The tendency of researchers to subtly and inadvertently affect the behaviours…

Concept: In experimental research, a bias is any systematic influence that distorts the measured relationship between the independent and the dependent variable. Such biases are classified by where the distortion originates — with the investigator who administers the procedure, or with the participant who responds to it.

Application: The stem fixes the origin of the distortion at the investigator — researchers who subtly and inadvertently affect participants' behaviour — and fixes the consequence at the independent variable, whose true effect is obscured. An investigator's expectations reach a study through two unintended channels: interpersonal cues such as tone of voice, facial expression and differential prompting, which alter what the participant does; and selective observation, recording or scoring, which bends the record of what the participant did. Either channel means part of the difference that appears in the data is produced by the experimenter rather than by the manipulation, and the stem names the first of them. The umbrella term is experimenter bias, also called the experimenter expectancy effect, demonstrated by Rosenthal and Fode (1963), whose student handlers obtained faster maze learning from rats that were merely labelled “maze-bright”.

Contrast: the three other named phenomena locate the distortion elsewhere.

• Response bias — a broad category of systematic deviation between a respondent's reported answer and an accurate one, arising at the point of responding through acquiescence, extreme or midpoint responding, order effects and item wording.

• Social desirability bias — the respondent reports what is socially approved rather than what is actually the case, driven by self-presentation on sensitive items.

• Hawthorne effect — the participant changes behaviour because of awareness of being observed at all, not because of any cue that tracks the researcher's hypothesis.

Cross-check: the standard control confirms the classification. Experimenter bias is countered by a double-blind design and standardised, scripted protocols, which withhold condition information from the person who interacts with participants. A safeguard aimed at the researcher's knowledge only makes sense for a distortion that originates with the researcher.