Why does knowing you're in an experiment change how you act?

Because people start guessing what the study is about and adjust to it, often without realizing. Psychologists call this a demand characteristic: an experimental artifact in which participants form an interpretation of the experiment's purpose and subconsciously change their behavior to fit it. One possible cause is that participants expect to be evaluated somehow, so they try to 'beat' the experiment, for example by giving answers that make them look more socially or morally responsible.

The pioneering research on this came from psychiatrist Martin Orne, whose 1950s study found that in most experiments, participants tell experimenters what they want to hear in hopes of pleasing them. The researchers Weber and Cook described roles participants can take on: the good-participant, or 'please-you' type, who tries to figure out the hypothesis and confirm it rather than 'ruin' the experiment; the negative participant, or 'screw-you' type, who tries to figure out the hypothesis just to undermine it; the faithful participant, who follows instructions to the letter; and the apprehensive participant, who is so worried about being judged that they behave in a socially desirable way.

These guesses don't come from nowhere. Participants pick up on rumors about the study circulating beforehand, the way the laboratory itself is set up, and any verbal or non-verbal communication from the experimenter, any of which can tip them off about what's supposedly being tested.

Orne, who was also well known for his research on hypnosis, took part in famous trials too. In 1979 he was a witness in the trial of Kenneth Bianchi, and showed that Bianchi had lied about having multiple personalities to avoid prosecution. Bianchi chatted with an imaginary lawyer Orne introduced; when the real lawyer walked in, a flustered Bianchi claimed the imaginary one had vanished. Bianchi pleaded guilty in October 1979.

Because demand characteristics can't be eliminated outright, researchers try to limit them: deceiving participants about the true hypothesis, keeping measurements unobtrusive, using a double-blind design so the person running the session doesn't even know the hypothesis, and asking participants afterward how much of the study's purpose they think they guessed.