Glossary terms for 'B'
|Beta||When designing a study, the preset maximum probability of committing a type II error, that is, failing to reject the null hypothesis when it is false. Only meaningful in the context of an effect size (see below). For example, if an investigator specifies a beta of 0.20 (and alpha of 0.05), she would need about 25,000 subjects per group followed for 10 years to show that daily aspirin halves the risk of colon cancer. Put another way, if aspirin actually had exactly that effect, her study of 25,000 per group would have a 20% chance of failing to reject the null hypothesis of no difference (at alpha = 0.05). See also power.|
|Between-groups design||A study design that compares the characteristics or outcomes of subjects in two (or more) different groups. For example, the investigator used a between-groups design to compare in-hospital mortality rates among patients treated in intensive care units that had round-the-clock intensivists with those among patients treated in units that used electronic monitoring of patients. See also within-group design. |
|Bias||A systematic error in a measurement, or in an estimated association, due to a shortcoming in a study?s design, execution, or analysis. For example, due to a bias in the way subjects remembered their exposure to toxic chemicals, patients with leukemia were more likely to report use of insecticides than were controls.|
|Blinding||The process of ensuring that subjects and/or investigators are unaware of the group (e.g., intervention or control) to which subjects are assigned, usually in the context of a randomized trial. For example, by using identical placebo pills and keeping the list of subject assignments off-site, neither the subjects nor the investigators (including research assistants) were aware which subjects were treated with the active medication. Also called masking, especially in ophthalmologic studies.|
|Blocked randomization||A method of assigning subjects to a particular intervention in blocks (groups) of a pre-specified size (e.g., four or six) to ensure that similar numbers of subjects are assigned to the intervention and control groups. Often used in multi-center studies in which the investigators want the total numbers of intervention and control subjects to be similar at each site. For example, patients within each clinic were randomly assigned to either the treatment or control groups in blocks of six, ensuring that the number of subjects per group would differ by no more than 3. See stratified blocked randomization.|
|Bonferroni correction||A technique to prevent type I errors by reducing alpha in a study to account for testing multiple hypotheses. For example, because the investigators were testing four different hypotheses, they used Bonferroni correction to reduce alpha from 0.05 to 0.0125.|
Glossary material from Hulley SB et al. Designing Clinical Research, 4th ed. Philadelphia, Lippincott Williams & Wilkins, 2013.