Bias is a term that is commonly referred to in epidemiological studies. It is a technical term and does not imply a partisan desire for or attempt to produce a particular outcome.
Bias is simply any factor that can distort the outcome of epidemiological work from its true value. There are a number of types of bias to consider:
• failure to record or identify factors (confounders) that could result in the same effect or prevent the effect of the causal hypothesis under study.
• inappropriate selection of study population.
• diagnostic and exposure measurement techniques can be under or over sensitive, under or over specific and plain wrong.
• measurements may be systematically biased. For example, an observer may improve in the accuracy of his observations with practice. If more cases than controls are observed at
the beginning of the study, the results could be biased.
Bias is particularly likely in studies that rely on exposure memory and/or self reported symptoms for diagnosis.
In some cases the inevitable random errors in study data have the effect of biasing the outcome away from the true measure of association of exposure and disease. Random error in dichotomous measures (i.e. Some or no exposure or, some or no disease) should bias the result toward the null (RR=1). More complex exposure and outcome measures have unpredictable dependencies on random error.
Statistical methods can be applied to check the sensitivity of the reported outcome to random error in model variables.