target.

Setting the number of trials, the stopping point and the success test before the run removes any chance to adjust them as results come in. Checking significance before deciding whether to collect more data, or stopping once results look good, are recognized questionable practices; a public, date-stamped plan written beforehand rules them out.

Repeating a study elsewhere helps check for chance, hidden artifacts and fraud. When a 2011 precognition report appeared in a major journal, three independent replication attempts all failed to reproduce it, though a later meta-analysis reported a smaller effect. Critics of remote viewing research raised a related point: the judges who scored it were not independent of the labs.

A direct replication repeats the original procedure as closely as possible to check the result; a conceptual one changes key features to test the idea another way. In practice they sit on a continuum. A conceptual failure is easy to explain away as a change in setting, so a string of conceptual successes can hide the fact that the original was never directly repeated.

When studies that find nothing stay in the drawer, the published record overstates the effect, and any meta-analysis built on it inherits the distortion. Parapsychologists argued over this early, and that debate led one of their journals to start reviewing plans before results existed. Journals willing to accept null results reduce the problem.

In an unbiased literature, small and large studies should scatter around the same effect. If small studies show bigger effects, a funnel plot looks lopsided, which can signal missing null results. Treat that as a clue, not proof: differences between studies can produce the same asymmetry, and visual inspection has been judged unreliable.

Pooling many studies gives a more precise estimate than any one of them, but it cannot repair what went into the pool. If published studies are inflated by selective reporting or flexible analysis, the pooled result is inflated too. A simulation of one ganzfeld database concluded such practices likely inflated its significance, though an unexplained effect remained.

Preregistration means filing your predictions and planned analysis publicly before the data exist, so exploratory findings are harder to pass off as tested predictions. The gap shows up in practice: in one parapsychology registry, preregistered studies reported significant results far less often than unregistered studies from the same years.

In a registered report, the journal reviews the question and methods before the study runs and promises to publish whatever the outcome, provided the plan is followed. That removes the reward for a positive result. A prototype ran in a parapsychology journal from 1976 to 1992, long before mainstream psychology took up the format.

Many effects in psychology and medicine shrink as studies are repeated, and parapsychologists described the same pattern in their own data long ago. Most researchers attribute it to inflated early results from questionable practices. Some psi researchers instead read it as a property of psi itself, while one recent ganzfeld meta-analysis reported no sign of decline; the question stays open.

A p-value is computed by assuming that chance alone is operating, and it tells you how unusual your data would be under that assumption. So it cannot be the probability that chance produced the result, nor the chance that your hypothesis is correct. A small one flags something unusual; it does not say which assumption failed.

Given enough trials, a tiny deviation from chance will still clear the significance bar. That tells you the data are unlikely under the chance model, not that the effect is large or useful. To judge importance, look at the estimated size and its confidence interval: a significant hit rate a few points above chance is still a few points.

Effect size is how big an effect is; power is a test's chance of detecting it if it is real. Small studies can miss even large effects, and a field of small studies chasing small effects produces results that are true less often. A warning sign is mismatch: estimates put neuroscience power at 31 percent at best, while 85 percent of reported results in neuroscience and behavior were significant.

Each extra outcome you test is another chance for a fluke to cross the line, so a study checking many measures can turn up something “significant” by luck alone. Findings are less likely to be true when more relationships are tested with less preselection. The remedies are to report every outcome, name the main one in advance and adjust the statistics for the number of tests.

If you check results as they come in and stop the moment they look significant, you raise the false-positive rate even though each step feels reasonable and no one means to cheat. It is common: in surveys in two countries, more than half of psychologists admitted collecting more data after checking significance at least once. The fix is a stopping rule set in advance.

Every analysis involves choices: which data to exclude, which outliers to drop, which covariates to include, which outcome to report. Made after seeing the data, each choice can nudge a result toward significance, and together they can turn noise into an apparent finding. Flexible designs, definitions and analyses make a field's findings less trustworthy.

HARKing, hypothesizing after the results are known, presents an unexpected finding as though it had been predicted all along. It turns exploratory work into what looks like a confirmed test, and in surveys roughly a quarter to a half of researchers across fields admitted doing it at least once. Preregistration helps by putting the original prediction on record.

Give something with one-in-a-million odds a hundred million chances and you should expect about a hundred occurrences. That is the law of truly large numbers: with a big enough sample, extraordinary coincidences are expected somewhere. The question to ask of a striking story is not “how unlikely?” but “how many chances were there?”

Ask how many people a party needs for an even chance that two share a birthday. Any pair counts, and the number of possible pairs grows much faster than the number of people, so a match arrives sooner than the headcount suggests. One test battery on paranormal belief included this problem, though errors on its dice and sample-size tasks, not this one, were what tracked belief.

A match is impressive only if the detail is rare. “Someone close to you had heart trouble” sounds specific, yet heart disease is the top cause of US deaths, so the guess often lands by chance. Base-rate neglect is underweighting how common something is when you take in new information, a well-documented source of error in medical and legal decisions.

