Direct comparison
Bias Vs Confounding: Key Differences & Comparison | CASRAI
Bias and confounding both distort study findings, but they arise differently. Bias is a systematic error introduced by how a study is designed, conducted, or measured. Confounding is a specific distortion caused by a third variable that is linked to both the exposure and the outcome, mixing their effects together.
Side-by-side comparison
| Dimension | Bias | Confounding |
|---|---|---|
| What it is | Systematic error from study design, conduct, or measurement | Distortion from a third variable mixing effects |
| Source | How participants are selected or data are gathered | A variable linked to both exposure and outcome |
| Nature of error | A flaw introduced by the researcher or process | A real but extraneous variable in the data |
| Common types | Selection bias, information/measurement bias, recall bias | A single class — confounding by a third factor |
| Fixable in analysis? | Usually not — must be prevented by design | Often yes — by adjustment or stratification |
| Prevented by | Careful sampling, blinding, standardised measurement | Randomisation, matching, restriction |
| Direction | Consistent, systematic departure from the truth | Can exaggerate, mask, or reverse an association |
| Requirement | Arises from procedure, not a specific variable | Requires a variable tied to exposure and outcome |
| Example | Volunteers differ systematically from non-volunteers | Coffee–cancer link confounded by smoking |
Common questions
FAQ
What is the key difference between bias and confounding?+
Bias is a systematic error created by how a study is designed, conducted, or measured — a flaw in the process. Confounding is a distortion caused by a specific third variable that is associated with both the exposure and the outcome, mixing their effects. In short, bias is about method; confounding is about an extraneous variable.
Can confounding be corrected after the data are collected?+
Often yes, provided the confounder was measured. Techniques such as stratification and multivariable adjustment can statistically control for a known confounder. Most forms of bias, by contrast, are baked into the data by flawed design or measurement and cannot be analysed away — they must be prevented up front.
How does randomisation help with confounding but not bias?+
Randomisation distributes both known and unknown confounders evenly across groups by chance, so on average it removes confounding — a major reason randomised trials are powerful. It does not, however, fix biases introduced later, such as measurement bias or loss to follow-up, which still require careful conduct, blinding, and standardised procedures.
Going deeper
