ReferenzklassenSchätzung

Referenzklasse, or reference class, is a concept used in statistics, risk assessment, and decision theory to denote the group of cases to which a particular instance is compared when estimating probability. The reference class provides a baseline, a source of empirical data from which a prior or an empirical estimate is drawn.

Choosing a reference class is often essential but contentious. It affects the resulting probability, because similar objects or events can have different observed frequencies depending on the scope and relevance of the class. In Bayesian reasoning, the reference class helps set priors; in reliability engineering and epidemiology, it conveys expected rates or incidences for comparison.

The reference class problem arises when there is no unique or objectively correct class. If the class is too broad, information is lost; if too narrow, the sample may be too small to be reliable. Practitioners attempt to justify the selection using domain knowledge, relevance to the decision context, and sensitivity analyses to show how conclusions change under alternate choices. Modern approaches include hierarchical models that borrow strength across related groups to reduce dependence on a single class.

Example: estimating the probability of a rare adverse event for a new drug. Possible reference classes include patients in the same trial, patients of the same age and sex with similar comorbidities, or patients treated with related medications. Each choice yields different risk estimates, illustrating the problem.

Critics point out that reference-class decisions can introduce bias and circular reasoning if the class is chosen to fit desired conclusions. Transparent reporting and explicit exploration of alternative reference classes are recommended. Related concepts include prior probability, base rate, and Bayesian inference.