The gambler's fallacy is expecting tails to become more likely after a run of heads, as if a fair coin owed a correction. Its mirror image is the hot-hand belief, which expects a streak to carry on. The hot hand is classically seen with runs of human performance, such as correct guesses, while the gambler's fallacy follows runs from devices like roulette wheels.

Truly random sequences contain more runs and clumps than people expect, so people rate sequences that alternate too often as more random than they are. A cluster in time or place therefore does not need a special cause. In one study, people who saw patterns in random coin tosses were also more likely to hold supernatural and conspiracy beliefs.

After a striking match, it is natural to ask how likely that exact match was, and the answer can be tiny. The fairer question is how likely it was that some match of that general kind would turn up, given every chance for one. With enough opportunities, extraordinary coincidences are expected, and an evidence pool searched for matches will find them.

Pareidolia is seeing a meaningful pattern, most often a face, in unrelated input such as clouds, stains or rock. It seems to come partly from fast face detection misfiring: in one eye-tracking study, food items with illusory faces drew the eyes faster, but only in peripheral vision. The face you see is a real experience, but it is evidence about vision, not about the stain.

Ambiguous sound invites you to fill in words, and being told what to hear changes what you hear. In one study, showing people an alternative written version of song lyrics could make them hear those words, and knowing the alternative beforehand strengthened the effect. So be wary of any noisy recording that arrives with its transcript attached.

People often fail to notice an unexpected object or event when their attention is busy elsewhere, even if it happens in plain view. In one magic study, a coin moved openly between napkins for about half a second and many viewers still missed it. That gap between what is visible and what is seen is what misdirection exploits.

Change blindness is failing to notice a large change between two views of a scene when a brief interruption, such as a flicker, separates them. The change is fully visible, yet it goes unseen. That is why a witness who insists nothing changed is reporting what they noticed, and that can differ from what happened.

Stare at a single dim light in a completely dark room and it seems to wander in random directions, even though it is fixed. This autokinetic effect may be the simplest visual illusion, and its cause is still debated, since two dots can drift independently, which eye movements cannot explain. A report that a lone light moved is weak evidence by itself.

To judge an object's size, the brain uses cues to its distance and corrects for them. A light in the night sky offers few such cues, so it is reasonable to expect its size, distance and speed to be uncertain together. A US intelligence review noted that some reports of unusual flight could reflect sensor errors or observer misperception.

What you expect changes what you detect, most of all when the input is weak. Told that most noisy displays would contain a human figure, people reported seeing one more often, especially in the noisiest displays. Researchers still debate how much expectation alters the percept itself rather than the decision about it.

Remembering is not playback. Each time you recall an event, your brain pulls together pieces of past experience and fills gaps, often adding content that was never in the original. In one well-studied case, people remember parts of a scene that lay beyond what they actually saw.

Information you meet after an event, from a news story, a conversation or a question, can work its way into your memory of the event itself. This is the misinformation effect, and witnesses who discuss an event with others, including online, can absorb details they never saw.

You can remember content correctly and still misplace where it came from, recalling something you only imagined as something you saw. People sort sources using cues such as detail: memories of real experience tend to be more detailed than imagined ones. Those cues are imperfect, so a vivid imagining can be filed as an event.

Imagination inflation is the rise in your confidence that something happened after you have imagined it happening. Researchers test it by having people rate how sure they are about childhood events, imagine some of them, and rate again. It is one route by which picturing an uncertain experience can make it feel more like a memory.

Researchers have led people to recall whole events that never happened, from being lost in a shopping mall to committing a crime, using trusted sources, false details, imagining exercises and repetition. The person is not lying; the memory feels real. When others watched recordings of true and implanted memories, they could not reliably tell them apart.

Memories of where you were when you heard surprising, emotional news feel vivid and permanent. Research finds these flashbulb memories decay and distort like ordinary ones, yet confidence in them stays high as the details shift. Vividness and certainty are therefore poor guides to accuracy.

Once you know how things turned out, you tend to remember your earlier forecast as closer to the outcome than it was. In classic hindsight studies, people asked to recall their original estimates assign more probability to what actually happened than they really did. By the same logic, a vague hunch may come to be remembered as a specific premonition, though that step is an inference.

Confirmation bias means seeking, picking out and interpreting information that backs up your existing belief while ignoring what could count against it. In a classic number-rule task, people tend to test examples that fit their guess. Trying cases that should fail works better there, though researchers debate how far that task measures the bias.

Francis Bacon said it centuries ago: people mark when a prophecy hits and never mark when it misses. So a record built from what people recall may already be filtered, though his old observation is not a measurement of how memory stores hits and misses. Lab work finds a related habit in how evidence is gathered: paranormal believers sampled more cases where the supposed cause was present.

Subjective validation is judging a statement accurate because it feels personally meaningful, not because it is specific to you. It is one reason vague personality readings win high marks, alongside flattery: most people readily accept descriptions that make them look good. The fit you feel is evidence about your reaction, not about the reader's insight.

