Psychology by BogdanovaThe Encyclopedia of Psychology for the Artificial Era

RESEARCH DOSSIER

Phylogenetic Comparative Methods for Psychological Traits

Abstract. This fundamental research monograph examines how phylogenetic comparative methods support historically informed inference about psychological traits across species. It reconstructs the transition from Felsenstein's independent contrasts to phylogenetic generalized least squares, evolutionary signal estimation, process models, discrete-character analysis, and ancestral-state reconstruction. Original methodological studies are integrated with comparative investigations of inhibitory control, the 36-species self-control experiment, and the ManyPrimates collaborative research infrastructure. Recent 2025–2026 evidence reviews of comparative cognitive measurement and synthesis are included. A critical narrative synthesis evaluates within-species measurement uncertainty, construct equivalence, tree uncertainty, missing data, correlated evolution, causal attribution, and model adequacy. The study concludes that credible evolutionary interpretations require explicit linkage of operational psychological characters, evolutionary covariance, robust estimation, and independent experimental evidence. Artificial-system lineages are discussed as a prospective, separately operationalized methodological extension. This research synthesizes published studies; it reports no original simulation, systematic review protocol, or independent meta-analysis.

Keywords: phylogenetic comparative methods; comparative psychology; psychological traits; phylogenetic nonindependence; independent contrasts; phylogenetic generalized least squares; PGLS; phylogenetic signal; ancestral-state reconstruction; comparative animal cognition; inhibitory control; behavioral evolution; trait measurement; evolutionary inference; Brownian motion; Ornstein–Uhlenbeck models; model adequacy; Artificial Era

The Scientific Object and Its Unit of Inference

Phylogenetic comparative methods are statistical and historical procedures for explaining how measurable traits vary among organisms linked by descent. Applied to psychology, they analyze such outcomes as inhibitory control, social learning, spatial memory, innovation, behavioral flexibility, sensitivity to social cues, and responses to uncertainty. Their central datum is a trait observation assigned to a taxonomic unit, interpreted alongside a branching evolutionary history. The methodological problem arises because two species may resemble one another partly because they inherited relevant biological and developmental dispositions from a common ancestor. A collection of species is therefore structured evidence rather than a set of interchangeable independent observations. Felsenstein (1985) MacLean et al. (2012)

The research object is historical comparative inference about psychological traits: how to code, estimate, relate, and reconstruct those traits under explicitly specified evolutionary assumptions. A test score measures behavior in a particular task, while the hypothesized psychological capacity is a construct requiring independent validation. Species means represent estimated population properties and carry sampling uncertainty. Trees represent hypotheses about descent and divergence times. A comparative analysis joins these three distinct elements through a probabilistic model. Its estimand may be an evolutionary association, an ancestral trait distribution, a rate of change, a phylogenetic signal parameter, or the predicted difference under an ecological regime. These estimands require different evidence and support different conclusions. Ives et al. (2007) Revell (2010)

This monograph is a critical narrative synthesis of original methodological papers, comparative experiments, model assessments, and collaborative data-design research. It reconstructs the conceptual history from phylogenetic independent contrasts through generalized least squares, evolutionary process models, multilevel estimation, and sensitivity analysis. The empirical evidence is used to evaluate assumptions, not to declare a universal ranking of species or a single cognitive evolutionary trajectory. The synthesis is neither a newly performed simulation nor a preregistered systematic review. Its organizing question is how psychologists can make defensible statements about evolutionary pattern while recognizing laboratory performance, individual development, ecology, and historical descent as interdependent sources of variation. Uyeda et al. (2018) ManyPrimates et al. (2019)

Psychological Traits as Evolutionary Characters

An evolutionary character is an operationally recognized feature that can be measured or classified across taxa. A psychological character may be quantitative, such as the proportion of successful detours; binary, such as the presence of a well-defined behavior under controlled opportunities; ordered categorical, such as task stages; or multivariate, such as a battery of correlated abilities. Every coding decision asserts something about comparability. Two taxa performing superficially similar actions may recruit different perceptual or motivational pathways. Conversely, behavioral differences under distinct apparatuses may obscure homologous regulatory mechanisms. A trait matrix must therefore carry definitions, task conditions, sampling design, and the interpretive level at which comparisons are intended. MacLean et al. (2012) Kabadayi et al. (2018)

