What is Cognitive Bias?

A cognitive bias is a systematic pattern of deviation from rational judgment: not random error but a consistent tilt, produced by the mental shortcuts (heuristics) that make everyday thinking fast and mostly good enough. The modern research programme began with Amos Tversky and Daniel Kahneman's work in the 1970s, which showed that these deviations are shared, directional, and predictable, and that expertise offers thin protection. For research the implications run in three directions at once: biases shape the users being studied, the participants doing the answering, and the researchers doing the interpreting. The third is the one this field most needs to take personally.

The Biases That Do the Damage in Research

Confirmation bias. The flagship: seeking, noticing, and remembering evidence that fits existing belief. In research it writes leading questions, hears agreement in ambiguous answers, and reads every transcript as vindication. It is the reason methods have rules at all.

Anchoring. First numbers and first opinions drag everything after them: the first participant's reaction frames the next nine, the first estimate in a workshop becomes the range's centre of gravity.

Availability. The vivid and recent feel frequent: one dramatic session outweighs eight quiet ones, last week's angry ticket outweighs the silent satisfied majority. It is the enemy of honest prevalence claims and the reason counting beats recalling.

Framing effects. Logically equivalent presentations produce different judgments (90% success versus 10% failure), which governs both how questions should be written and how findings should be reported.

Recency and peak-end effects. Experiences get remembered by their peaks and endings rather than their averages, shaping what participants report about sessions and what stakeholders retain from readouts.

Survivorship and selection blindness. Reasoning only from who's present (current customers, completed responses) while the departed carry the real story: the cognitive companion to sampling bias and non-response bias.

Participant-side biases (social desirability, acquiescence, and their relatives) are covered under response bias; user-side biases are half of what fields like behavioural design study. The list above is the researcher's own occupational hazard sheet.

Defences That Actually Work

1. Decide before you look.
Pre-registering questions, metrics, and thresholds (even informally, in a dated document) is the single strongest anti-confirmation device: it separates hypothesis from wish while separation is still possible. The disciplines of honest significance testing are largely this.

2. Count instead of recalling.
Systematic coding across the full dataset, tallies of task outcomes, and structured severity ratings replace memory (availability's home turf) with procedure.

3. Seek the disconfirming case on purpose.
Interview the churned as well as the loyal, hunt the sessions that contradict the emerging story, and assign someone the explicit job of arguing the other reading of the data.

4. Use independent judgments before group ones.
Private estimates and separate first-pass analyses, compared afterwards, blunt anchoring; a second coder catches what one analyst's frame filtered out.

5. Let structure carry the discipline.
Neutral scripts, consistent tasks, and standardised instruments encode debiasing into the method so it works on tired days: a study template with balanced questions is confirmation-bias resistance you don't have to re-summon each time.

The Takeaway

Cognitive biases are not a character flaw to overcome but a permanent operating condition to engineer around: the researcher's mind is an instrument with known distortions. Decide in advance, count everything, hunt the disconfirming, judge independently, and let method structure do the remembering. The goal is not unbiased researchers (there are none) but studies built so the biases have nowhere to work.