DataTerminology.comNEW
A bias arising when sample selection is not representative of the target population.
DESCRIPTION
Selection bias occurs when the dataset or experimental sample systematically differs from the intended population, skewing estimates and conclusions. It undermines external validity and model fairness. Causes include nonresponse, self-selection, and survivorship. Mitigation uses randomized sampling, stratification, weighting, and careful inclusion criteria. Transparent reporting and audit are critical in AI systems.
ALTERNATIVE DEFINITIONS
Encyclopaedia Britannica
Selection bias, distortion of statistical analysis results caused by the method of collecting samples.
Frames bias as distortion originating from sample collection methods.
Investopedia
Selection bias is the distortion of a statistical analysis resulting from the method of collecting samples.
Emphasizes distortion due to sample collection and analysis.
CDC
Selection bias occurs when participants included in a study differ systematically from the target population, leading to erroneous conclusions.
Highlights participant differences and erroneous inference in epidemiology.
RELATED TERMS
Sampling Bias Survivorship Bias Confounding Stratified Sampling Superclass External Validity Nonresponse Bias Causal Inference A/B Testing
COMPETING TERMS
RELATIONS
- Broader: Bias
KNOWLEDGE AREAS
Data Science AI Governance Testing & Validation Mathematics
TARGET AUDIENCE
Data Scientist AI Engineer MLOps Engineer Data Analyst AI Product Manager Chief AI Officer
EXTERNAL SOURCES
- Wikipedia - Selection bias
- Centers for Disease Control and Prevention - Epidemiology lesson on selection bias
- Stanford Encyclopedia of Philosophy - Bias effects in probabilistic causation
- Nature Human Behaviour - Article discussing sampling bias impacts
TOP QUESTIONS
Why is Selection Bias harmful in AI?
It skews model training, reduces fairness, and degrades generalization to real-world populations.
Category: Relevancy
What are common sources of Selection Bias?
Self-selection, nonresponse, convenience sampling, and survivorship are typical causes.
Category: Overview
How can Selection Bias be detected?
Compare sample demographics to population, analyze response patterns, and inspect covariate distributions.
Category: Testing
How to mitigate Selection Bias?
Use random/stratified sampling, weighting, oversampling underrepresented groups, and resurvey strategies.
Category: Requirements
What risks arise when ignoring Selection Bias?
Models underperform, mislead decisions, and can violate fairness regulations.
Category: Risk
What is a competing concept to Selection Bias?
Random sampling aims to produce representative samples that minimize bias.
Category: Alternatives
What hot topics relate to Selection Bias today?
Fairness auditing, demographic parity metrics, and causal adjustments are prominent.
Category: Topic
Where to start learning about Selection Bias?
Begin with epidemiology and statistics tutorials, then explore fairness literature in ML.
Category: Example
How is Selection Bias managed operationally?
Govern sampling processes, track representativeness, and document risks in model cards.
Category: Management