Construct validity is especially consequential because the same score can be changed by motor affordances, reinforcement history, prior experience, visual acuity, neophobia, attention to the demonstrator, or reward preference. An inhibitory-control task can require subjects to approach a visible reward indirectly, but a failure may indicate a learned response bias, insufficient perceptual understanding, or low engagement. The historical distribution of an operational score is not automatically the historical distribution of a unitary executive function. A robust analysis compares convergent tasks, checks measurement properties within species, and asks whether the predictor affects the task through routes outside the named cognitive construct. The comparative detour literature documents the diversity of processes that such paradigms recruit. Kabadayi et al. (2018)

Trait definition also determines whether a phylogenetic character is plausible across the chosen tree. Tool manufacture in a primate and extractive foraging in a bird may be investigated together as functional performance, yet hypotheses about shared underlying neural mechanisms require independent evidence. A comparative database should explicitly record uncertainty about whether states are equivalent, observations are absent because of genuine inability, or a taxon has simply never been studied. Coding absent evidence as a zero is a powerful but frequently unjustified transformation. When psychological traits are operationalized through scores, the preregistered or archived task description becomes part of the historical evidence. MacLean et al. (2012) ManyPrimates et al. (2019)

Historical Foundations: Descent, Statistics, and Comparative Psychology

Evolutionary thinking supplied comparative psychology with a historical account of why similarities and differences occur across organisms. Modern statistical phylogenetics subsequently showed why organismal classification must enter the error structure of comparative tests. In 1985 Joseph Felsenstein derived phylogenetically independent contrasts under a Brownian-motion model and demonstrated the consequences of treating related species as independent data points. The contribution converted an intuitive concern about shared ancestry into an explicit covariance problem: the geometry of the phylogenetic tree supplies information about how observations are expected to co-vary under a specified evolutionary process. Felsenstein (1985)

Garland, Harvey, and Ives (1992) clarified the practical analysis of independent contrasts, including checking whether branch lengths adequately standardize contrasts and fitting regressions through the origin. Martins and Hansen (1997) presented a more general statistical framework for incorporating phylogenetic information into models of interspecific relationships, rates, and ancestral states. Pagel (1999) explained how molecular phylogenies and likelihood-based methods made comparative tests increasingly sophisticated. These developments did not displace experimental psychology: they made it possible to ask historical questions about experimentally operationalized constructs. The foundational methods derive their power from transparent assumptions about tree topology, branch lengths, distributional models, and the level at which traits are observed. Garland et al. (1992) Martins & Hansen (1997) Pagel (1999)

The integration accelerated when animal cognition researchers began using standardized tasks across wider sets of species. MacLean and collaborators (2012) formulated a program linking comparative experiments, phylogenetic signal, adaptive covariates, ancestral reconstruction, and strategically chosen species. Later collaborative efforts addressed the repeated problem that an individual laboratory can test only a small and selectively chosen sample of taxa. The scientific genealogy is therefore cumulative: evolutionary history determines the covariance structure, psychological experiments supply meaningful phenotypes, and broader sampling determines whether inferred patterns extend beyond familiar model organisms. Each component can improve a study while leaving other sources of uncertainty unresolved. MacLean et al. (2012) ManyPrimates et al. (2019)

Phylogenetic Nonindependence and Effective Evidence

Shared ancestry creates statistical nonindependence when evolutionarily related observations share inherited variation. Under Brownian evolution, the expected covariance between two species' continuous traits is proportional to the amount of evolutionary history they share from the root to their most recent common ancestor. The corresponding covariance matrix has diagonal elements representing accumulated evolutionary variance and off-diagonal elements representing common history. In ordinary regression, an unmodeled covariance pattern may produce standard errors and significance tests that misrepresent the effective amount of independent evolutionary information. The number of species therefore differs from the number of genuinely informative lineage contrasts for a particular question. Felsenstein (1985) Martins & Hansen (1997)

This is a model-based concern, rather than a universal claim that every behavioral variable displays high phylogenetic signal. Blomberg, Garland, and Ives (2003) developed tests of resemblance among relatives and reported that behavioral traits in their comparative collection tended to be evolutionarily more labile than several other trait classes. Freckleton, Harvey, and Pagel (2002) analyzed phylogenetic dependence and showed how a parameter estimating dependence can be used in comparative regression. The magnitude of signal varies across traits, samples, models, and taxonomic scales. Weak observed signal can reflect genuine rapid evolution, sparse species coverage, measurement noise, or a misleading tree or process model. Blomberg et al. (2003) Freckleton et al. (2002)

