A topic in the Open Knowledge Graph — a free, open map of 15,290 topics and the order to learn them in.
Experimenter expectations and subtle behavioral cues can influence participant responses, creating self-fulfilling prophecies. Demand characteristics signal the study's hypotheses, allowing participants to adjust responses toward expected outcomes. Blinding experimenters to conditions and hypotheses, standardizing procedures rigorously, and using objective measures reduce these threats, though complete elimination is impossible.
In an ideal experiment, the researcher is a passive observer: conditions are assigned randomly, treatments are administered identically, and outcomes are measured objectively. But researchers are human, and humans with expectations behave differently than humans without them — often unconsciously. Experimenter bias is the systematic distortion of results that occurs when a researcher's hypotheses, hopes, or beliefs influence how an experiment is conducted, observed, or recorded. The defining characteristic is that the bias operates through subtle, often unintentional channels, not through deliberate fraud.
The canonical demonstration is Robert Rosenthal's Pygmalion experiment (1968), in which schoolteachers were told that certain students had been identified by a test as "late bloomers" likely to show exceptional intellectual growth. In reality, these students were chosen randomly. Yet at the end of the year, the "late bloomers" showed significantly greater IQ gains — because teachers interacted with them differently, providing more warmth, more challenging material, and more positive feedback, all without realizing they were doing so. The expectation became self-fulfilling. Rosenthal also demonstrated experimenter expectancy effects in laboratory rat studies: experimenters told their rats were "bright" got better maze performance than those told their rats were "dull," even though the rats were randomly assigned. The experimenter's expectations somehow transmitted to the animal through handling differences too subtle to observe directly.
The participant-side counterpart is demand characteristics — cues in the experimental setting that signal to participants what the study is "about" and what response seems expected or appropriate. Participants are not passive: they arrive with social motivations, they observe the setup, listen to instructions, and draw inferences. If an experiment obviously pairs an aggressive film with a measurement of hostility, many participants will guess the hypothesis and may either confirm it (to be cooperative) or disconfirm it (to resist being manipulated). Either way, demand characteristics threaten internal validity. From your prerequisite on blinding, you know that the standard solution is single-blind (participants unaware of condition assignment) and double-blind (both participants and experimenters unaware) designs. Double-blinding targets both threats simultaneously: experimenters who do not know which condition a participant is in cannot transmit differential expectations, and participants who do not know their condition cannot play a role relative to it.
Several additional controls reduce these threats. Standardized protocols — scripted instructions, computerized administration, pre-recorded stimuli — remove the experimenter's moment-to-moment behavioral variability. Objective outcome measures (reaction time, physiological recordings, behavioral observation with coded video) are harder to bias than subjective ratings made by someone who knows the hypothesis. Pre-registration — publicly posting hypotheses and analysis plans before data collection — prevents post-hoc reinterpretation of results. None of these controls is individually sufficient; robustness against experimenter effects comes from layering them. Even then, complete elimination is impossible, which is why replication by independent labs with no stake in the original finding remains the gold standard for establishing a result's reliability.