Selection bias in observational and experimental studies

Stat Med. 1994;13(5-7):557-67. doi: 10.1002/sim.4780130518.

Abstract

There has been a heightened awareness of the dangers of selection bias over the past two decades. Certainly coverage in statistical and 'statistics for medicine', and epidemiology textbooks have allocated pages to warn investigators and readers of investigations to be aware of its presence. The scientific community has not, however, yet accepted the necessity for critical assessment of the method of sample selection in the planning and execution of studies as a fundamental underpinning of observational and experimental studies. To wit, we are faced with a plethora of research studies receiving funding, being published in peer-reviewed journals and influencing future studies, that may be reporting entirely spurious associations. It is the intent of this paper to present examples of selection bias in a variety of areas which have resulted in misleading or entirely incorrect results. We hope to help make such research scientifically 'politically incorrect' to the degree that the scientific community 'just says no' to such studies, either proposed or reported.

Publication types

  • Review

MeSH terms

  • Clinical Trials as Topic / statistics & numerical data*
  • Cohort Studies*
  • Humans
  • Patient Dropouts / statistics & numerical data
  • Randomized Controlled Trials as Topic / statistics & numerical data*
  • Selection Bias*
  • Treatment Outcome