{
  "slug": "selection-bias",
  "term": "Selection bias",
  "aliases": [],
  "definition": "Selection bias is systematic error caused by how units enter, remain in or are excluded from an analyzed sample relative to its target population.",
  "note": "Start with the target: a failure rate among currently observed drives and lifetime reliability among all installed drives ask different questions. A restricted sample need not be biased for its stated conditional target. Transferring that estimate to a different population requires justified observation and sampling assumptions. [cdc-interpretation] [schisterman-2013]\n\nSurvivor-only inclusion can omit failures before observation starts. Left-truncation describes that entry pattern; right-censoring retains a known unit with incomplete follow-up after entry. Neither term is a synonym for selection bias. Correct delayed-entry risk sets require assumptions about entry, and cannot identify survival before the earliest supported entry age from those observations alone. [schisterman-2013]\n\nMore records and narrower conventional confidence intervals do not remove an unmodelled selection mechanism. Record eligibility, entry times, exit reasons and missing-data exclusions; compare retained and excluded outcomes and exposure before extrapolating. A selection-adjusted analysis still needs defensible assumptions. [cdc-interpretation]",
  "status": "shipping",
  "advantages": [],
  "limitations": [],
  "related": [
    "left-truncation",
    "length-biased-sampling",
    "right-censoring",
    "confidence-interval"
  ],
  "supersedes": [],
  "supersededBy": [],
  "firstSeen": null,
  "lastSeen": null,
  "citations": [
    {
      "id": "cdc-interpretation",
      "title": "CDC Field Epidemiology Manual, chapter 8: confidence intervals and epidemiologic interpretation",
      "url": "https://www.cdc.gov/field-epi-manual/php/chapters/analyze-interpret-data.html",
      "accessed": "2026-11-19"
    },
    {
      "id": "schisterman-2013",
      "title": "Schisterman et al. (2013): Accuracy Loss Due to Selection Bias in Cohort Studies with Left Truncation (doi:10.1111/ppe.12073)",
      "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC6151356/",
      "accessed": "2026-11-19"
    }
  ],
  "provenance": [
    "The methodological sources concern epidemiologic inference; the drive examples apply those distinctions without estimating bias in a storage fleet."
  ],
  "updated": "2026-11-20"
}
