Optimality criterion¶

An objective measure used to compare candidate statistical models for a hypothesis and designate the model with the best criterion value.

Version: v2 · 2026-09-06 · History
Domain-specific #: 2429
Origin domain: statistics
Subdomain: model and phylogenetic-tree selection
Aliases: Model optimality criterion

Core Idea¶

Optimality criterion is an objective measure used to compare candidate statistical models for a hypothesis and designate the model with the best criterion value. ^[1]

An optimality criterion defines a scalar or ordered objective over candidate models so one can search for and compare best-fitting explanations under a declared standard. In phylogenetics, likelihood, parsimony, and distance-based criteria can rank the same tree space differently because they encode different models, penalties, and summaries of the data. The criterion is distinct from the search algorithm used to optimize it.

Its operative boundary is not supplied by the name alone. Preserve this identity: An objective measure used to compare candidate statistical models for a hypothesis and designate the model with the best criterion value. Validity boundary: The criterion and optimization direction must be specified and applied comparably to all candidates; an unexplained preference is insufficient. The entry therefore captures a reusable specialist role structure rather than a topic label, a single historical instance, or a loose analogy.

Structural Signature¶

Sig role-phrases:

the observed data — the evidence against which candidates are evaluated
the candidate model space — models, trees, or parameterized hypotheses eligible for comparison
the criterion function — the declared score or ordering rule
the modeling assumptions — probability, cost, loss, or evolutionary commitments built into the score
the optimization direction — maximize or minimize under a fixed convention
the search procedure — the algorithm used to find high-scoring candidates
the optimum set — one or more tied candidates attaining the best score
the adequacy check — evaluation of whether score improvement answers the scientific question

Recognition test. A case qualifies only when the analyst can map the declared the observed data, the candidate model space, the criterion function, the modeling assumptions, the optimization direction and preserve the specialist validity conditions. Shared vocabulary, a similar output, or a generic instance of one parent relation is insufficient.

What It Is Not¶

Not the search algorithm. Hill climbing or branch-and-bound explores the space; the criterion says what counts as better.
Not a universal definition of truth. Different criteria embody different assumptions and inferential goals.
Not goodness of fit without penalty. Some criteria trade fit against complexity or other costs.
Not a unique answer guarantee. Ties and near-equivalent optima are common.
Not validation data. A criterion can be optimized on data and still require external or predictive checking.

Scope of Application¶

The abstraction recurs literally within statistical and phylogenetic model spaces whose candidates can be scored under a declared inferential objective.^[1] The following habitats preserve the same recognition machinery; they are not invitations to extend the name metaphorically.

Maximum likelihood. models are ranked by the probability assigned to observed data.
Maximum parsimony. trees are ranked by the minimum changes required.
Residual criteria. models minimize squared, absolute, or weighted discrepancies.^[1]
Information criteria. fit is adjusted by a complexity penalty.^[1]
Tree inference. topologies and nuisance parameters are jointly or separately optimized.
Sensitivity analysis. rankings are compared across plausible criteria and assumptions.

Clarity¶

A report should name the score, its sign convention, its assumptions, and the candidate space. Saying that software found the 'optimal model' is incomplete when the criterion and search completeness are unknown. A global optimum under one score need not be adequate or preferred under another.

A practical identification audit begins with the typed roles rather than the title: establish the observed data, verify the candidate model space, then test the remaining conditions and exclusions. If the case retains only the portable skeleton described below, it should be named through a parent abstraction rather than as Optimality criterion.

Manages Complexity¶

The criterion converts a heterogeneous comparison into an explicit ordering that algorithms can optimize and researchers can contest. Separating criterion, model, data, and search exposes whether disagreement comes from evidence, assumptions, or incomplete exploration.

The compression remains accountable because each simplification has a named failure condition. Disagreement can be localized to a missing role, an invalid assumption, an ambiguous measurement, or a neighboring abstraction instead of being hidden inside an unanalyzed label.

Abstract Reasoning¶

R1. Define the candidate space before interpreting an optimum. R2. Write the objective and optimization direction explicitly. R3. Identify assumptions and penalties embedded in the score. R4. Separate score evaluation from the algorithm that searches candidates. R5. Assess ties, uncertainty, sensitivity, and scientific adequacy after optimization.

These moves separate definition, derivation, measurement, and interpretation. A formal consequence does not by itself prove that an observed case instantiates the abstraction, while an observed resemblance does not relax the formal or institutional recognition conditions.

Knowledge Transfer¶

Optimality criteria transfer literally among model-selection problems with explicit candidates and objectives. Optimization and selection are broader primes; calling a personal preference, vague ideal, or unmeasured goal an optimality criterion removes the score and model-comparison machinery.

The transfer boundary is explicit: DOMAIN-SPECIFIC PASS / PRIME FAIL: Optimality criteria recur across datasets, competing models, hypotheses, phylogenetic trees, and objective functions. Literal recognition retains the specialist vocabulary and validity conditions of statistical model selection; outside that setting only broader parent operations transfer. The safe move beyond the home habitat is to carry the applicable parent relation and leave the specialist name behind unless every defining role remains literal.

Examples¶

Canonical: maximum-likelihood tree selection¶

For each candidate topology, branch lengths and substitution-model parameters are optimized and the resulting likelihood of the observed sequence data is recorded. The highest maximized likelihood ranks best under that criterion. A heuristic tree search may fail to find the global maximum, so the criterion value and search procedure must be reported separately. ^[2]

Mapped back: the observed data; the candidate model space; the criterion function; the modeling assumptions; the optimization direction; the search procedure.