When two unrelated events often happen together, people tend to read a causal link into the pattern, and the illusion grows when you mostly see cases where the supposed cause was present. In one study, paranormal belief went with sampling more of those cases and with stronger illusions, but only when no real relationship existed; real ones were detected equally well.

An ambiguous sound or movement can easily be read as someone's doing. One account holds that expectations combined with unclear input produce false detections of agents, including supernatural ones. In a lab test, people told to expect a human figure in visual noise said yes more often, most strongly when the display was noisiest.

On one influential view of self-deception, desire, fear and anxiety can bias how evidence is read with no intent to deceive yourself. A grieving person who longs for contact, or a parent who fears bad news, may misread evidence that an uninvolved observer would weigh calmly. If that view is right, the bias can sit alongside complete sincerity.

Beliefs often survive after the evidence that created them is retracted, a pattern called belief perseverance, so announcing that a claim was false may not undo it. In one experiment, a refutation pointing to arguments for the opposite view worked better than two other debiasing techniques, though all three reduced the bias. Comparisons of such techniques are still scarce, so treat that as one result.

A Barnum statement sounds personal but fits almost anyone: “You have had a setback lately that still stings.” Since nearly everyone has, most listeners relate. People rate whole sets of such vague, general statements as highly accurate descriptions of themselves when they believe the sketch was prepared just for them.

In the classic demonstration, each student receives what looks like a personal profile, rates how well it fits, and only then learns everyone got the same text. One modern version reused most of the original 1949 sentences and still drew high ratings, around 80 on a 100-point scale. The reveal is the lesson: the sense of fit was not evidence of accuracy.

Cold reading leans partly on playing the odds: a general statement is likely to fit someone, and every extra guess is another chance for a match. A match then gets noticed while a miss does not, the old pattern Bacon described when he wrote that people mark when prophecies hit and never mark when they miss.

Fishing for information is among the named techniques of cold reading: rather than knowing a detail, the reader draws it out of the sitter. A century-old account describes a stage mind reader taking clues from signs a person was sure they never gave. Blinded designs keep the reader away from the sitter for this reason: no contact, no feedback to fish from.

Readers lean on what is statistically common. A mentalist classic asks for an animal starting with E, its color, then a country starting with D; most people land on a gray elephant in Denmark, so the “reveal” is safe. The performer is not reading minds, only betting on the likeliest answer, and the prevalence of names and medical conditions can be exploited the same way.

A vague profile can run two ways at once. One stock personality result used in a lab study called people assertive and independent, yet sometimes uncertain for no apparent reason. A line that covers a trait and its opposite is hard to get wrong, which makes it a close cousin of the deliberately vague, ambiguous statements in the Barnum profiles people rate as accurate.

Some readers present information gathered in advance as if it came from intuition. A 1924 expose by Harry Houdini described mediums' helpers trailing clients, searching court records and reading newspaper proofs. The safeguard is a reader who never meets the sitter and works under blind conditions, so research done beforehand has nowhere to attach.

In this control, the reader never meets the sitter, a stand-in asks the questions, and each sitter later scores two anonymized readings, one meant for them and one for someone else, without knowing which is which. Real insight should earn the target reading higher scores. One such study reported positive results for some mediums; the wider field has no consensus.

Misdirection steers attention so the secret move happens unnoticed. Magicians aim it at perception, memory and reasoning alike: drawing the eyes with a look at the other hand, letting a detail be forgotten, or encouraging a wrong inference. It works because attention, not eyesight alone, decides what you register, which is why the move can happen in full view.

Audiences remember the effect, the coin vanished, and then supply a method from imagination. In one study, most children who watched a vanishing pen tried to explain the secret, none got it right, and both young children and older adults were especially overconfident. Magicians also aim at memory, letting the real method be forgotten or misremembered.

Scientific training teaches you to read nature, which never sets out to fool you; it does not teach the specialized tricks of a performer who does. A classroom study found that educated adults, scientists among them, could be led to credit ordinary magic tricks to psychic power, and spotting sleight of hand reliably took magicians years of practice.

Adding a magician to a testing team guards against deliberate cheating, a threat scientific training does not prepare you for. Experience matters: in one study, only magicians with more than five years in the craft spotted sleights of hand better than non-magicians. Over a century ago the psychologist Joseph Jastrow already called this work for experts in conjuring and detectives.

Some of what looks impossible on stage was prepared before it started: facts gathered about audience members, and helpers placed among them. Old accounts of seance mediums describe confederates trailing sitters for information and “committees” of friends used as tools or accomplices. A fair test has to control who enters the room and what is known in advance.

In billet reading, sitters write names on slips of paper, fold them, and the performer appears to read them unopened. A 1920 account describes one medium swapping the folded slips for blanks, then reading the originals in the shelter of his hands while relighting a cigar. The control is simple: slips that never leave the writer's hands.

Metal bending is a familiar psychic demonstration, and part of its power may lie in what viewers think they saw. In one study, verbal suggestions made people more likely to report seeing a spoon bending. Reports of an object softening by mind alone therefore need recording and close, independent observation before they count as evidence.

