Information matrix test

Diagnose parametric likelihood misspecification by testing whether score outer-product and negative expected-Hessian information estimates agree under the fitted model.

Version

v2 · 2026-09-19 · History

Domain-specific #

2066

Domain group

Social Sciences

Origin domain

Economics & Finance

Subdomains

Econometrics, Econometric Specification Testing → Economics & Finance

Aliases

White information matrix test, IM test, Information matrix equality test

Core Idea

The information matrix test is a specification test for a parametric likelihood model. Under correct specification and standard differentiability and integrability conditions, the expected outer product of the individual score vector equals the negative expected Hessian of the log likelihood. White's test forms sample analogues of selected elements of their discrepancy at a fitted quasi-maximum-likelihood estimate and tests the null that the discrepancy has expectation zero. Rejection indicates that at least one maintained distributional or functional-form condition is inconsistent with the data-generating process.

How would you explain it like I'm…

 

No faithful explanation at this level. At this level the test collapses into a check that tells you whether a model is right and what is wrong with it. In fact it can only flag that some assumption fails, without saying which, and passing it proves nothing.

Two Measurements Should Match

Economists build math models that guess how likely different data are. If a model is built correctly, there are two different ways of measuring how sharply it fits the data, and on average they should give the same answer. The information matrix test checks whether those two measurements disagree by more than chance would explain. If they do, something in the model is wrong, but the test doesn't say what. If they agree, that doesn't prove the model is right; it just means no problem showed up.

Score-Versus-Curvature Specification Test

A likelihood model gives each observation a probability based on some parameters. The score is how fast that log-probability changes with the parameters. The Hessian is how fast that change itself changes, which you can think of as the curvature. When the model is correctly specified (and some technical conditions hold), the average of the score times itself equals minus the average Hessian. This is the information-matrix equality. White's test computes both from the data at the fitted estimates and checks whether the gap is bigger than random noise would produce. A big gap means some assumption, such as the distribution shape or the functional form, is wrong, but it doesn't tell you which. A small gap isn't proof the model is correct.

 

The information matrix test is a specification test for parametric likelihood models. For each observation, define the score s_i (the gradient of the log-likelihood contribution) and the Hessian H_i (its matrix of second derivatives). Under correct specification and standard regularity conditions, E[s_i s_iᵀ + H_i] = 0, the information-matrix equality. White's test evaluates sample averages of selected unique elements of s_i s_iᵀ + H_i at the quasi-maximum-likelihood estimate. It scales and standardizes them using an estimated covariance matrix, which yields a statistic that is asymptotically chi-square under the null. Chesher's interpretation as a score test motivates convenient auxiliary-regression (nR²) versions. Implementation choices, however, affect finite-sample behavior, and Monte Carlo evidence shows some nR² variants have poor size in small samples. The test is omnibus. Rejection signals misspecification without identifying whether it comes from heteroskedasticity, nonnormality, dependence, omitted variables, or something else. Non-rejection does not establish correct specification, since power depends on sample size, alternatives, parameterization, and covariance estimation.

Scope of Application

The abstraction is literal wherever practitioners can identify the same constitutive roles, apply the same boundary tests, and obtain the same kind of output. The following habitats are uses of Information matrix test itself, not metaphors based only on resemblance.

• Maximum-likelihood model checking. Testing distributional and functional-form implications jointly.
• Regression diagnostics. Applying the equality to likelihood-based linear or nonlinear regression models.
• Econometric specification analysis. Screening maintained conditional-density assumptions after estimation.
• Auxiliary-regression implementation. Computing score-interpreted forms of selected discrepancy moments.
• Simulation studies. Evaluating finite-sample size and power of competing variants.
• Robust-inference workflow. Using rejection to motivate targeted diagnostics or alternative covariance and model choices.

Clarity

A clear account of Information matrix test must preserve the recognition invariant stated in the Core Idea rather than rely on the title alone. State the likelihood contributions, parameterization, fitted estimator, and regularity assumptions. Write the score, Hessian, selected discrepancy elements, covariance estimate, and degrees of freedom. Name the exact White, Chesher–Lancaster, or other variant rather than treating all implementations as interchangeable. Interpret rejection as evidence of misspecification and nonrejection as limited, not affirmative, support.

Manages Complexity

Information matrix test manages complexity by replacing a diffuse field of observations or possible operations with a bounded role structure: parametric likelihood supplies a maintained density family defines observation-level log-likelihood contributions.; fitted parameter supplies maximum or quasi-maximum likelihood supplies the evaluation point.; score outer product supplies first derivatives yield one estimator of Fisher information.; negative hessian supplies second derivatives yield the alternative information expression.; discrepancy vector supplies selected unique elements of score products plus Hessians should average to zero..

Abstract Reasoning

1. Specify the conditional or joint likelihood and the null model family.
2. Estimate parameters consistently under the maintained null conditions.
3. Compute observation-level scores and Hessians at the fitted value.
4. Form a nonredundant vector of information-equality discrepancies.
5. Estimate its covariance under the sampling assumptions actually used.
6. Construct the quadratic statistic and verify its asymptotic degrees of freedom.
7. Assess small-sample calibration through appropriate simulation or bootstrap when material.

Knowledge Transfer

The strict upward abstraction is Hypothesis Testing Null Vs Alternative. Information Matrix Test instantiates Hypothesis Testing (Null vs. Alternative): it constructs a null reference law for a statistic whose departure supports a misspecification alternative, with explicit type-I-error calibration. Within econometric specification testing, the full mechanism transfers literally when the same roles and boundary tests recur. Beyond that domain, only the parent-level skeleton should travel. Reusing the label Information matrix test after removing its constitutive vocabulary would hide a change of mechanism behind an analogy.

Relationships to Other Abstractions

Current abstraction Information matrix test Domain-specific

Parents (1) — outgoing typed relations

• Information matrix test is a kind of Hypothesis Testing (Null vs. Alternative) Prime

  Information Matrix Test instantiates Hypothesis Testing (Null vs.

Hierarchy paths (4) — routes to 4 parentless roots

• Information matrix test → is a kind of Hypothesis Testing (Null vs. Alternative) → is a kind of Statistical Inference → is a kind of Inductive Reasoning

Show alternative paths (3)

Neighborhood in Abstraction Space

Information matrix test sits in a sparse region of the domain-specific corpus (89^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (2769 abstractions)

Nearest neighbors

• Observed information — 0.82
• Formation Matrix — 0.80
• Fisher Kernel — 0.80
• Fisher information — 0.80
• Structural Break — 0.79

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