Skip to content

Publication Bias

A scientific record becomes systematically unrepresentative when the probability that a study, result, or outcome becomes publicly available depends on its direction, magnitude, statistical significance, novelty, or sponsor-favoredness.

Core Idea

Publication bias arises when a result-bearing property—direction, magnitude, statistical significance, novelty, or sponsor favorability—changes whether conducted research becomes visible. The accessible literature is then a selected sample of the evidence that exists. A synthesis that treats the visible record as the conducted population can overstate effects or support conclusions that the full record would weaken.

Scope of Application

The abstraction applies to journal publication, conference acceptance, outcome reporting, indexing, sponsor-controlled disclosure, and other gates between conducted research and the record available to readers or meta-analysis.

Clarity

It separates the population of research that was done from the population that can be found. More published studies do not remove the problem when admission remains result-dependent.

Manages Complexity

The pattern reduces many dissemination failures to evidence population, result-bearing selector, visibility decision, accessible record, and downstream inference. Prospective registration and results-independent review target the gate; sensitivity models only estimate its past effects.

Abstract Reasoning

The accessible record is conditioned on visibility. If positive or significant results are more likely to cross the gate, their share in the record exceeds their share among all conducted results even when every published study is internally sound.

Knowledge Transfer

The logic of selection bias transfers directly, but the named abstraction stays within scientific dissemination. Researchers can distinguish prevention at the gate from retrospective diagnosis through funnel asymmetry or selection models.

Example

Twenty teams test the same weak effect. The two studies with unusually large positive estimates cross a significance threshold and are published; the others remain unavailable. A reader sees two apparently concordant positive studies rather than twenty noisy attempts.

Relationships to Other Abstractions

Local relationship map for Publication BiasParents appear above the current abstraction, mutual partners to the right, and children below. Node labels state whether each abstraction is prime or domain-specific; colors identify relation types.Publication BiasDOMAINPrime abstraction: Selection on Noisy Estimates — is part of, conditionalSelection onNoisy EstimatesPRIMEPrime abstraction: Selection Bias — is a kind ofSelection BiasPRIMEDomain-specific abstraction: File Drawer Problem — is a kind ofFile DrawerProblemDOMAIN

Current abstraction Publication Bias Domain-specific

Parents (2) — more general patterns this builds on

  • Publication Bias is a kind of Selection Bias Prime

    Publication bias is selection bias specialized to admission of scientific results into the visible literature.

  • Publication Bias is part of, conditional Selection on Noisy Estimates Prime

    Significance-filtered publication bias contains selection on noisy estimates when visibility depends on a threshold-crossing estimated effect.

    Condition / exception Publication selection can instead depend on direction, novelty, sponsor preference, or whole-study availability without selecting on estimate extremeness.

Children (1) — more specific cases that build on this

  • File Drawer Problem Domain-specific is a kind of Publication Bias

    Drawer Problem is the whole-study null-suppression species of the broader results-dependent publication and reporting filter.

Not to Be Confused With

  • File Drawer Problem is the whole-study, usually null-result suppression species.
  • Selective outcome reporting hides outcomes or analyses within a visible study.
  • Small-Study Effects and Funnel Plot Asymmetry are possible symptoms with multiple causes.
  • Selection Bias is the cross-domain genus and does not require a scientific record.

Notes

Initial canonical draft created from the mixed-DAG missing-node adjudication. Editorial re-authoring and citation verification are assigned in CHATGPT_2_CLAUD_TODO_LIST.