A magician who says “this is a trick” is entertaining; someone who presents the same trick as real psychic power is making a factual claim. The framing matters: when the same magic show was introduced as psychic rather than conjuring, audiences gave more psychic explanations, and over half considered psychic explanations even when told it was a trick.

A receiver relaxes in an even, unpatterned field of light, with translucent covers over the eyes and steady noise in headphones, and describes whatever images come to mind. Elsewhere, a sender views a randomly chosen picture or video clip. At the end, the receiver picks which of four candidates was the target. The setup masks outside input; it cannot by itself rule out cues leaking through the people running it.

With one real target among four, guessing alone hits 25 percent of the time, so that is the line any claimed effect must beat. Ganzfeld databases have reported pooled hit rates of around 31 percent. To mean anything, that excess must survive selective reporting, flexible analysis and leakage of cues, which is where the dispute lies.

In the mid-1980s, a critic and a proponent argued in print over flaws in the early ganzfeld studies. Their joint communique agreed the database showed an overall effect that selective reporting or multiple analysis could not readily explain, and set guidelines that later studies followed. The agreement was narrow: the critic went on faulting that early database as too flawed.

The autoganzfeld studies that followed handed target randomization, selection and judging to a computer, aiming to avoid the flaws critics had raised, and they followed the joint guidelines from that earlier exchange. They reported a hit rate above chance. Automation answers procedural objections, but it cannot by itself show that independent labs get the same result.

Analysts pooling overlapping ganzfeld studies have reached different verdicts depending on which studies they include, which they drop as outliers, which statistical model they use and whether old and new databases may be combined. One review found no significant effect for a decade of studies; others found one. Read the inclusion rules before the bottom line.

Whether the ganzfeld result replicates is the crux. A 2024 meta-analysis by authors of earlier positive reviews reported a small effect with no decline over time; an earlier simulation by other researchers concluded questionable practices likely inflated the significance, though an unexplained effect remained. Skeptics hold that errors and bias explain the record, and there is no consensus.

Forced-choice tests of ESP and precognition, with work dating to the 1930s, have a guesser pick from a fixed set of possible targets, so the chance rate can be computed exactly. Proponents cite significant meta-analyses of these guessing studies. Critics of parapsychology have focused on safeguards against sensory leakage and on randomization, and early researchers already discussed selective data and multiple analyses.

Some psi researchers propose that certain experimenters can consistently obtain positive results where others do not. That can be tested, for example by running one study with ten experimenters and comparing them. The trouble is interpretation: a difference could reflect the experimenter, whoever first looked at the data or something else, so one researcher in the field warns the claim is hard to falsify.

In remote viewing, a person tries to describe a distant target, often a place, a person or a fact, while cut off from ordinary sensory access to it. In the classic design, judges then rated how well each description matched the targets, so the result rests on that matching step. Programs funded by US intelligence agencies ran for over two decades before being declassified in the mid-1990s.

Free descriptions are loose enough that matching them to targets takes judgment, and critics objected that the original program's judges were not independent of the labs that ran it. One later study dropped that judging design as highly unstable and used forced-choice scoring with sealed targets that an independent technician handled and checked.

When the US programs were declassified in the mid-1990s, two outside reviewers assessed them for a research organization and disagreed. Utts judged remote viewing empirically validated; Hyman did not consider it scientifically established and criticized some early methods. Both agreed the later, more rigorous studies showed consistent effect sizes, so the sharpest dispute was over the conclusion to draw from them.

The programs aimed to learn whether remote viewing was consistent and stable enough for espionage. A statistically significant lab effect does not answer that: in a large enough study a very small effect can reach significance, and significance does not show the effect is important. Applying this to remote viewing is an inference from statistics, not a measured result about field use.

A micro-PK design asks whether intention alone can shift a physical random source, usually one driven by quantum events, away from the 50 percent that chance expects. Early pooled analyses mixed quantum generators with weaker algorithm-based ones, so a later meta-analysis kept only the quantum studies.

Before an intention effect means anything, the generator must pass randomness checks such as the NIST tests and give chance-level output when nobody watches. One large study ran its whole protocol with no observers as a control; a critic replied that a single control run is too few, since a pattern only stands out against many shuffled or simulated datasets.

Because any effect would be tiny, studies need thousands of participants or millions of trials to detect it, and at that scale a minute deviation can reach statistical significance. A 2006 meta-analysis of 380 quantum-generator studies found a very small, uneven effect that weakened as studies grew larger; researchers still disagree about what that pattern means.

A benchmark study from Jahn's group ran to millions of trials. When teams from Princeton's PEAR laboratory and two German institutes tried to repeat it, with over 2 million more trials, the attempt failed to confirm the original results. A later paper lists it among high-powered replication failures, the test of whether a result travels beyond one setting.

A micro claim is a small statistical tilt in random output; a macro claim, such as a key bending, is a visible event a conjurer can fake. In one study, people who watched a faked key bend were more likely to report that it kept bending after hearing a verbal suggestion, so eyewitness accounts of macro effects need testing too.

