Golem Effect

Elisha Babad, Jacinto Inbar, and Robert Rosenthal named it in a 1982 study, after the golem of Jewish folklore — a creature brought to life by its maker that turns destructive. It is framed as the negative counterpart to the Pygmalion Effect, the term Rosenthal and Lenore Jacobson gave in 1968 to the finding that teachers told a random set of students were “bloomers” saw those students gain more.

Babad, Inbar, and Rosenthal had teachers nominate high- and low-expectancy students, then measured how those teachers treated them. Teachers who were prone to bias treated the low-expectancy students more negatively and got weaker performance from them; teachers who showed little bias produced much smaller differences. The effect depended heavily on the individual teacher, not just the existence of a label.

Mixed. McNatt's 2000 meta-analysis found management-context expectation effects “fairly strong” on average, but built mostly from military and training settings — the authors themselves called for more research in ordinary, for-profit workplaces. On the other hand, Jussim and Harber's 2005 review argued that most expectations are reasonably accurate, so the truly self-fulfilling slice of the effect is real but modest on average — and Smith, Jussim, and Eccles (1999) found such effects often fade across successive supervisors or teachers rather than piling up. Sorhagen (2013) found the harm lands disproportionately on children from lower-income households even when the average effect is small. Ethical limits also mean researchers rarely induce a true Golem Effect on purpose — Oz and Eden's 1994 work with military fitness instructors, and Davidson and Eden's 2000 study raising expectations for disadvantaged women trainees, both worked by changing expectations upward, not by proving how much damage a new low expectation could do.

Sometimes. Self-verification theory (Swann & Read, 1981) suggests people actively seek feedback that matches their existing self-view, which can look identical to a Golem Effect from the outside. Regression to the mean is a separate statistical trap: an unusually bad first impression, measured on a noisy test, tends to look better on its own next time — which can be mistaken for the effect of raised expectations, or hide the effect of lowered ones. A careful diagnosis has to rule these out before blaming expectations.

McNatt's 2000 meta-analysis found management-context expectation effects were actually stronger when people started out with low expectations placed on them — suggesting the downward pull may be at least as powerful as the upward boost. But that analysis pools Pygmalion and Golem-direction studies together and leans heavily on military settings, so treat the comparison as suggestive rather than a precise head-to-head number.

Deliberately inducing a harmful low expectation in a real classroom, clinic, or workplace raises obvious ethical concerns. Most Golem Effect evidence therefore comes from observing naturally occurring bias (as in the original 1982 study) or from correlational surveys (like Leung and Sy's 2018 work), rather than from controlled experiments that would isolate cause from effect as cleanly as the original Pygmalion studies did.

Cooper and Good (1983), in Pygmalion Grows Up, describe a related but distinct pattern: an authority figure forms an expectation, the person's real performance later changes, but the authority figure keeps responding to the old expectation instead of the new evidence — for example, continuing to hand a student only easy drill work after they have actually improved. This is subtly different from the Golem Effect itself — it is about a belief that fails to update, not one that actively lowers performance — but the two often travel together in practice. Separately, Rubie-Davies and colleagues' later work (2018) found that teachers who over- or under-estimated a class tended to keep that same pattern across different cohorts over several years, which shows how persistent an initial misjudgment can be even without claiming a single, named "effect."