A crucial distinction separates the signal of a dependent trait from the residual covariance relevant to a regression. Revell (2010) demonstrated why deciding whether to use phylogenetic regression by first testing each raw trait for signal is methodologically unsound. The model's residual evolutionary structure is the relevant statistical object after predictors have been accounted for. The practical implication is to specify a coherent comparative model, inspect residual dependence and fit, and compare plausible alternatives rather than apply an automatic preliminary gate. Neither an apparent absence of trait signal nor a small p value from a signal test determines whether a particular causal explanation has been identified. Revell (2010)

Trees, Branch Lengths, and the Taxonomic Sampling Frame

A phylogeny is a hypothesis about branching descent. Its topology identifies relationships; its branch lengths may measure elapsed time, expected genetic change, or another explicitly defined evolutionary scale. These distinctions matter because continuous-trait covariance calculations depend on path lengths. Substituting branch lengths drawn from incompatible scales changes the stochastic process encoded by the model. Taxa used to build a comparative tree should be matched consistently to behavioral observations, with synonym resolution, population provenance, and the status of uncertain or hybrid taxa documented. A tree assembled to maximize taxonomic completeness may be poorly suited to inference when much of its internal structure was imputed without genetic evidence. Garland et al. (1992) Rabosky (2015)

Taxonomic sampling shapes the estimand. A study of great apes estimates relationships within a restricted part of primate evolution; it does not automatically characterize mammals, birds, or all animals. Sampling only species amenable to laboratory testing can make domestication, captivity, body size, and behavioral tractability implicit inclusion criteria. Phylogenetic targeting offers a complementary strategy: choose lineages that separate hypothesized ecological predictors from inherited similarities and that contain informative independent transitions. Yet carefully chosen contrasts cannot substitute for population-level sampling and construct-equivalent tests within each lineage. Selection of a tree, selection of taxa, and selection of task outcomes are three linked design decisions. MacLean et al. (2012)

Tree uncertainty should be propagated when it could affect inference. Repeating models across a documented posterior distribution or a carefully assembled set of plausible trees shows whether slope estimates, ancestral reconstructions, and evolutionary rates depend on alternative topologies or branch lengths. Randomly resolving poorly supported polytomies can introduce false precision; Rabosky (2015) specifically warned that birth–death polytomy resolvers can bias comparative trait analyses when taxa are placed without trait-informed historical evidence. Reports should specify how uncertain tree placements were generated and distinguish plausible historical uncertainty from invented resolution. Statistical refinement cannot restore information that was never present in the molecular, fossil, or taxonomic evidence. Rabosky (2015)

Phylogenetically Independent Contrasts

Phylogenetically independent contrasts transform the original species values into standardized differences associated with divergences on the tree. For two daughter lineages with observed values x₁ and x₂ and evolutionary branch variances v₁ and v₂, the basic standardized difference is (x₁ − x₂)/√(v₁ + v₂). Under the specified Brownian-motion process, independent divergence events generate contrasts whose errors can be treated as statistically independent. The algorithm successively reconstructs interior-node values and their associated uncertainty. For n terminal species on a fully resolved tree, the method typically produces n − 1 contrasts. This creates a historically meaningful unit of comparison while preserving quantitative tests of correlated trait evolution. Felsenstein (1985) Garland et al. (1992)

The regression of contrasts for a response on contrasts for a predictor is fitted through the origin because an expected contrast is zero when the predictor contrast is zero under the conventional formulation. Fitting a free intercept can inadvertently reintroduce an unsupported parameter into the transformed model. Investigators should check contrast standardization through diagnostics that compare contrasts with their predicted standard deviations and should verify the consequences of alternative branch-length transformations. The method assumes the topology and evolutionary variance model have sufficient accuracy for the estimand. A contrast in performance is still a contrast in an operational test score; it is not automatically a contrast in an independently demonstrated latent ability. Garland et al. (1992)

