Common Types of Bias
Core summary
Research is vulnerable to many specific bias types. The most important to know are selection bias, information bias, recall bias, observer bias, attrition bias, and publication bias. Each has specific causes and prevention strategies.
Detailed explanation
Detailed explanation
Understanding specific bias types helps you both design better studies and critically appraise published research. Selection bias occurs when participants are not representative of the target population, or when groups being compared are systematically different. Subtypes include healthy user bias (people who choose an intervention tend to be healthier), Berkson's bias (hospital-based studies select sicker patients), and volunteer bias (study volunteers differ from the general population). Prevention: random sampling, clear eligibility criteria, and randomization. Information (measurement) bias occurs when data are collected or measured inaccurately or differently between groups. Subtypes include recall bias (participants remember past exposures differently — common in case-control studies), observer bias (researchers assess outcomes differently based on group assignment), and social desirability bias (participants give answers they think are expected). Prevention: blinding, standardized instruments, objective outcomes. Recall bias specifically refers to participants remembering or reporting past events differently depending on their current health status. Patients with disease tend to recall exposures more thoroughly than healthy controls. Prevention: use prospective designs or medical records instead of patient recall. Observer (detection/ascertainment) bias occurs when researchers assess or detect outcomes differently between groups. Prevention: blinding outcome assessors to group assignment. Attrition bias occurs when participants drop out differentially between groups. If sicker patients leave the treatment group, the remaining participants look healthier — falsely inflating effectiveness. Prevention: intention-to-treat analysis, minimizing dropout. Publication bias occurs when studies with positive results are more likely to be published than negative studies, distorting the overall evidence. Prevention: study registration, including unpublished studies in reviews, funnel plots. Lead-time bias and length-time bias are specific to screening studies. Lead-time bias makes screened patients appear to survive longer simply because their disease was detected earlier, not because screening changed the outcome.
Clinical example
In a case-control study of smartphone use and brain tumors, patients with tumors might recall their phone use more carefully than healthy controls (recall bias), and interviewers might probe tumor patients more thoroughly (observer bias). Both biases could exaggerate the apparent association.
Research example
The 'file drawer problem' — a form of publication bias — was illustrated in antidepressant trials. A meta-analysis of FDA-submitted data (which included unpublished negative trials) showed much smaller effects than published literature alone suggested. The positive results were published; the negative ones stayed in file drawers.
Knowledge check
Q1. Which bias is MOST likely in a case-control study that relies on participant interviews?
Q2. What is the BEST way to prevent observer bias?
Q3. Publication bias is detected using: