### Glossary terms for 'O'

Observational study | A general term for a research design in which the investigators simply observe the subjects without making any interventions. Thus this term includes cross-sectional, case-control, and cohort studies, but not randomized trials or before-after studies. For example, the examiners performed an observational study to determine the risk factors for melanoma. |

Observer bias | The situation in which an investigator (or research assistant) makes an non-objective assessment that is affected by her knowledge of one or more of the subject?s attributes, such as whether the subject is a case or control, or was exposed or not exposed to a particular risk factor. For example, observer bias was apparently responsible for the finding that, based on an interview, Hispanic teenagers were more likely to be characterized as having issues with anger management than Asians, since a self-administered survey and a review of school records found no differences between the two groups. |

Odds | The risk of a disease (or other outcome) divided by 1 ? risk. For example, if the lifetime risk of breast cancer among women is 15%, then the lifetime odds of developing breast cancer are 0.18 (0.15/0.85). Risk (see below) and odds are similar for rare diseases. |

Odds ratio | The ratio of the odds of a disease (or other outcome) in those exposed to a risk factor to the odds of that disease in those not exposed. For example, the odds ratio for renal failure among those with hypertension is 2.0, meaning that hypertensive patients are about twice as likely to develop renal failure as those who are not hypertensive. The risk ratio (see below) and the odds ratio are similar when a disease is rare in both the exposed and the unexposed, because the odds and the risks of the disease are similar. |

One-sample t test | A statistical test used to compare the mean value of a variable in a sample to a fixed constant (a particular number). The most common type of one-sample t-test is a paired t-test, in which the sample mean for the difference between paired measurements (e.g., on the same subject at different points in time)is compared with zero. For example, the investigators found that men gained a mean (? SD) of 4 ? 3 kg in weight during their residencies (P = 0.03, by one-sample t test). See also two-sample t-test. |

One-sided hypothesis | An alternative hypothesis (see above) in which the investigator is interested in evaluating the possibility of committing a Type I error (see below) in only one of the two possible directions (e.g., greater or lesser risk, but not both). For example, the investigator tested the one-sided hypothesis that smoking was associated with an increased risk of dementia. See also two-sided hypothesis. |

Ordinal variable | A categorical variable whose values have a logical order. For example, current alcohol use was treated as an ordinal variable: The values were no alcohol consumption, 1 or 2 drinks per week, > 2 but < 7 drinks per week, 1 - 2 drinks per day, and = 3 more drinks per day. See also nominal variable. |

Outcome | A general term for the endpoint(s) of a study, such as death or the occurrence of a disease. For example, in a study of whether radiosurgery was beneficial for patients with solitary brain metastasis, subjects were followed for the outcomes of death or placement in a skilled nursing facility. See also predictor. |

Outcome variable | The formal definition of the outcome for each subject. For example, in a study of the effects of different types of exercise on bodyweight and body composition, the outcome variables were defined as the change in weight in kg from baseline to the final measurement after one year, and the change in waist circumference in cm during that same time period. |

Overmatching | The situation in which matching beyond that necessary to control for confounding reduces the ability of the investigator to determine whether a risk factor is associated with an outcome because the controls have become too similar to the cases. For example, because the controls were matched to cases by age (? 3 years), sex, race, and socioeconomic status, overmatching made it impossible to determine whether education was associated with the risk of stroke among subjects ages 65 years and older, since the matching variables are major determinants of education in that age group. |

Glossary material from Hulley SB et al. *Designing Clinical Research*, 4th ed. Philadelphia, Lippincott Williams & Wilkins, 2013.