An economics and business concept defining a measure, method, or organizational practice used for analysis and decision-making. It specifies how information is generated or used to guide allocation of resources and evaluation of outcomes. It does not ensure correctness without clear assumptions, reliable inputs, and appropriate review of results. It materially affects planning, performance, and risk by shaping decisions and incentives within organizations and markets. The concept is generally stable, though methods and tools evolve over time.

Principle

When the mechanism that determines whether an observation appears in the data correlates with variables of interest, naive estimates will be biased unless the selection process is modeled or properly adjusted for.

Demonstration

In hiring analysis, using only currently employed workers underestimates turnover risk because those who left are excluded (survivorship bias); in credit risk modeling, training only on accepted loan applicants causes acceptance bias that distorts default rate estimates in the general applicant pool.

Misapplication

Ignoring the selection mechanism or attempting to correct selection by conditioning on a collider; using post-stratification weights without correcting measurement or nonresponse mechanisms leads to residual bias.

Consequence

If unaddressed, selection bias produces misleading estimates and poor decisions—e.g., overestimating treatment effectiveness or underestimating risk—thus misallocating resources or misinforming policy.

Reversal

A random or properly weighted sample representative of the target population eliminates selection bias for estimands defined on that population; well-designed experiments that randomize inclusion avoid many selection issues.

Boundary

Selection bias refers specifically to biases stemming from the sampling or inclusion mechanism; it is distinct from measurement error, confounding due to omitted variables, and model misspecification, though these problems can coexist and interact.

Semantic Tension

Often conflated with confounding or missing-data mechanisms; the key tension is between selection as a sampling/inclusion process and other sources of bias that operate through causation or measurement.

Synthesis

Selection bias is the class of distortions arising when the process that determines who or what is observed is correlated with the quantities we aim to estimate; diagnosing it requires understanding inclusion mechanisms and correcting via design, weighting, modeling, or sensitivity analysis.