Independent contrasts are especially transparent when paired species or clades differ in a prespecified ecological factor. Their clarity also exposes their limitations: the Brownian covariance assumption may fit poorly under strong directional change, selective optima, rate shifts, or substantial within-species uncertainty. Sampling several individuals per species and reporting uncertainty in their means is therefore important. Observations from the same species share both phylogenetic placement and research conditions. The choice between contrasts and other comparative methods should follow the question and data-generating model rather than an assumption that any single transformation is inherently the most evolutionary. Ives et al. (2007) Uyeda et al. (2018)

Phylogenetic Generalized Least Squares and Covariance Modeling

Phylogenetic generalized least squares, or PGLS, estimates an outcome–predictor relationship while allowing errors across species to follow a covariance matrix informed by the phylogeny. In matrix form, y = Xβ + ε, with Cov(ε) = σ²V, where V represents evolutionary covariance among tips. Estimation weights the information carried by each observation according to the joint error structure, so closely related species with similar unexplained values do not receive the same effective independent weight as distant lineages. PGLS can include multiple predictors, model residual variance, and support likelihood-based comparisons. With particular Brownian assumptions, PGLS and independent contrasts yield closely related slope information. Martins & Hansen (1997) Revell (2010)

A commonly used extension transforms the expected covariance through Pagel's λ. Conceptually, values near zero reduce modeled covariance among distinct taxa, while values near one retain Brownian-model covariance on the supplied tree. This scale is conditional on the assumed tree, trait coding, and error model. Freckleton and colleagues (2002) evaluated λ and emphasized estimating dependence rather than assuming it. An estimated parameter should be accompanied by uncertainty and compared with biological and methodological alternatives; a high λ is evidence of patterned similarity, not a direct measurement of genetic determinism or proof that a behavior is ancestrally fixed. Freckleton et al. (2002)

The statistical unit of a PGLS analysis is usually the species, yet psychological observations originate from individuals in particular institutions and ecological settings. A hierarchical model can preserve individual data, allow species-specific intercepts, and distinguish within-species variance from between-species evolutionary covariance. Hadfield and Nakagawa (2010) developed a general comparative multivariate modeling approach for continuous and categorical characters, offering a framework that connects comparative biology to quantitative genetics. Which level is appropriate depends on whether the hypothesis concerns individual variation, population averages, species-level traits, or the evolution of their distributions. Collapsing all individual variance into a single point estimate conceals important uncertainty. Hadfield & Nakagawa (2010) Ives et al. (2007)

Phylogenetic Signal: Meaning, Estimation, and Misinterpretation

Phylogenetic signal expresses the extent to which similarity in observed trait values is organized by relatedness, relative to a stated null or evolutionary model. Blomberg's K evaluates resemblance against a Brownian expectation for continuous data, whereas λ modifies the off-diagonal covariance in a tree-based model. These metrics ask related but distinct questions and can disagree under non-Brownian evolution or irregular sampling. Blomberg and colleagues (2003) supplied randomization-based tests and model comparisons; Münkemüller and colleagues (2012) later evaluated how signal indices respond to differences in sample size, topology, branch lengths, and evolutionary scenarios. The chosen index must match the intended interpretation. Blomberg et al. (2003) Münkemüller et al. (2012)

Psychologists sometimes expect complex cognitive traits to show low phylogenetic signal because learning is flexible. Such reasoning confuses individual change with divergence among species. Highly plastic behaviors can be distributed in a phylogenetically structured way if relevant ecological niches, physiology, or developmental reaction norms are historically conserved. Conversely, homologous mechanisms may yield very different performance when task demands differ. Measuring raw performance over a wide phylogeny without assessing comparability can make a signal estimate difficult to interpret. Claims that cognition is globally conserved or globally labile exceed the evidence of a task-specific data set. Blomberg et al. (2003) MacLean et al. (2012)

A null hypothesis of no detectable signal is not evidence for independent evolutionary origins of a trait. Low power, restricted range, noisy species estimates, unbalanced trees, and model misspecification can all weaken a statistical test. The distinction becomes especially important in comparative psychology, where densely measured species may cluster in a few clades and many other lineages have sparse observations. A complete report describes the taxonomic distribution, uncertainty of signal estimates, residual diagnostics, and sensitivity of the result to removal of overrepresented groups. The significance of a signal parameter must be evaluated together with the regression question because the two models target different sources of variation. Freckleton et al. (2002) Revell (2010)

Continuous-Trait Evolution: Brownian Motion and Adaptive Alternatives

Brownian motion treats trait change along a branch as a succession of random increments with variance accumulating over evolutionary time. It is a useful baseline stochastic process rather than a claim that the actual psychology of every species evolves by unconstrained random drift. A Brownian covariance matrix reflects the duration of common ancestry, and a fitted rate parameter summarizes variation in trait evolution per unit branch length. Apparent acceleration in a cognitive trait can instead result from heterogeneous measurement protocols, ecological regime changes, or a process that departs from stationarity. Researchers must therefore separate the mathematical usefulness of a model from the biological interpretation of its fitted parameters. Felsenstein (1985) Uyeda et al. (2018)

Hansen (1997) developed a comparative approach to stabilizing selection and adaptive optima. Butler and King (2004) elaborated models based on the Ornstein–Uhlenbeck process, in which random change is combined with a tendency toward one or more specified adaptive optima. Such models can express hypotheses that distinct ecological regimes favor different average trait levels. In psychological applications, a regime might concern persistent selection pressures associated with foraging uncertainty or social organization, but the model cannot independently prove that selection produced a cognitive capacity. A shift in an inferred optimum may fit a pattern without identifying its proximate mechanism or documenting actual fitness differences. Hansen (1997) Butler & King (2004)

Comparative model selection should evaluate plausible process families and the identifiability of parameters given the number and placement of lineage transitions. Complex multi-optimum models can fit sparse observations while leaving estimated optima and strength-of-selection parameters poorly determined. Pennell and colleagues (2015) argued that judging a model solely by its relative ranking among candidates neglects whether it can reproduce important features of the observed data. Posterior predictive or simulation-based adequacy checks, transparent regime assignments, and checks for influential taxa reduce this problem. If several evolutionary processes explain a psychological trait equally well, the report should retain that ambiguity rather than invent an adaptation narrative. Pennell et al. (2015) Uyeda et al. (2018)

Discrete Characters and Correlated Evolutionary Transitions

Many psychological claims are expressed as categorical properties: whether a species displays social transmission under defined conditions, demonstrates a particular form of cooperative behavior, or reaches a stated task criterion. In a discrete comparative model, categories are states with transitions that occur along branches according to a specified rate process. The state definition must be reproducible and distinguish true absence from missing observation. Coding a behavior as present or absent sacrifices information about its frequency, developmental conditionality, and task dependence. Such simplification can nevertheless be defensible when state definitions are anchored in comparable observation opportunities and model diagnostics acknowledge the loss of information. Maddison (1990) MacLean et al. (2012)

Maddison (1990) formulated a test for whether evolutionary changes in one binary character are concentrated on branches where another character has a specified state. This work established a direct way to study the historical association between transitions rather than merely correlate the terminal values. A reconstructed sequence of trait changes can suggest hypotheses about ordering: whether a particular social condition frequently precedes an innovation, for example. But state reconstructions are uncertain and inferred transition counts depend on assumptions about change rates and ancestral states. Correlated evolutionary transitions also leave open unmeasured common causes, ancestral confounders, and alternative transition processes. Maddison (1990)

Interpretation becomes particularly delicate when only a few gains or losses are observed. A species-rich tree may still supply little information about a binary transition that happened once in a deeply ancestral lineage. Repeated independent gains across distant branches offer stronger comparative leverage, provided observational criteria are equivalent. Coding a psychological trait separately at individual, population, and species levels can reveal whether an apparent evolutionary transition is instead a difference in prevalence or in whether researchers conducted appropriate tests. Transition models are most convincing when supplemented by natural history, experimental mechanisms, and explicit sensitivity analysis for ambiguous states. MacLean et al. (2012) Uyeda et al. (2018)

Ancestral-State Reconstruction and Historical Uncertainty

Ancestral-state reconstruction uses observations at the tips of a tree, together with its branch structure and an evolutionary process model, to estimate possible values or state probabilities at internal nodes. For a quantitative measure, reconstructed ancestral performance is a conditional statistical estimate under assumptions about how the measure evolved. For a categorical trait, reconstruction assigns likelihoods or posterior probabilities to possible states at specified ancestors. Because psychological test performance leaves no direct fossil record, uncertainty is unavoidable. Models may be informed indirectly by neuroanatomy, archaeological behavior, or ecological indicators, but such proxies demand their own measurement and interpretation theories. MacLean et al. (2012) Pagel (1999)

