research methods
Research bias
What is research bias, and where does it creep in?
Research bias is any systematic influence that pushes a study's results away from the truth - not random error that averages out, but a consistent lean introduced by how you recruit, ask, observe or interpret. It's not a single mistake but a whole family of them, lurking at every stage of research, and the first defence is simply knowing where each kind tends to strike.
Also known as: research bias, bias in research, study bias
Prefer to watch?
Watch the recap 0:56
The demo
Bias isn't one slip - it's a family, with a different member waiting at each stage of a study. Click any to see where it strikes and what it does.
What this demo shows (text version)
An interactive map organises research biases by the stage of a study where they strike. Sampling: survivorship bias (your data only includes those who stuck around). Asking: leading questions, acquiescence bias (yes-saying) and social-desirability bias (answering to look good). Observing: the observer-expectancy effect (your expectations nudging participants) and the Hawthorne effect (people behaving differently because they're watched). Analysing: confirmation bias, hindsight bias and the false-consensus effect. Clicking any shows a one-line summary and links to its full entry.
The point is that bias is systematic, not random - it leans results consistently in one direction, so more data just makes you more confidently wrong - and it infiltrates every phase, not just one. You can't eliminate it, but you can anticipate it stage by stage and design against it with neutral methods, representative sampling, blinding, pre-registration and triangulation.
Bias isn't one error, it's a family that infiltrates every stage of research: who you study (sampling, survivorship), how you ask (leading questions, acquiescence, social desirability), how you observe (observer-expectancy, the Hawthorne effect), and how you interpret (confirmation, hindsight, false consensus). You can't eliminate it, but naming where each one strikes - and designing against it - is what separates research from elaborate self-deception.
Laid out by stage, the biases stop being a vague worry and become a checklist: this one attacks your sample, that one your questions, another your reading of the results. You can't remove them all, but you can see them coming - and a bias you've named is one you can design against.
Bias maps onto the research lifecycle. Recruiting and sampling: who's in your study (and who's silently missing - survivorship bias). Asking: how questions are worded (leading questions, acquiescence, social desirability). Observing: how your presence and expectations change things (the observer-expectancy and Hawthorne effects). Analysing: how you read the data (confirmation, hindsight and false-consensus bias). Each stage has its own traps.
What unites them is that they're systematic, not random. Random error scatters around the truth and shrinks with more data; bias pushes consistently in one direction, so a bigger biased study just gives you a more confident wrong answer. That's why you can't fix bias by collecting more - you fix it by changing the method.
You'll never eliminate it - the researcher is human and embedded in what they study - but you can manage it: neutral wording and scripts, representative and inclusive sampling, blinding where possible, pre-registered predictions, and triangulation across methods so one method's bias can't carry the conclusion. The goal isn't bias-free research; it's research whose biases you've anticipated and countered.