What Is Bias?
Core summary
Bias is a systematic error in the design, conduct, or analysis of a study that produces results that are consistently different from the truth. Unlike random error, bias cannot be reduced by increasing sample size.
Detailed explanation
Detailed explanation
In research, bias refers to any systematic deviation from the truth. The key word is 'systematic' — bias pushes results in one direction consistently, unlike random error which scatters results unpredictably around the true value. Random error decreases as sample size increases (the law of large numbers). Bias does not. If your scale is consistently 2 kg too heavy, measuring more people does not fix the problem — every measurement is still 2 kg too high. This is why preventing bias through careful study design is far more important than simply enrolling more participants. Bias can enter a study at any stage: during participant selection (selection bias), during data collection (information/measurement bias), during analysis (analytical bias), or during reporting (publication bias). The direction and magnitude of bias are often unpredictable, making prevention through good design the primary strategy. Importantly, bias is not the same as confounding (covered in an earlier lesson). Confounding involves a third variable that distorts the association. Bias involves flaws in how the study was designed, conducted, or analyzed. Both are threats to validity, but they require different solutions.
Clinical example
A hospital survey about patient satisfaction is distributed only to patients who returned for follow-up visits. Patients who had bad experiences and did not return are excluded — systematically skewing results toward higher satisfaction (selection bias). Adding more responding patients does not fix this — the bias is in who was sampled.
Research example
The Nurses' Health Study initially suggested that HRT reduced heart disease. This finding was influenced by selection bias (healthy user bias) — women who chose HRT were systematically healthier than those who did not. The randomized WHI trial, which eliminated this selection bias through randomization, reached the opposite conclusion.
Knowledge check
Q1. Bias in research is BEST defined as:
Q2. Increasing sample size can eliminate bias.
Q3. A crooked ruler always gives wrong measurements. This is analogous to: