Selection bias is an error with the methodologies behind recruiting and retaining participants in studies, or analyzing the data obtained, that makes the results less reliable. It is one of a number of biases that can discolor a study if the researchers do not anticipate them and take steps to avoid them. In a writeup of a sound scientific study, researchers can discuss all the methods used to allow readers to judge whether biases may have tainted the results.
One example of selection bias is a sampling bias, where the candidates for a study are not chosen randomly, which would tend to skew the data. A truly random sampling method pulls in a broad assortment of people from the target population to avoid problems that might arise with a narrow sample, like false correlations that are actually the result of who participated, rather than what is being studied. For example, recruiting for a study on pet health that focuses on veterinary offices would create a sampling bias, because people with healthy pets would not be recruited.
A selection bias can also come into play with retention. Over the course of a study, especially a long one, a certain amount of attrition tends to occur as people drop out or become ineligible for various reasons. If this rate is high, it can skew the final results by narrowing the sample and making it less random. If a study does not have adequate measures in place to encourage participants to see it through to the end, it might have a selection bias problem.
Stopping a trial early can interfere with the time interval and may create false or misleading data. Likewise, not controlling data adequately, and using poor methods of statistical analysis, can create a selection bias. Researchers may also confuse cause and effect, create false correlations, or otherwise misinterpret study results. If they analyze the data in a way that confirms the false conclusions, their end results may be less valuable.
Some degree of bias can be difficult to avoid with scientific research. Before a project starts, researchers may sit down to discuss possible biases and ways for dealing with them, so they can plan ahead to address issues like selection bias. They monitor the study as it occurs to check for signs of emerging bias and are careful about how they evaluate and discuss data. Peer review is an important part of this process, as it permits input from third parties who are less likely to have an interest in the results, and thus can be honest in their assessments of the validity of a project.