In short
The observer-expectancy effect is the change in a result that happens when the person running a study expects a particular outcome and, without meaning to, passes that expectation on to the people or animals being studied. This page is about that route in particular, the expectation traveling from the researcher to the subject through handling, tone and timing; the related entries cover the wider set of ways a researcher's beliefs tilt a study and the way an expectation changes what is recorded.
How the observer-expectancy effect works
The effect needs a route from the researcher's head to the subject's behavior, and the route is made of small unplanned acts. A handler who expects a clever animal picks it up more readily and sets it down more carefully. An interviewer who expects a particular answer leans forward at the start of it, pauses a beat longer, or reads the question with the emphasis that suggests one option. None of this is deception, and the people doing it generally cannot report it afterward, which is why asking researchers to be careful does not remove it.
The classic demonstration is a horse. Clever Hans appeared to do arithmetic by tapping a hoof, and worked only when the questioner knew the answer and was in view, because the horse was reading the small change in posture that came as the right number of taps approached. The questioners were sincere and had no idea they were signaling. Hans was reading a cue nobody knew they were sending, and that is the shape of the effect in every setting since.
A second route travels under the same name but is a different mechanism. Expectation also changes what the observer records: an ambiguous trial is resolved in the expected direction, a borderline symptom scored as present, a stopwatch stopped a fraction early. Here nothing reaches the subject; the tilt is entirely in the measurement. The two routes usually run together, and the evidence for the first comes from designs that rule the second out, such as automated scoring or coders who do not know the condition.
The size of the effect is not fixed. It is largest where the measurement calls for human judgment, where the subject can see or hear the experimenter, and where the subject has no firm answer of their own and is looking for a cue about what is wanted. It shrinks toward nothing when the outcome is machine-recorded, the condition concealed and the procedure scripted. The effect is a property of the design, not of the honesty of the person running it.
The original study
Robert Rosenthal and Kermit Fode published "The effect of experimenter bias on the performance of the albino rat" in Behavioral Science in 1963. Twelve students in a laboratory course were each given five rats, 60 in total, and told the animals came either from a strain bred to be good at mazes or from a strain bred to be poor at them. The two descriptions were false: the rats came from the same stock, and the groups were made up so that their average ages matched closely. The task was an elevated T-maze in which a correct response was recorded when the animal went to the darker platform, and each student ran their rats over five days.
The results followed the labels. The students who believed they were working with maze-bright animals recorded significantly better performance than the students who believed theirs were maze-dull. Because the labels stood for no difference in breeding, the authors looked for the source in how the sessions were run and recorded. Two informants who had watched most of the sessions found no false entries and five instances of a student prodding a rat, two in the bright group and three in the dull, a split the authors judged unlikely to explain the gap. The explanation they proposed was handling: students running bright rats described themselves as handling their animals more, and more gently, than students running dull ones.
Read the diagram and evidence as text
Matched rat silhouettes receive bright or dull tags before the same T-maze task. Score sheets carry opposite ordering chevrons for better versus lower recorded performance, with no numerical performance scale or invented maze routes. The sample labels are 12 experimenters and 60 rats. Handling as the causal explanation was proposed, not established.
Elevated T-maze, five days
The labels were false: the rats came from the same stock
Label given to the student: “Bred maze-bright” or “bred maze-dull”
Student expects a clever or a dull rat: 12 students, each running five rats
Handling as the causal route: Post-experiment self-reports suggested handling differences; mediation was not established
Recorded performance: Significantly better for the rats labeled bright
Different selective breeding: The 60 experimental rats were assigned bright/dull labels within matched groups
Label given to the student → assigned expectation → Student expects a clever or a dull rat
Student expects a clever or a dull rat → different recorded outcomes → Recorded performance
Recorded performance → proposed explanation, not established → Handling as the causal route (error transition)
| Condition | Value and unit | Evidence quote |
|---|---|---|
| Student experimenters, each given five rats | 12 students | Twelve of the thirteen students enrolled in a senior division course in experimental psychology served as Es. [Display conversion: 12 is the numeric transcription of Twelve.] |
| Rats, all from the same stock | 60 rats | N of Ss 5 30 30 [Display conversion: 30 + 30 = 60 experimental rats; the assistant’s separate five rats are not included.] |
Rosenthal went on to look for the same thing in classrooms. In the Pygmalion study with Lenore Jacobson, published in 1968, teachers were told that certain pupils, picked at random, were about to show an intellectual spurt, and the pupils named gained more on a later IQ test than their classmates. That study has been argued over since. Stephen Raudenbush's 1984 meta-analysis of the experiments that followed found the effect depended on timing: an expectation induced in teachers who had barely met the pupils produced a measurable gain, and one induced in teachers who already knew their class produced almost nothing. What the follow-up experiments support is that an induced expectation can move behavior when it has nothing to compete with, not that labels rewrite children.
The methodological response was blinding. Keeping the people who deliver and measure a treatment ignorant of the condition is now standard in trials. Schulz and colleagues examined 250 trials from 33 meta-analyses in 1995: those that were not double blind produced treatment-effect estimates about 17 percent larger than those that were, which is the clinical version of this page's problem. The same paper reports a separate and larger distortion from failing to conceal the allocation sequence, a different fault, in how cases are enrolled rather than in how they are handled and scored.
