Selection, Information and Confounding Bias: Sources of Systematic Error in Research

Selection bias arises from how participants are chosen or retained in a study. Information bias arises from how exposure or outcome data is measured or recorded. Confounding arises from a third variable associated with both exposure and outcome. All three can distort a result even when the study is conducted honestly and carefully in every other respect.

Selection bias and the composition of the study sample

Pannucci and Wilkins describe selection bias as arising when the study population, or the subgroup analysed, is not representative of the population to which the conclusions are intended to apply, for example through non-random recruitment, differential loss to follow-up between groups, or volunteer bias, in which those who choose to participate differ systematically from those who do not1. Once introduced, selection bias is often difficult to correct in analysis and is better addressed through study design, for example by minimising differential attrition and reporting flow diagrams transparently.

Information bias and the measurement of data

Information bias (also called measurement or observation bias) arises from how exposure or outcome data are collected, and includes recall bias (participants in one group remembering past exposures differently, often because of their outcome) and observer or detection bias (an unblinded assessor measuring or interpreting outcomes differently depending on known group assignment)1. Standardised, ideally blinded, data-collection procedures are the primary defence against this bias.

Confounding by a third variable

Confounding occurs when a third variable is associated with both the exposure and the outcome, without being on the causal pathway between them, creating a spurious or distorted association. Pannucci and Wilkins note that, unlike selection and information bias, confounding can often be addressed after data collection through statistical adjustment (stratification, multivariable regression or matching), provided that the confounder was measured; an unmeasured confounder remains a limitation that cannot be adjusted for, only acknowledged1.

Key points

Practical application

Frequently asked questions

@misc{selection_information_confounding_bias_2026,
  author       = {{Pubrica Academy}},
  title        = {Selection, Information and Confounding Bias: Sources of Systematic Error in Research},
  howpublished = {Pubrica},
  year         = {2026},
  month        = {oct},
  url          = {https://pubrica.com/academy/concepts-definitions/selection-information-confounding-bias/},
  note         = {Accessed: <span data-cite-acc="iso">2026-10-15</span>}
}

[ ] Recruitment and retention reported with a flow diagram
[ ] Data-collection procedures standardised and, where possible, blinded
[ ] Plausible confounders identified in advance and measured
[ ] Adjusted and unadjusted results both reported
[ ] Unmeasured confounders acknowledged explicitly as a limitation