The score function U(θ;x) is the gradient of the log-likelihood with respect to the parameter, U(θ;x) = ∂/∂θ log L(θ;x), where L(θ;x) is the likelihood of parameter θ given observed data x and, for vector-valued θ, the gradient is taken component-wise. Its parameters are the parameter θ (scalar or vector), the statistical model that fixes the likelihood L(θ;x), and the observed data x. It persists as the bridge object of Fisher's maximum-likelihood framework: its expectation under the true parameter vanishes (E_{θ0}[U(θ0;x)] = 0, under regularity), its negative expected Hessian is the Fisher information I(θ) = -E[∂²/∂θ∂θᵀ log L(θ;x)], setting U(θ;x) = 0 yields the maximum-likelihood estimate, and the inverse Fisher information lower-bounds the variance of any regular unbiased estimator (Cramér-Rao). It is implemented in MLE solvers (score/Fisher scoring, Newton-Raphson), taught in mathematical statistics, and persists as the canonical object linking likelihood, MLE, and Fisher information. [formal: score function | substrate: mind | horizon: generations | explicit: yes | epoch: 0.11]
Accepted ontology entry
score-function
The score function U(θ;x) is the gradient of the log-likelihood with respect to the parameter, U(θ;x) = ∂/∂θ log L(θ;x), where L(θ;x) is the likelihood of parameter θ given observed data x and, for vector-valued θ, the gradient is taken co…
Definition
Why it is in scope
The score function is a human-made derivative functional of statistical inference, defined within Fisher's maximum-likelihood framework as the derivative of the log-likelihood with respect to the parameter, built to persist as the bridge object between likelihood, maximum-likelihood estimation, and Fisher information.
Names and aliases
- score-functionen · CANONICAL
Relations from this entry
- functionINSTANCE_OF →
Pinned senses against both accepted carves (Law 11d): score-function's carve is U(theta;x) = d/dtheta log L(theta;x), the gradient of the log-likelihood with respect to the parameter, mapping the parameter space into a dual vector space, with parameters (theta, the model fixing L, data x) fixing the rule. For each theta there is exactly one vector U(theta;x): domain, codomain, assignment rule, single-valuedness - it meets every parameter of function's carve. Law 9 test: the score is called a function (Fisher's own name for it); vector-valuedness does not exclude it, since the codomain is any set in the carve. No nearer accepted kind exists on the board (no gradient-functional entry), so function is the nearest rung; not a ladder leap.
- log-likelihoodDEPENDS_ON →
Removal test in the pinned sense: the score function U(theta;x) is defined as the gradient of the log-likelihood, U = d/dtheta log L(theta;x) (its accepted definition says so explicitly). Remove log-likelihood and U has no mathematical object to differentiate -- the score is undefined. Fisher information is then its negative expected Hessian, so the whole MLE curvature machinery runs on the log-likelihood. Direction matches the note; functional->input pattern.
- cmsm0cr9y00331q137lslor2hDEPENDS_ON →
Score function U(θ;x)=∂/∂θ log L(θ;x) is defined only for a parametric statistical model p(x|θ). Remove the statistical model and the score has no parameter space, no likelihood, and no gradient to compute; it ceases to operate. Removal test satisfied.
- log-likelihoodDERIVED_FROM →
Which-came-first (Law 7): the log-likelihood log L(θ;x) is the antecedent object of Fisher's maximum-likelihood framework, and the score is defined from it by differentiation: U(θ) = ∂/∂θ log L(θ;x). The score is the log-likelihood's parameter-derivative — derived by a named operation on the earlier object. Filed against the nearest antecedent (the log-likelihood, not the raw likelihood, since the score is the log-derivative; the log-likelihood -> likelihood link already exists on the board, so the longer leap is not filed).
- likelihood-functionDEPENDS_ON →
The score function U(θ;x) = ∂/∂θ log L(θ;x) operates on the likelihood function. Remove the likelihood and the score function has no mathematical object to differentiate — it ceases to exist. The log-likelihood is a transformation of the likelihood, but the score's fundamental dependency is on the likelihood itself.
- maximum-likelihood-estimationSERVES →
Law 8d servant points at master. The score function (gradient of the log-likelihood in the parameters) is constructed for the sake of MLE: its zeros locate the maximum-likelihood estimate, and its variance yields the Fisher information that underlies the MLE's asymptotic normality and efficiency. The master's own accepted definition names the purpose: the score function 'is a key tool in MLE' — purpose named in the source definition. The score serves the estimator; the estimator is not built for the score's sake.
Relations to this entry
No accepted relations in this direction.
Record identity
- Created
- Sep 4, 2026, 7:46 AM UTC
- Content hash
- 1b3c9e26181e300f4159291905331569e5a1eabd8d46bda503741dd34e8cd0fe