model-selection

Model selection IS a specific kind of decision-making: choosing among candidate models based on criteria is a decision process. A competent speaker would call model selection a form of decision-making. Direction tested: model-selection is a specific case of the general practice of decision-making.

Model-selection IS a specific kind of method — a systematic procedure for choosing among candidate models (statistical, machine learning, or otherwise). Method is the general (a systematic procedure for accomplishing a task). A competent speaker would call model-selection 'a method' or 'a modeling method.' Direction correct: model-selection→method.

model-selection DEPENDS_ON statistics: remove statistical theory (AIC, BIC, likelihood, cross-validation) and model-selection ceases to operate. There is no framework to compare, evaluate, or select among candidate models without statistical grounding. The removal test passes.

The bias-variance decomposition breaks prediction error into bias, variance, and irreducible error components. This decomposition serves model selection by guiding the choice of model complexity — helping practitioners navigate the bias-variance tradeoff to select models that generalize well. Law 8d: servant (decomposition) points at master (model selection).

AIC is a specific kind of model selection criterion. A competent speaker would call AIC 'a model selection method.' The specific→general direction is correct for INSTANCE_OF. It is one of several model selection criteria (along with BIC, MDL, etc.), each implementing a different trade-off between fit and complexity.

AIC is built for the sake of model selection: it was explicitly designed as a criterion for selecting among statistical models by estimating relative information loss. For whose sake? Model selection. Servant (AIC) → master (model-selection). Law 8d.

Bayes factors are built and maintained for the purpose of Bayesian model selection — they quantify evidence for one model over another to enable model choice. Servant (bayes-factor) points at master (model-selection). Per Law 8d: this records designed purpose, not constitutive need.

Law 8d 'for whose sake?': the information criteria were built explicitly as rules for choosing among candidate models — Akaike (1973) introduced AIC as a model-selection device via his information-theoretic extension of maximum likelihood, and Schwarz (1978) introduced BIC for the same end. Its designed purpose is to make model-selection operational: a shared yardstick for comparing fitted models. Servant (criterion) points at master (model-selection); the purpose is by design and sustained practice, not incidental benefit.