Demand Characteristics

In brief

An experiment is not a neutral window onto behavior. It is a social occasion, and the people inside it are trying to make sense of it. Martin Orne (1927 to 2000) argued in 1962 that the totality of cues in a study, its instructions, apparatus, rumors, and the very fact of being tested, convey to a participant what response is expected. He called these cues the demand characteristics of the situation. A cooperative participant reads them and, wanting to be a good subject and confirm the hypothesis, supplies the behavior the study seems to want. The result then measures the participant's theory of the experiment, not the process the experimenter meant to study. A twin problem sits on the other side of the table: the experimenter's own expectations, documented by Robert Rosenthal, leak through tone, posture, and unnoticed cues and nudge the data toward the hoped-for answer. Both threaten internal validity, and both are why serious method treats the observer as part of the system.

The full treatment

The problem it answers

Underneath every experiment is a hope: that if you hold everything constant and vary one thing, the change in behavior you observe was caused by that one thing. Demand characteristics name a way this can quietly fail. The participant is not an inert preparation. Orne noticed that people in experiments are remarkably compliant, willing to perform tedious or pointless tasks for hours simply because a scientist asked, which means they are attending closely to what the scientist seems to want. Any hypothesis a participant forms about the study becomes a possible cause of the behavior, running in parallel with, or instead of, the variable the experimenter cares about. The concept answers a foundational question of method: how do you know the effect you found belongs to your manipulation and not to your subject's guess about your manipulation?

How it works

The mechanism is interpretation. Given cues, participants build a hypothesis about the study's purpose and then act on it. The direction they push is not always flattering; Orne noted the "good subject" who tries to confirm the theory, but there is also the participant who resents being manipulated and pushes the other way. Either way the behavior is a response to the perceived demand rather than to the variable alone. The cues are often subtle: a difference in how two conditions are introduced, a leading debriefing question, campus gossip about what the study is "really" testing. Orne's diagnostics were sharp. Ask, after the fact, what participants thought the study was about (the postexperimental inquiry), or tell non-participants about the procedure and ask them to predict how a real subject would behave. If those predictions match the actual results, the finding may be driven by shared expectations rather than by the manipulation.

Clever Hans, the founding cautionary tale

The purest demonstration predates the term by half a century. In early 1900s Berlin, a horse called Clever Hans appeared to do arithmetic, tapping out answers with his hoof, and a commission of experts found no fraud. The psychologist Oskar Pfungst (1874 to 1932) then ran controlled tests and cracked it: Hans was right only when the questioner knew the answer and was in the horse's view. As Hans neared the correct number of taps, the questioner made a tiny involuntary movement, a slight relaxation or lift of the head, that told Hans to stop. Hide the questioner or keep them ignorant of the answer, and the horse failed. Hans had learned nothing about numbers; he had learned to read people. The lesson generalizes far beyond horses: an observer who knows the desired outcome can transmit it without meaning to, and a subject can pick it up.

The bright and dull rats

Rosenthal (1933 to 2024) turned the Clever Hans problem into an experiment about experimenters. With Kermit Fode in 1963, he gave psychology students rats to run through mazes, telling half that their animals were bred to be "maze-bright" and half that theirs were "maze-dull." The labels were fiction; the rats were randomly assigned. Yet the rats run by students who expected bright animals performed better than those run by students who expected dull ones. Nothing differed but the belief in the handler's head, presumably transmitted through differences in handling, care, and recording. Rosenthal and Lenore Jacobson then carried the idea into schools in Pygmalion in the Classroom (1968), telling teachers that certain randomly chosen pupils were "bloomers" poised to spurt intellectually, and reporting later IQ gains for those children. The claim, that expectancy alone can shape outcomes, became one of the most famous and most contested findings in the field.

The countermeasure: blinding

If the danger is that expectations and guessed purposes contaminate the result, the defense is to remove the knowledge that carries them. In a single-blind design the participant does not know which condition they are in, which starves demand characteristics of their cue. In a double-blind design neither the participant nor the person administering the procedure and recording the data knows the assignment, which also closes off experimenter expectancy: you cannot signal an answer you do not possess. This is the Pfungst fix applied by design rather than after the fact. It is standard in drug trials, where the placebo effect and the double-blind control are two sides of one coin, and it is why a well-run study keeps the code that links subjects to conditions sealed until the data are in.

