bias
Meaning of bias: Adding participants fixes random error but never bias, so a very large flawed study can be more confidently wrong than a small one.
Definition of bias
Bias is a systematic flaw in how a study is designed, conducted, analyzed or reported that pushes the result away from the truth in one consistent direction. It differs from random error, which scatters measurements unpredictably and shrinks as a study grows. Bias does not shrink with size, so a larger study carrying the same flaw simply produces a more precise wrong answer.
Where does bias enter a study?
| Type | How it arises |
|---|---|
| Selection bias | The people studied differ systematically from the population the result is applied to |
| Recall bias | People with an illness remember past exposures more thoroughly than people without it |
| Observer bias | Knowing the group assignment shifts how an outcome is judged or recorded |
| Attrition bias | Those who drop out differ from those who remain, distorting the group left behind |
| Publication bias | Studies with striking results are published more readily than those finding nothing |
Confounding sits close by and is often listed alongside these, though strictly it is a mixing of effects rather than a flaw in method: a third factor linked to both the exposure and the outcome creates an association that is real in the data and false as an explanation.
What reduces it?
Study design is the main defense. Random assignment, as in a randomized clinical trial, distributes both known and unknown differences evenly across groups, which no amount of statistical adjustment afterward can reliably imitate. Randomization works only if the upcoming assignment cannot be foreseen, so the allocation sequence is concealed from the person enrolling participants. Blinding keeps participants, clinicians and outcome assessors unaware of who received what, which blocks observer bias and expectation effects. Analyzing everyone in the group they were assigned to, whatever they actually received, blocks a further distortion.
Why does the direction matter?
Random error is symmetrical, so it widens the uncertainty around an estimate without moving its center. Bias moves the center. This is why selection bias in a study of a treatment given preferentially to healthier people can make an ineffective treatment look beneficial, and why publication bias inflates the apparent effect of everything in a field at once. Since bias cannot be measured directly from the data, appraisal focuses on the methods described rather than on the size of the result.
Used in a sentence
The study's design included measures intended to reduce the risk of selection bias among enrolled participants.
Good to know
Bias, a systematic distortion of results, is distinguished from random error, which varies unpredictably between measurements and tends to average out in a sufficiently large study rather than pushing results in one consistent direction.