| Astrophysics (Index) | About |
The term selection bias refers to a type of problem in the selection of a sample (i.e., set) of cases or objects from which statistical inferences will be drawn, i.e., some means by which the set lacks some randomization aspect that a stated inference depends upon. A number of types of inherent selection-bias mechanisms have been identified as relevant to astronomical studies using information from surveys, including:
Volume weighting is a method of compensating for some such biases.
Other selection biases can occur throughout science, and careful scientists generally look for selection bias (and for documentation of what has been done to compensate for it) in any study-report they face. One ubiquitous psychological phenomenon is a tendency to notice (and use) evidence that confirms something you already believe (confirmation bias or expectation bias), a reason that more careful statistical confirmation is often demanded before accepting anecdotal evidence. Another common issue is surveys collected from people who volunteer or who are able to avoid participating, from which inferences are drawn as if the group was a random set of people: those who participate and those who don't can differ statistically. While these can apply to astronomy, the above-mentioned types of bias refer to means by which seemingly-uniform collections of raw observation data can be non-representative.
The term sampling bias (or sampling error) is also commonly used for such cases of selection bias, but specifically implies some kind of sampling is taking place: for example, cases where the above-described flaws occur during the selection of a subset of some larger set, to be used to derived statistics applicable to the larger set. The term selection bias is also applicable to cases where such sampling is not involved, such as an attempt to produce statistics about all Harvard graduates, but using such a flawed method when gathering data intended to include every Harvard graduate.