Bias

Error happens. Scales are set incorrectly. Memories fade. Study participants try to give the ‘right’ answer. When these mistakes are systematic, that is, not random, they will likely cause a biased result. The term bias can be translated fairly as ‘wrong’ with the additional refinement of ‘wrong due to systematic error.’ Thus, in the general statistical and scientific languages, a biased estimate is an incorrect estimate. In epidemiology, biased estimates typically refer to the distortion of a measure of association between exposure and outcome. This entry describes these measures, including the rate ratio, relative risk, attributable risk, and odds ratio.

The following examples will help illustrate bias. Underestimation would be caused by a scale that always weighs people 10 lb less than their true weight. A survey with leading questions (e.g., ‘Do you believe that smoking is bad?’) may draw the desired answers more frequently than the study population actually believes. Study participants with a condition linked to the research may be more likely to follow study procedures or be available for follow-up than participants without the condition.

Causes of biased estimates in epidemiologic studies are generally categorized as selection bias and information bias. Selection bias refers to ways an estimate may be incorrect due to how participants were enrolled in a study or dropped out of a study (e.g., lost to follow-up). Information bias refers to ways an estimate may be incorrect due to measures made during the study. More than 100 terms for information bias exist, many with similar or overlapping meaning. This entry describes common forms of bias that affect estimates in epidemiologic studies. Confounding can also bias measures of association and is discussed in a separate entry.