In the laboratory, the claim is that your responses can be influenced by a stimulus picked at random only after you have responded. The design records the response first and lets a computer choose the target afterward, so there is no ordinary way for you to anticipate which trial is coming.

In 2011 a high-impact psychology journal published a set of experiments by Daryl Bem reporting this time-reversed influence, with an editorial explaining the decision and hoping for replication attempts. A later meta-analysis by Bem and colleagues pooled 90 experiments from many laboratories to defend the claim, which remains disputed.

Three preregistered replications of one of Bem's memory experiments found no significant effect. A later multi-laboratory replication of his first experiment, built with extra transparency so that neither side could easily dispute it, came in close to the chance rate and below the original's figure.

Critics noted that effect sizes in Bem's experiments shrank as samples grew, a pattern that can arise when data collection ends once a result looks good. They also faulted the experiments for not allowing for multiple analyses and argued a Bayesian analysis would not treat the results as evidence. Bem and colleagues replied that the priors in those analyses were unrealistically low.

Bem's paper is often named as an origin of psychology's replication crisis. Three failed replications were turned down by four journals, including the one that published the original, as not novel, which strengthened the case for registered reports, where a journal reviews a study's plan before any data exist.

To judge a predictor, score every published prediction, not only the hits that get quoted later, and decide in advance what counts as a miss. One philosophical account lists selective attention to confirmations as a mark of pseudoscience, and Popper described how a failed prediction can be explained away with ad hoc excuses.

A credible test fixes its analysis and stopping rules before data collection, randomizes targets by machine, and sends data directly to a trusted third party as it arrives, so nobody can adjust it afterward. Open data and logged sessions then let outsiders confirm that the protocol ran as written.

Leakage is any ordinary route by which target information reaches a participant, and it need not be deliberate. A commission ruled out intentional tricks in the case of Clever Hans and believed familiar unintentional signs were excluded too, yet the horse turned out to read his questioner's unintended movements.

Ganzfeld protocols test the participant in a room isolated from outside sound and light, and in sender designs put the sender in a separate, distant room. The protocol has to say how that isolation is achieved: Clever Hans failed once large blinders blocked his view, which showed his cues were visual.

Pfungst traced the horse's counting to small movements: a slight forward lean started the tapping and a tiny upward jerk of the head stopped it. Animal-cognition studies now add controls such as blindfolding the handler, so an animal cannot succeed by reading cues nobody meant to give.

A protocol states who picks the target, how, and when. In ganzfeld research the target is drawn by a random number generator, the assistant who deals with the participant stays blind to it until the ratings are done, and some designs choose the target only after judging, so no one can know it beforehand.

Research on subliminal priming reports that stimuli flashed too briefly to reach awareness can still influence later responses, so a correct response does not prove information arrived by an unknown route. Clever Hans is the classic warning: thousands of spectators watched for months without spotting the small unintended movements he read.

A control named before the data arrive is part of the test; an objection invented after an unwelcome result is not. Clever Hans's owner explained away failed blind trials with stories about the horse's moods, the move Popper called saving a theory with ad hoc hypotheses. Agreeing controls in advance with both sides avoids it.

Scientists have been tricked into accepting a magician's psychic abilities, and in one classroom study highly educated adults credited ordinary magic tricks to psychic power. Simply introducing the same performer as a psychic instead of a magician shifts how audiences explain an effect, so a laboratory must plan for trickery it cannot see.

Assume a participant may cheat and ask what a skilled performer could do at each step. Stage mediums once invited committees to bind them and mark around their feet, and they still beat those checks. Houdini wanted claims tried under test conditions before expert mystifiers as well as open-minded committees.

A confession counts most when it comes with a method you can check. In 1888 one of the Fox sisters confessed in print and then produced the rapping sounds on stage in front of an audience. Weigh what was demonstrated: a later retraction, or a confession with no method behind it, is only testimony.

Spirit photography left a paper trail in court, by Houdini's account. In one early fraud trial, expert photographers showed how such pictures could be produced; the defendant was acquitted for lack of proof but refused 500 dollars to repeat his pictures in another studio under test conditions. Another practitioner, once arrested, told the court how his had been made.

The US Office of Research Integrity defines research misconduct as fabrication, falsification or plagiarism, and excludes honest error and differences of opinion. One review argues that peer review and replication rarely catch fraud, so prevention rests on data handling: archived raw data, audit trails, blinding and preregistered analyses.

A single unwatched interval can be enough. Stage mediums did their work inside a closed cabinet, and witnesses to one conjurer's spirit-writing trick swore they never lost sight of his hand on the table while the writing was produced elsewhere. A protocol has to cover every second and every place a result could come from, not just the spot everyone is watching.

The idea of an outcome can trigger small muscle movements with no felt intention. Because you sense no act of will, a moving pointer or rod feels pushed by something outside you. In 1852 William Carpenter used this to explain dowsing rods and pendulums without spirits, and the same account is now applied to Ouija boards.

With several people's hands on a pointer or a table, small unconscious pushes add up, and a wheeled pointer magnifies them. In 1853 Faraday placed a roller device between sitters' hands and the table that showed any sideways push; with it in place, a table that had turned stayed still.

