A researcher says “great” when a participant finds the intended button, then falls silent when they choose another route. Neither reaction appears in the task script, but both may teach the participant what the session rewards. Moderator bias is systematic influence from the researcher’s conduct that affects behaviour, answers or their interpretation.
It can enter through wording, tone, follow-up questions, assistance and selective attention. Good intentions do not remove it. The practical response is to examine what the moderator contributes to the session and build procedures that make consequential influence easier to notice.
Inspect the question for an answer already supplied
“Did the new layout make that easier?” introduces both a comparison and a favourable interpretation. “How did you go about that task?” leaves room for an account that may not concern the layout at all. Nielsen Norman Group’s guidance on leading questions explains how a prompt can supply the answer the researcher then appears to discover.
Follow-up questions deserve the same review as the prepared script. Reflecting a participant’s own phrase can help clarify meaning, but replacing it with the team’s vocabulary can change the account. An interview guide supports consistency while leaving room to investigate unexpected details.
Neutrality does not require an unfriendly performance. A researcher can acknowledge the contribution, show care and clarify practical arrangements without praising a particular answer or defending the product.
Decide how assistance will be handled
In a fictional test of an appointment system, a participant cannot find how to change a booking. A hint may allow the session to reach the next screen, but the completed task is now an assisted attempt. Record the intervention and its timing rather than counting the outcome as independent success.
Agree assistance rules before the study, including technical problems, distress and the point at which continuing would no longer serve the research. Apply those rules consistently enough for the intended comparison. The right rule depends on whether the study is exploring a later interaction or measuring unassisted performance.
Where several researchers moderate, rehearse together and review a pilot. Differences in phrasing and help can otherwise become differences attributed to the design.
Make the moderator visible during analysis
Record the questions around a finding, not only the participant’s answer. A quotation following a strongly leading prompt needs a different interpretation from the same statement offered spontaneously. Observers can note interventions, and recordings can help the team examine moments that felt unremarkable at the time.
Where feasible, keep the moderator unaware of the team’s preferred version or use neutral labels for alternatives. This can reduce a source of expectation without making the whole study blind to every relevant detail.
An unmoderated format may suit tasks that do not need live probing, but its instructions still carry the researcher’s choices. Removing the person from the session does not remove the need to review the study design. The aim is credible evidence with an explicit account of how it was collected, rather than a claim that human influence has disappeared.