Endogeneity¶

The condition in which a regressor is correlated with a model's error term — through confounding, simultaneity, or measurement error — so OLS coefficients are biased and inconsistent for the causal effect, collapsing the coefficient's causal reading while leaving its predictive one intact.

Version: v3 · 2026-09-28 · History
Domain-specific #: 333
Domain group: Formal Sciences
Origin domain: Experimental Design & Statistics
Subdomain: Causal Inference and Econometrics → Experimental Design & Statistics

Core Idea¶

Endogeneity is the condition in which a regressor is correlated with a model's error term, so ordinary least squares produces coefficient estimates that are biased and inconsistent for the causal effect. It arises through three structural mechanisms: omitted-variable confounding (an unobserved common cause of regressor and outcome), simultaneity (regressor and outcome co-determine each other), and measurement error in the regressor (attenuating the coefficient toward zero). All three violate exogeneity — the assumption the regressor is set independently of everything else affecting the outcome.

How would you explain it like I'm…

The Sneaky Hidden Cause

On sunny days, kids eat more ice cream and also get more sunburns. If you only looked at ice cream and sunburns, you might think ice cream causes sunburns. The hidden sun is making both happen. Endogeneity is when a hidden helper like that tricks you about what really causes what.

Tangled-Up Causes

Scientists often try to measure how much one thing changes another, like how much studying changes test scores. Endogeneity means the thing you are measuring is tangled up with other things you did not account for. A hidden cause might push on both, the two might push on each other, or you might have measured the first thing badly. When that happens, your answer about the effect comes out wrong, and collecting more of the same data does not fix it.

Regressor–Error Correlation

In a regression, you predict an outcome from explanatory variables, and everything you left out goes into an 'error term.' Endogeneity means an explanatory variable is correlated with that error term, so the estimated effect is biased and doesn't shrink toward the truth even with more data. It happens in three main ways: a hidden variable causes both the explanatory variable and the outcome (confounding), the two variables cause each other (simultaneity, like police presence and crime), or the explanatory variable is measured with error, which pushes the estimate toward zero. The opposite condition, exogeneity, means the variable behaves as if a researcher set it independently, and only then does the coefficient mean 'the effect of changing this, holding everything else fixed.' Correlation alone isn't the issue; the issue is correlation with what you left out.

Endogeneity is the condition where a regressor is correlated with the model's error term, which makes OLS (and most regression-type estimators) biased and inconsistent for the regressor's causal effect. Its three standard sources are omitted-variable confounding, where an unobserved variable drives both regressor and outcome so the coefficient absorbs the confound's effect; simultaneity, where outcome and regressor are jointly determined, as with price and quantity in supply and demand; and measurement error in the regressor, which produces attenuation bias toward zero. All three violate exogeneity, the assumption that the regressor can be treated as if set independently of everything else affecting the outcome. Under exogeneity, the coefficient is the expected change in the outcome when the regressor changes with the error held fixed; under endogeneity, that interpretation fails. Diagnosing endogeneity motivates the quasi-experimental toolkit: instrumental variables, regression discontinuity, difference-in-differences, and panel fixed effects. Which remedy fits depends on the source of endogeneity and on what identifying variation the data or institutional setting provides.

Scope of Application¶

Because endogeneity is a diagnostic condition on a model-and-data pairing with an attached remedy-menu, it applies wherever causal effects are estimated from a model whose right-hand-side variables may be entangled with the outcome.

Labour economics — the textbook case: returns to schooling and the unobserved-ability confound.
Industrial organisation — price-quantity simultaneity in demand estimation.
Development economics — programme placement endogenous to baseline characteristics.
Macroeconomics — policy variables responding to the conditions they explain.
Epidemiology — confounding-by-indication, selection-into-treatment, and reverse causation.
Political science, sociology, and finance — endogenous take-up, peer effects, and simultaneous determination.

Clarity¶

Naming endogeneity replaces the coarse slogan "correlation is not causation" with a mechanism — the regressor is correlated with the error — that says exactly how and why a coefficient can describe well yet cause falsely. It collects several pathologies under the violation of exogeneity while keeping them separable, and shifts the burden from "is the coefficient significant?" to "why should this regressor be treated as exogenous?"

Manages Complexity¶

A double sprawl of biases and repair techniques collapses to a single hinge: is the regressor correlated with the error? The analyst tracks one binary property plus a three-way source classification, and the toolkit becomes a menu indexed by the diagnosed source. The causal warrant, bias direction, and appropriate design all read off the exogeneity condition.

Abstract Reasoning¶

The construct licenses a boundary-drawing move separating description from causation by the exogeneity condition, a diagnostic move classifying which of three sources broke it, a predictive move on bias sign and shape, an interventionist move mapping source to remedy, and a burden-shifting move forcing the identifying assumption into the open.

Knowledge Transfer¶

Within causal inference the diagnostic condition and remedy-menu transfer literally across labour economics, IO, development, epidemiology, and finance. The portable lesson — a variable used to explain an outcome may be entangled with it, breaking the causal reading — belongs to the entanglement primes Feedback, Confounding/common_cause, Selection Bias, and measurement_error. The econometric apparatus stays home, and the macroeconomic "endogenous variable" is a distinct homonym.

Relationships to Other Abstractions¶

Current abstraction Endogeneity Domain-specific

Parents (1) — outgoing typed relations

Endogeneity presupposes Regression Domain-specific

Endogeneity presupposes a regression-style model whose regressor, error term, and exogeneity assumption make the violation definable.

Children (3) — abstractions linked here by typed relations

Attenuation Bias Domain-specific presupposes Endogeneity

Classical regressor-measurement attenuation presupposes the endogeneity created when the noisy observed regressor correlates with the composite error.
Instrumental variable Domain-specific presupposes Endogeneity

Instrumental-variable identification presupposes an endogenous treatment whose correlation with the model error destroys the naive causal coefficient.
Omitted Variable Bias Domain-specific presupposes Endogeneity

Omitted-variable bias presupposes endogeneity because its two-condition gate makes an included regressor correlated with the regression error.

Hierarchy paths (10) — routes to 6 parentless roots

Endogeneity → presupposes Regression → is a kind of Signal Extraction

Show alternative paths (9)

Neighborhood in Abstraction Space¶

Endogeneity sits in a sparse region of the domain-specific corpus (75^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Statistical Bias & Inference Pitfalls (7 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-25