A pendulum is one of the devices Carpenter explained as ideomotor action, and dowsers use it as well; old claims held that pyrite pendulums swing near metal. A fair test hides the answer from the person holding it, so expectation has nothing to work on and only a real signal could produce a hit.

Where rainfall and geology are favorable, almost any spot yields water, so a dowser finding water proves little. A test needs targets placed by a random procedure, a chance rate you can calculate, and a hit criterion written down before the first trial, so success cannot be redefined afterward.

A US Geological Survey leaflet says that published work on dowsing includes scientifically controlled experiments and, from those findings, concludes that further tests are not worth the expense. It explains field success by ground water lying beneath most land, not by dowsers sensing veins, which ground water rarely forms.

In many areas a dowser can hardly fail, because some water lies under the surface almost everywhere. Dry holes get blamed on hills, a short circuit of current or a careless driller, and a success rate such as 8 in 10 says nothing about how much water was found or of what quality.

Ideomotor movement needs no intent to deceive, so an honest practitioner can produce it. Clever Hans was questioned by people of unimpeachable honor who swore they gave no sign, yet their cues were real. Judge the method with a blind test, not by the character of the person using it.

In facilitated communication a helper physically supports a person's hand or arm while they spell out words. Supporters present it as giving a voice to people with complex communication needs, often autistic, with the stated aim of fading the support toward independent typing.

A message-passing test gives the communicator information the facilitator does not have, such as an object or question shown only to them, then checks whose knowledge appears in the typing. In such tests users have largely failed to pass on information the facilitator lacked.

Over about 25 years, more than 19 rigorous experimental studies traced the messages to the facilitator, not the person with a disability. Proponents cite over a hundred naturalistic and text-analysis studies and call for new research, but education reviews still treat the method as discredited.

The standard account treats facilitated typing as ideomotor action, the mechanism behind a Ouija pointer, and some researchers liken it to Clever Hans reading his trainer. Recent studies propose that users make some identifiable contribution, which leaves authorship to controlled tests, not impressions.

Critics call facilitated communication dangerous as well as ineffective, and argue that such approaches can draw families away from supports with evidence behind them. One review of autism treatments lists it among approaches not recommended because controlled trials failed to show benefit.

The practice continues under other names, including spelling to communicate and the rapid prompting method, which critics count as the same technique, and systematic reviews of the adapted forms again report no evidence of effects. A new name does not change the question: a message-passing test can still check whose knowledge the words carry.

An authorship test asks whether a technique carries a person's own words; it does not measure their intelligence or worth. Some autistic advocates argue for presuming communicative competence and for new FC research done with autistic people as partners, still aimed at assessing whether it is valid.

Electronic voice phenomena are apparent ghost voices found on ordinary audio recording equipment. Researchers treat them as a likely illusion, in which a listener perceives a meaningful pattern in sound that has none, and in one study listeners did not agree on what the voices said.

What you are told before listening shapes what you hear. In one study more than a third of people heard voices in random noise after watching videos about EVP, and in another, people told they would hear ghost voices heard more voices in EVP clips than in degraded speech, though that effect replicated only in part.

The control for priming is to have several listeners transcribe a clip with no hint of what it might say. If words are really present, independent transcripts should agree; in one study using purported EVP recordings, listeners showed no agreement about the spoken content.

Before treating a sound as a voice, account for the noise every recording carries. EVP may be nothing more than random audio signals within white noise that listeners hear as speech, which is why one study compared EVP clips with plain white noise and degraded speech.

An EMF meter reads the local electromagnetic field, which weakens quickly with distance, so moving the meter two or three feet can change the reading a lot. A thermometer at floor level can add a slow downward drift, which in one study may have produced apparent temperature correlations. Neither reading identifies a cause by itself.

A photographic anomaly means little until it is compared with somewhere nothing is reported. One field study set a reputed haunt against a nearby control house, had independent judges rate apparent anomalies across several film types, and checked them against room features such as mirrors and windows.

A review of haunt research found that few studies measured several environmental variables at once, compared haunted areas with suitable control areas, or paired investigators holding opposing views. Rarely triangulating field readings also made results hard to compare. Fixing those steps beforehand turns a visit into a test.

Infrasound is sound below the normal range of hearing, typically under 20 Hz. A 1998 proposal held that sound near 19 Hz could vibrate the eyeball and produce ghostly sights. A 2002 study tested this in a reputedly haunted cellar, but later measurements at the same venue did not match, and critics found the original measurements poorly documented.

Carbon monoxide poisoning causes headache, dizziness, weakness and confusion, and sleeping people can die before they notice symptoms. A 1921 case report traced a couple's headaches, phantom footsteps and figures to a faulty furnace, which is why a safety check comes before any ghost hunt.

One physics analysis argues that naturally occurring vibration and low-frequency sound can rattle windows and create whispering-gallery effects. Yet a comparison of haunted and non-haunted rooms in a historic mansion found no differences in humidity or number of air vents, so a plausible cause still needs measuring.