Everyday examples
The rows below are settings where one person's expectation can reach the person assessed, not findings from the studies cited here.
| Situation | What the bias does | What helps |
|---|---|---|
| A job interview after the hiring manager has read a glowing referral | The candidate gets warmer questions and more room to recover from a weak answer, and performs better for it | Same questions in the same order for every candidate, scored before the next one starts |
| A teacher told which pupils are in the top set | Harder questions and longer waits for an answer go to the pupils expected to manage them | Rotate who gets asked, and mark written work without names |
| A user test run by the designer who built the thing | The tone of the prompts steers participants past the parts that confuse them | A written script, and a moderator who did not build it |
| A clinician assessing response to a treatment they recommended | Ambiguous signs read as improvement, and the patient is asked whether they feel better rather than how they feel | Open questions, a standard measure, and a second rater who does not know the treatment |
| A manager checking whether a new process is working | The team hears which answer is wanted and gives it | Collect the measure before asking for opinions, and have someone outside the team collect it |
How to counter it
Resolve to be objective and the effect survives, since it runs below the level you can watch. What removes it is structure.
- Blind whoever handles the subject and whoever takes the measurement. If neither knows which condition a case is in, an expectation has nothing to attach to. This is the single most effective step.
- Script the interaction. Fixed wording, fixed order, fixed timing. Anything left to the moment is where expectation gets in.
- Write the scoring rule before you see the data: what counts as a borderline case, and what happens to it.
- Decide exclusions in advance. The animal, person or data point that disagrees with you is the one most likely to be dropped for a reason that sounds good later.
- Separate the roles. Whoever has a stake in the result should not collect it; use automated recording where you can.
- When blinding is impossible, say so and treat the result as weaker evidence. Naming the limit is better than claiming care removed it.
What it is not
Three nearby terms get used as synonyms. They differ in who holds the expectation and where it acts.
- It is not experimenter's bias, the wider set of ways a researcher's beliefs tilt a study, from the design and the exclusions to the analysis and what gets published. This effect is one path inside it, the one that runs through the subject.
- It is not expectation bias, where what a person expects changes what they themselves perceive and record. There the expectation stays with the observer; here it travels to the subject and changes their behavior.
- It is not the observer effect, the change in behavior that comes from knowing you are watched. That happens with a neutral observer who expects nothing, since it depends only on being observed.
Tests and tools that use this
Related biases
Observer-Expectancy Effect FAQ
What is the observer-expectancy effect?
It is the change in a study's result that comes from the researcher's expectations reaching the people or animals being studied. The researcher does not intend it and usually cannot report it: the expectation travels through handling, tone, emphasis and timing, and the subject's behavior shifts toward what is expected. It is also called the experimenter-expectancy effect or the Rosenthal effect.
How is it different from experimenter's bias?
Experimenter's bias is the whole family of ways a researcher's beliefs can tilt a study, including the design, the choice of measures, the handling of outliers, the analysis and the decision to publish. The observer-expectancy effect is one route within that family, the one in which the expectation reaches the subject and changes what the subject does, rather than changing what the researcher records.
What was the Clever Hans case?
Clever Hans was a horse that appeared to answer arithmetic questions by tapping a hoof. Careful testing showed he could only do it when the questioner knew the answer and was visible to him: he was reading the slight change in the questioner's posture that came as the correct number of taps approached. The questioners were sincere and unaware they were signaling, which is the pattern the effect describes.
How do you prevent the observer-expectancy effect?
Keep the expectation away from the measurement. Blind whoever runs or scores the study to the condition, use a fixed script and fixed timing, write the scoring and exclusion rules before you see the data, and automate recording where possible. Instructions to stay objective are not a safeguard: the students in the 1963 experiment described their own handling as gentler for the animals they had been told were bred for the task.
Was the Pygmalion classroom study real?
Rosenthal and Jacobson reported it in 1968, and it has been contested ever since. Raudenbush's 1984 meta-analysis of the follow-up experiments found that an induced expectation moved pupil IQ scores when teachers had barely met the pupils and had almost no effect once teachers already knew their class. Expectancy effects in classrooms are real but narrower and more conditional than the original coverage suggested.
How did the demo on this page score me?
It recorded the four decisions that let an expectation into a result: handling animals by feel rather than by a fixed routine, scoring an ambiguous trial in the direction you expected, removing or re-running the case that disagreed, and relying on effort or sample size instead of blinding the handling and the scoring. Expecting good scores from a strain described as bred for the task is not counted, because that is reading the sheet you were handed.
References
- Rosenthal, R., & Fode, K. L. (1963). The effect of experimenter bias on the performance of the albino rat. Behavioral Science, 8(3), 183-189. doi.org
- Rosenthal, R., & Jacobson, L. (1968). Pygmalion in the classroom. The Urban Review, 3(1), 16-20. doi.org
- Raudenbush, S. W. (1984). Magnitude of teacher expectancy effects on pupil IQ as a function of the credibility of expectancy induction: A synthesis of findings from 18 experiments. Journal of Educational Psychology, 76(1), 85-97. doi.org
- Schulz, K. F., Chalmers, I., Hayes, R. J., & Altman, D. G. (1995). Empirical evidence of bias: Dimensions of methodological quality associated with estimates of treatment effects in controlled trials. JAMA, 273(5), 408-412. doi.org
Cite this source
Psychology.com. (2026, October 23). Observer-Expectancy Effect. Psychology.com. https://psychology.com/biases/observer-expectancy-effect