A bias created when using data from a sample that differs systematically in its characteristics from the general population owing to a feature of the selection process. Such a bias will enter, for example, if the selection rule is 'take those whose names start with A', or 'those who live on the corner', or 'those who respond to a mailed questionnaire'. It also refers to the selection by patients or physicians of treatments that they believe will confer the greater benefits. It may also arise as a direct form of scientific fraud. Blinding is no safeguard against this form of bias. Measures taken to prevent it include taking the appropriate steps to minimize specific forms of it (such as volunteer bias), to require full disclosure of scientific procedures and possible conflict of interest, or to correct for it by multiplying an observed odds ratio by the inverse of the 'selection odds ratio'. Cf. Randomizing, Sample Selection Bias, Sample Size.
Was this article helpful?