Lineage

The idea belongs to a mid-century reckoning with the experiment as a social system. Its concrete ancestor is Pfungst's Clever Hans work of 1907, long cited as the origin of the "Clever Hans effect." Orne's 1962 paper gave the participant's side its name and its diagnostic tools; Rosenthal's program through the 1960s gave the experimenter's side its evidence and the label "experimenter expectancy effect." The two strands converge on a principle already implicit in the placebo-controlled trial: the people running and undergoing a study can create the effect they are looking for, so their knowledge must be controlled like any other variable. Demand characteristics thus sit at the root of the modern apparatus of blinding, standardized scripts, and cover stories, all built to keep the subject from correctly guessing the game.

The strongest case for it

The concept earns its place because it is constructive, not merely skeptical. It does not say experiments are worthless; it identifies a specific, testable confound and hands you tools to catch and neutralize it. The quasi-controls Orne proposed (postexperimental inquiry, the simulating-subject design) let a researcher estimate how much of a result could be produced by expectation alone. Rosenthal's rats supply an existence proof that is hard to wave away: with random assignment and a fictional label, belief still moved the numbers, so the effect is real and not a philosopher's worry. And the fix works. Double-blind procedure is now the near-universal standard in clinical trials precisely because it demonstrably strips out these effects, the strongest endorsement any methodological idea can have: it changed how the most consequential experiments in the world are run.

The strongest case against it

The sharpest attacks land not on the concept but on its most famous demonstration. Pygmalion in the Classroom drew immediate fire. Robert L. Thorndike, reviewing it in 1968, argued that the study rested on an IQ measure so unreliable at the ages and score ranges involved that the reported gains could not bear the weight placed on them; some class averages implied nonsensical baseline scores. Later reviewers found the classroom expectancy effect to be small, inconsistent, and concentrated in the earliest grades rather than the sweeping force the popular version implied. The lesson is not that expectancy is fictional but that a striking single study can outrun its own evidence.

A second line of criticism questions how much demand characteristics actually distort results in practice. Some researchers argue the "good subject" is partly a myth: participants often do not correctly guess the hypothesis, and when they do, they do not reliably comply. Attempts to manufacture effects purely through demand cues sometimes fail, suggesting the phenomenon, while real, is weaker and more situation-bound than its reputation. Overstated, the worry even becomes unfalsifiable: any inconvenient result can be dismissed as "just demand characteristics" without evidence that expectation actually drove it.

Finally there is a reflexive irony. The tools for detecting demand characteristics, especially asking participants what they thought the study was about, are themselves suggestive social interactions that can plant the very hypothesis they aim to detect. And blinding, the clean solution in principle, is often impossible in the psychology and social science that most need it: you cannot blind a participant to whether they received therapy, exercise, or meditation. Where the cure cannot be applied, the disease has to be reasoned about rather than removed.

Where it stands now

Demand characteristics and experimenter expectancy are settled parts of the methodological canon, taught in every research-methods course and built into good designs as a matter of routine. The concepts gained new force in the replication crisis: unblinded, expectancy-prone designs are now seen as one channel through which flimsy findings enter the literature, and preregistration and blinded analysis answer the same problem Orne and Rosenthal named. The scope of specific claims, above all the classroom Pygmalion effect, has been trimmed by decades of scrutiny. What remains is the durable core: the observer and the observed are participants in a shared situation, and a result that does not control for what they expect may be measuring the expectation.

Test yourself

Think of a time you gave feedback on something, a survey, a taste test, a friend's draft, when you already sensed which answer the person wanted. Did you report what you actually perceived, or what you read the situation as asking for? Now flip it: when you have run a small test of your own, a diet, a change at work, did you measure the outcome without knowing which condition you were in, or did you already know what you hoped to find? The concept is not only about laboratories. It is about how easily the wish for a result becomes the result itself.