Applied / In Practice: parsimony and likelihood disagree¶

The same alignment is scored under minimum character changes and under a probabilistic substitution model. The criteria prefer different trees. The disagreement is not repaired by declaring one algorithm more efficient; it requires examining the evolutionary assumptions, weighting, and predictive behavior of the two objectives. ^[3]

Mapped back: the criterion function; the modeling assumptions; the optimum set; the adequacy check.

Structural Tensions¶

T1: Fit vs complexity. A flexible model can improve in-sample score while generalizing poorly. Diagnostic: Where and how is complexity penalized?

T2: Criterion vs search. A precise objective can still be optimized incompletely. Diagnostic: Is the reported optimum global, bounded, or heuristic?

T3: One optimum vs model uncertainty. Selecting a winner can conceal near-ties and unstable rankings. Diagnostic: Are score differences and uncertainty reported?

T4: Assumption transparency vs convenient scalar. One number simplifies choice while hiding its modeling commitments. Diagnostic: Which premise changes the ranking most?

T5: Statistical optimum vs scientific adequacy. Best within a candidate set can still be substantively wrong. Diagnostic: What residual or predictive check tests adequacy?

T6: Domain autonomy vs prime reduction. Optimization and selection omit the scored statistical model space and inferential assumptions. Diagnostic: Would an arbitrary objective function still be this model-selection abstraction?

Structural–Framed Character¶

The five-criterion aggregate is 0.45 (mixed). The judgment is criterion-specific:

Vocabulary travels — material (0.50). The complete vocabulary remains tied to the typed roles in the Structural Signature.
Evaluative weight — low (0.25). Application carries the stated degree of normative or interpretive judgment beyond structural recognition.
Institutional origin — material (0.50). The abstraction depends to this degree on a scholarly, technical, legal, or social convention.
Human-practice bound — material (0.50). Recognition depends to this degree on organized practice, language, measurement, or institutional action.
Import versus recognize — material (0.50). Beyond its home habitat, use of the full name increasingly becomes analogy rather than literal recognition.

The portable skeleton is a declared objective orders competing representations so search and choice become explicit and auditable. The named abstraction remains mixed because that skeleton alone does not supply its specialist objects, constraints, or tests.

Structural Core vs. Domain Accent¶

Structural core: A declared objective orders competing representations so search and choice become explicit and auditable.

Domain accent: Statistical models, phylogenetic trees, likelihood, parsimony, residual loss, complexity penalties, nuisance parameters, and model adequacy.

Why it does not clear the prime bar: Optimization is portable; an optimality criterion is the inferential scoring rule attached to a statistical candidate space and its assumptions. Generalization therefore routes through parent abstractions; preserving the specialist name requires the full accent.

Optimization (prime:optimization). The criterion supplies the objective whose extremum defines the preferred candidate.
Selection (prime:selection). Candidates are retained or ranked according to an explicit comparison rule.

These are prose placement proposals only. They create no dag_edges; endpoint, redundancy, and cycle checks are recorded separately in the bundle's placement memo.

Relationships to Other Abstractions¶

Current abstraction Optimality criterion Domain-specific

Parents (2) — more general patterns this builds on

Optimality criterion is a kind of Evaluation Prime

The accepted reference-grade review places Optimality criterion under Evaluation because the child instantiates or depends on the parent's broader structure while retaining its own constitutive identity.
Optimality criterion presupposes Selection Prime

Selection (prime:selection).

Children (1) — more specific cases that build on this

Deviance Information Criterion Domain-specific is a kind of Optimality criterion

Optimality Criterion is the proposed immediate parent.

Hierarchy paths (2) — routes to 2 parentless roots

Optimality criterion → Evaluation → Comparison → Self Checking

Show alternative path (1)

Neighborhood in Abstraction Space¶

Optimality criterion sits in a sparse region of the domain-specific corpus (72^nd percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Inference Bias & Multiple Testing (10 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-09-08

Not to Be Confused With¶

Search heuristic. an algorithm for exploring candidate models. Tell: Does it define better, or merely try to find it?
Loss function. a discrepancy function often used inside a criterion. Tell: Is the term naming the component loss or the full model-ranking rule?
Information criterion. a penalized-likelihood subtype such as AIC. Tell: Is complexity penalty part of the score?
Goodness-of-fit test. a test of compatibility rather than a ranking objective. Tell: Is the output a hypothesis test or candidate ordering?
Decision criterion. a broader rule for actions under consequences. Tell: Are statistical explanatory models being ranked against data?

References¶

[1] Kenneth P. Burnham and David R. Anderson, Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach, 2^nd ed., Springer, 2002. General reference for statistical candidate-model comparison, residual fit, information criteria, complexity penalties, and adequacy-aware model selection. registry ↩a ↩b ↩c ↩d

[2] Mike Steel and David Penny, “Parsimony, Likelihood, and the Role of Models in Molecular Phylogenetics”, Molecular Biology and Evolution 17(6) (2000), 839–850. registry ↩

[3] David Penny, “Criteria for Optimising Phylogenetic Trees and the Problem of Determining the Root of a Tree”, Journal of Molecular Evolution 8 (1976), 95–116. registry ↩