Evolutionary Analysis

Evolutionary analysis asks how traits covary across related taxa, which modes of change are supported, what can be inferred about internal nodes, and how results change across plausible trees and model assumptions.

flowchart LR
    tree["Tree or tree set"]
    traits["Continuous or<br/>discrete traits"]
    models["Covariance · signal<br/>mode · transition model"]
    results["Regression · comparison<br/>ancestral states · uncertainty"]

    tree --> models
    traits --> models
    models --> results

Begin With What Was Actually Observed

Most evolutionary outputs concern quantities that were not directly observed. Name the boundary before interpreting a result:

Quantity	Evidence status	Required conditioning record
measured tip trait or state	observation for an admitted sampling unit	source, unit, uncertainty, mapping, missingness, and exclusions
fitted coefficient, rate, or covariance parameter	model estimate	tree, population, design, likelihood, constraints, and diagnostics
internal-node value or state probability	conditional reconstruction	fitted model, rooted node identity, uncertainty, and tree sensitivity
branch transition or movement	derived historical summary	endpoint or sampled-history rule, branch identity, opportunity, and ambiguity
ecological or geographic narrative	interpretation	explicit claim linking only the accepted observations and reconstructions

This ordering prevents a reconstruction from being described as a recovered observation. It also prevents a visual narrative—such as an annotated tree or map—from silently becoming the source of a transition that was only weakly or conditionally supported in the structured result.

Comparative And Ancestral Methods covers PGLS, phylogenetic signal, continuous and discrete evolutionary modes, ancestral reconstruction, diversification summaries, and sensitivity review. Discrete Ancestral States And Transitions follows state identity, Mk-family fitting, node probabilities, branch transition ledgers, opportunity denominators, and tree-set sensitivity. Cross-Domain Ecology And Geography follows a result across host, niche, region, coordinate, and presentation boundaries. Use the dedicated guides for Biogeography And Migration, Continuous Phylogeography And Maps, and Host Association And Niche Evolution when one of those contracts owns the scientific question.

Choose The Evolutionary Object

Object being inferred	Owning guide	Primary uncertainty
trait association, signal, mode, or ancestral state	Comparative And Ancestral Methods	covariance, tree, model, state, and parameter uncertainty
discrete node state or branch transition	Discrete Ancestral States And Transitions	state order, root prior, transition constraints, node identity, opportunity, and tree-set stability
region state, constrained transition, chronology, or migration event	Biogeography And Migration	state coding, root prior, transition policy, time bins, and tree set
coordinate estimate, branch movement, outlier, or map ledger	Continuous Phylogeography And Maps	spatial error, branch scale, node identity, and filter policy
host switch, niche transition, or clade shift burden	Host Association And Niche Evolution	ancestral-state ambiguity, transition constraints, opportunity, and sampling
joined ecological and geographic narrative	Cross-Domain Ecology And Geography	whether every handoff preserves the same tree, taxa, states, and branch identities

Establish The Observation Population

Before fitting an evolutionary model, identify what one row means. A species mean, an individual, a population, a branch event, and a spatial observation are different sampling units even when all carry the same taxon label.

Population decision	Record	Consequence when omitted
unit of observation	species, individual, population, branch, locality, or aggregate	uncertainty and independence are misrepresented
repeated measurements	subject/taxon grouping and within-group error model	pseudo-replication or discarded variation
taxon reconciliation	retained, excluded, duplicated, aggregated, and reordered rows	model changes are confounded with population changes
trait scale	units, transform, centering, standardization, factor coding	coefficients and reconstructed values lose meaning
missingness	missing mechanism, exclusion/imputation rule, affected columns	fitted population and uncertainty become irreproducible
tree uncertainty	selected tree or weighted tree set plus failed members	one topology is treated as known history

From Reconstructed State To Historical Claim

flowchart LR
    fit["Fitted evolutionary<br/>model"]
    state["Node states or<br/>coordinates"]
    history["Branch transitions<br/>or movement"]
    stability["Tree and model<br/>sensitivity"]
    claim["Bounded ecological or<br/>geographic conclusion"]

    fit --> state --> history --> stability --> claim

A branch transition inherits uncertainty from the tree, state model, root treatment, and observations. An ecological or geographic interpretation must retain that chain rather than treating a rendered map as primary evidence.

Keep Three Result Layers Separate

Layer	Examples	What it establishes	What it cannot establish alone
fitted model	covariance parameter, transition rates, regression coefficients, likelihood	behavior under the declared tree, data and model	literal truth of the evolutionary process
reconstruction	node value/state, branch transition, stochastic history, coordinate path	conditional historical estimate or sample	observed history or certainty at an internal node
interpretation	trait association, host shift, niche transition, geographic movement	bounded biological reading of the retained results	causation, completeness, or robustness outside sensitivity checks

Move forward only when identity and uncertainty survive the handoff. A map or annotated tree is a projection of reconstruction records; it must not become a new source of node states or branch events.

Match The Question To The Model Family

Scientific question	Model family	Interpretation that must remain explicit
Does a trait covary with predictors after phylogenetic dependence?	PGLS or comparative regression	design matrix, covariance transformation, coefficient scale
Is trait resemblance associated with phylogeny?	signal statistics and fitted transformations	null model, tree scale, boundary estimates
Which continuous evolutionary mode is most supported?	Brownian, OU, early-burst, or regime-aware comparison	candidate set, parameter bounds, likelihood convention
How do discrete states change?	Mk and transition-model comparison	state coding, root prior, forbidden transitions
What values or states are supported at internal nodes?	continuous or discrete ancestral reconstruction	node identity, uncertainty, conditioning model
Do conclusions survive tree or model uncertainty?	tree-set and sensitivity review	aggregation rule, failed members, instability
Where are host, niche, or region changes reconstructed?	ecological and phylogeographic branch review	state certainty, transition constraints, branch identity, exclusions
Which continuous movements are supported?	coordinate reconstruction and movement review	coordinate units, uncertainty, tree scale, outlier policy

Do not select a method by output shape. Several models produce coefficients, rates, node values, or information criteria with different conditioning assumptions and meanings.

Sensitivity Axes

At minimum, consider taxon inclusion, tree choice, branch-length scale, rooting, trait transformation, factor coding, covariance or transition model, parameter bounds, and missing-data policy. A result that changes materially along one axis should report that dependence rather than hide it behind the preferred fit.

Refusal And Qualification Boundaries

Refuse a clean evolutionary conclusion when the analytical population is ambiguous, the tree lacks required branch semantics, the design matrix is rank-deficient, an estimated covariance/transition parameter is unidentified, node identities cannot be reconciled, or plausible trees/models reverse the conclusion. Qualify rather than suppress boundary estimates, unstable clades, failed tree-set members, weakly supported transitions, and scale-dependent effects.

Interpretation Boundary

Trait results inherit the taxon mapping, tree scale, covariance or transition model, transformation policy, and uncertainty treatment used to produce them. Model ranking describes the declared candidate set; ancestral estimates remain conditional on both model and tree.

Comparative results also inherit the observation unit. Species means, individual observations, repeated measures, and aggregated counts imply different error and random-effect structures; they are not interchangeable representations of the same analysis.