Expectation may be enough to produce haunt-like reports. In one experiment, people told they might be exposed to fields or infrasound reported many odd sensations in an empty room, but how often they did had nothing to do with whether the exposure was on, though dim light and cool air may also have played a part.

Some field studies report different magnetic readings where people describe odd experiences, and one lab claimed to induce experiences with patterned fields. A failed replication pointed to suggestion and prior belief, which that lab disputes. In double-blind tests of phone-mast signals, people reporting field sensitivity had no more symptoms than under sham, though one signal raised their reported arousal.

A natural explanation also has to earn its place with evidence. A review weighing infrasound, fields and air quality found the carbon monoxide link documented in only a couple of cases, and said recent field-measured anomalies still need independent validation before anyone calls them the answer.

In sleep paralysis the muscle atonia of REM sleep carries on after you wake, so you are aware but cannot move. Recordings show waking brain rhythms together with the muscle quiet of REM, while breathing and eye movements continue.

Hypnagogic imagery comes as you fall asleep and hypnopompic imagery as you wake, and either can involve any sense. During sleep paralysis these hallucinations may include floating, seeing your own body from outside, or seeing, hearing and sensing an intruder in the room.

Researchers group sleep-paralysis hallucinations into intruder experiences, a felt evil presence; incubus experiences, pressure on the chest and suffocation; and floating or movement feelings. The first two usually come with fear, and failed attempts to breathe deeply may add to the pressure.

The same bodily events get read through local belief, from demon or ghost attacks to alien abduction. Researchers describe the brain weaving chest pressure, paralysis and a shadowy figure into a personal story, and where a culture offers no framework the episode is sometimes taken for an ordinary dream.

Abduction accounts share features with sleep-paralysis episodes, and researchers propose that some of them begin that way. The overlap shows that an ordinary mechanism can produce such a story; it does not show that every account had that origin, so each still needs its own evidence.

Details recovered under hypnosis in abduction cases cannot be assumed to be memory. Hypnosis can increase how much people report by loosening their standard for remembering without improving accuracy, and suggestions can inflate confidence in false memories, so vivid detail may be newly made.

Sleep paralysis is common: estimates for at least one episode run from about one person in five up to 40 percent of the general population. Online surveys can read far higher, but one that did said its figure was probably biased by recruiting through sleep paralysis forums. An experience that ordinary makes an everyday explanation the first one to test.

Researchers still argue whether hypnosis is a special state of consciousness or ordinary social behavior, and work in the Barber tradition produced hypnotic effects with only a short instruction to imagine. What it does not do is unlock accurate memory: hypnosis can make people report more, but not more accurately.

Hypnotic suggestibility is a stable trait that varies widely, can be measured with standard scales and is partly heritable. In one sample about 11 percent scored high, and harder effects such as suggested paralysis or hallucination appear only in highly hypnotizable people, so a dramatic result tells you about that person, not about what hypnosis does to everyone.

Hypnosis can lead people to report more by relaxing their standard for what counts as remembered, but the extra recall is not more accurate, and suggestions can raise confidence in false memories. In one US survey, more than half agreed hypnosis helps witnesses recall crime details accurately; nearly all the experts surveyed disagreed.

Past-life regression uses hypnosis to take a person back to supposed earlier lives. One medical ethics paper argues the method rests on an unsupported reincarnation hypothesis and risks implanting false memories; cryptomnesia, material once stored and then forgotten, is one ordinary source of seemingly unknown detail.

Researchers still dispute whether trauma can be unconsciously repressed and recovered later: one side says the belief persists in clinical and legal settings and can do harm, the other that attacks on repression are misguided. Both treat suggestive therapy aimed at digging up memories as a hazard, and some people later retract recovered abuse memories.

In child investigative interviews, research favors open prompts such as “Tell me what happened,” saving direct questions until open ones are exhausted. Early directive questions can turn leading, and props such as anatomically correct dolls can lead to false reports. That evidence comes from children's accounts, so it guides other interviews by analogy.

Astrology holds that the positions of stars and planets can predict the future or describe what a person is like. Popper took it as an unusually clear example of pseudoscience, and later philosophers add that claims this checkable are not untestable: they can be, and have been, put to the test.

A fair test of astrology asks whether a birth chart can be matched to its owner's personality profile above chance while nobody knows the pairing. Philosophers cite a double-blind test published in Nature among the studies behind their statement that astrology has been tested and refuted.

The same design tests any reading: hide the real target among decoys, have an independent judge make the match, and compare the hit rate with what guessing alone would give. A system earns support only by beating that baseline reliably, under rules fixed before anyone sees the results.

People tend to rate vague personality descriptions as highly accurate when told they were written for them, even though they fit almost anyone. In one class exercise, students received a supposedly personal profile adapted from the sentences of the classic demonstration, shuffled so classmates comparing notes would not spot the trick.

A birth-month effect is not a sun-sign effect. One study of 2,962 college students found links between birth season and a few temperament subscales, with small coefficients, and suggested that environmental factors in early development may explain them, a route that needs no planets. Its authors say the cause still needs more research.

Earth's axis wobbles against the fixed stars over about 25,800 years, so the stars behind a given season slowly shift, and the stars of a constellation are not physically connected. Those objections concern mechanism; whether chart readings predict anything is a separate question, answered by blind matching tests whatever the mechanism.

Handwriting analysis claims to read character from script and can face the same blind-matching tests as astrology. Even a 2020 paper defending it concedes that validation evidence is thin and earlier studies were methodologically weak, while arguing that this is not enough to reject it.

Eyewitness reports and footprints are offered for creatures such as the yeti and bigfoot, but no bodies or recent fossils have ever been authenticated. Science neither accepts nor rejects such an animal without examining the evidence, which is why hair and tissue samples matter.

A 2014 survey sequenced 30 hair samples attributed to anomalous primates and found known mammals; two Himalayan hairs matched an ancient polar bear, a result later questioned. A 2017 study placed its purported yeti samples with local bears, except one museum tooth that came from a dog.

Genetic work keeps pointing to known animals: one study concluded the biological basis of the yeti legend is local Himalayan and Tibetan bears. Two earlier papers, later described as tongue-in-cheek, sequenced a Nepalese yeti sample and hair attributed to an Alaskan sasquatch, and both identified hoofed mammals.

New species enter the record when researchers collect and describe specimens, often in little-explored places. Exploring hydrothermal vents in the Southern Ocean with a remotely operated vehicle, scientists found ecosystems dominated by a new species of crab and other animals, and most ocean species are estimated to be still unclassified.

People singled out for an extreme value, such as a high symptom score, tend to measure closer to average next time with no treatment at all. The more extreme the first value, the larger the expected change, which is why a remedy taken at the worst moment looks effective unless untreated people are measured the same way.

Many conditions vary or improve on their own, so improvement in a treated group alone may reflect spontaneous change, practice on the test, or placebo. The FTC holds claims about naturally varying conditions to a stricter standard because consumers cannot judge them on their own.

A placebo effect is a benefit that comes from expecting an intervention to help, and a clinician's manner may add to it. Because it can follow any treatment, claims open to placebo effects need a comparison arm given a placebo or sham, and individual stories cannot rule it out.

Even genuine customer stories cannot show a product works, since the change may come from placebo or unrelated causes, and advertisers may pick the people with the best results, leaving failures unseen. FTC guidance asks for proof that featured results are typical, or a clear statement of what people generally get; results-not-typical notes do not suffice.

The FTC requires health claims to rest on competent and reliable scientific evidence, which for a health benefit generally means randomized, controlled human clinical testing. Consumer surveys and anecdotes never qualify, and a practitioner's observations of patients are anecdotal too.

A surrogate endpoint, such as blood pressure, stands in for a clinical outcome like stroke. The FDA treats one as validated only after extensive evidence links the two, though a merely reasonably likely surrogate can support accelerated approval. Reporting only the measures that came out well is a related trap; FTC guidance asks for outcomes defined in advance and all of them reported.

An unproven remedy can cost more than its price if it delays or replaces one that works. The FTC notes that an unsupported therapeutic claim could lead someone to forgo a more effective treatment, and it says traditional-use claims should not be made for serious medical conditions.

Writing down conditions, controls and success criteria before testing heads off later arguments about what a result means. When believers and skeptics agree the design together, including which outcome would count for each side, neither can easily disown the result once it is in.

A success criterion has two parts, both fixed before the first trial: the chance rate, which follows from the number of options, and the result that will count as success. Presetting the analysis points and the evidence level that ends data collection also stops anyone from quitting while ahead.

A pilot run lets you fix procedure, train staff and settle practical problems before anything counts. Piloting, checklists and session logs also give you a record that the formal test ran as intended, so a later dispute can turn on what happened and not on memory.

Standing cash offers for a demonstration under test conditions are old: in 1924 Houdini counted over forty years of offers from 500 to 5,000 dollars for a single proven case, and he recorded claimants who declined to try. An unclaimed prize still shows only that no one passed, not that every claim is false.

A failed test is a result about one claimant, one protocol and one occasion; it counts against the claim without refuting it everywhere. Reports of failed Bem replications say the results do not support psychic ability, and a large Bayesian study reports graded evidence for no effect, both narrower than a blanket verdict.

Scientific skepticism means keeping an open mind while accepting only claims that survive rigorous tests; cynicism closes the question in advance. Houdini, writing against mediums, still insisted he was not a scoffer, and students are urged to ask what evidence would convince them instead of dismissing a ghost claim out of hand.

Report how strong the evidence is, not only a verdict. Some analyses state how many times more likely the data are under no effect than under an effect, which gives a graded answer. And flag what you noticed afterward: a commentary later showed that one post hoc pattern in a micro-PK study would arise by chance about one time in three.

Some paid claims reach people in grief or illness. Houdini recorded a medium paid 1,000 dollars by a man for a few minutes with the spirit of his dead wife, and the FTC warns that an unsupported health claim can lead someone to forgo treatment that works.