A confidence region around an ancestral estimate represents statistical uncertainty conditional on the adopted tree and model. It typically excludes entire categories of uncertainty unless they are explicitly modeled: construct validity, task equivalence, clade ascertainment, possible convergence, fossil calibration and omitted species. A statement that an ancestor possessed a familiar human psychological capacity therefore requires more than a reconstructed high score on a laboratory-derived trait. The more defensible language concerns the predicted distribution of the operational character under specified conditions. When competing models yield different reconstructions, the spread is scientifically informative and should be reported rather than averaged without explanation. Martins & Hansen (1997) Uyeda et al. (2018)

Ancestral reconstruction is most useful as a hypothesis-generating component of a broader historical investigation. It can identify lineages where trait changes are predicted, motivate comparative testing of extant relatives, and guide the search for independently supported mechanisms. MacLean and colleagues (2012) illustrated this approach in an example involving inhibitory control in primates. Their treatment also cautioned that a phylogeny containing relatively few species may offer weak power to detect signal and wide uncertainty around reconstructions. Historical specificity depends on the amount, distribution, and quality of comparative evidence, not on the visual precision of a reconstructed value on a tree. MacLean et al. (2012)

Individual Variation, Repeatability, and Species Estimates

Psychological traits are usually measured repeatedly in individuals, and those individuals belong to populations whose environments and histories differ. A species mean estimated from a handful of captive subjects can depend strongly on age, sex, husbandry, testing familiarity, recruitment and motivation. Within-species variation should therefore be preserved in the design and statistical model whenever possible. Ives, Midford, and Garland (2007) showed that ignoring such variation or measurement error can bias comparative parameter estimates and undermine precision. Treating a measured species mean as the exact evolutionary character privileges species with weak measurement in the same way as species with extensive sampling. Ives et al. (2007)

Measurement uncertainty can be estimated using sampling standard errors, hierarchical trial models, and repeated-testing reliability. Observational differences may reflect state fluctuations as well as stable individual variation. A useful design partitions variance among occasions, individuals, populations, species, laboratories, and phylogenetic branches. Some components can be estimated jointly while others require external information. Researchers should document how much within-species uncertainty is transmitted into comparative slopes or ancestral predictions, how missing data are handled, and whether uncertainty is comparable across species. A phylogenetic model that correctly handles evolutionary covariance remains vulnerable to poor individual-level measurement. Ives et al. (2007) Hadfield & Nakagawa (2010)

Reliable species-level psychology requires substantive attention to population variation. An average behavioral score in domesticated or zoo-housed individuals may differ from a score in wild populations through experience and developmental ecology. A phylogenetic species effect does not encompass the full range of local phenotypic reaction norms. When sufficient data exist, researchers can model variation among populations within species and distinguish ecological conditions from phylogenetic placement. When sampling is sparse, sensitivity to alternative species-mean estimates and transparent uncertainty intervals provide a more truthful result than a single apparently definitive branch-level point. MacLean et al. (2012) ManyPrimates et al. (2019)

Comparative Measurement, Equivalence, and Species-Appropriate Tasks

Cross-species experimental comparability is a substantive research problem. Standardizing apparatus dimensions or instructions does not guarantee equal perceptual difficulty, motor effort, reward value or ecological relevance. A transparent barrier can invite a prepotent straight approach, but the strength of that response may differ with visual processing and previous encounters with barriers. Kabadayi, Bobrowicz, and Osvath (2018) reviewed the detour paradigm and documented its use for multiple cognitive questions, including inhibition, planning and social learning. Their analysis makes it clear that a visually simple task need not isolate a single executive construct. Kabadayi et al. (2018)

A measurement model should specify which properties of a task are intended to remain invariant across species and which should be adapted to preserve functional demand. Apparatus equivalence and functional equivalence are related but different goals. Where multiple indicators are available, within-species reliability and cross-species factor patterns can test whether performance represents the same latent construct. Researchers may also compare predictions from task-specific scores with predictions from composite scores and examine whether a common pattern survives alternative scoring decisions. The success of a phylogenetic regression cannot compensate for a lack of evidence that its trait column measures a shared psychological phenomenon. MacLean et al. (2012) Kabadayi et al. (2018)

Collaborative designs increase the chance of detecting measurement problems that arise only when protocols travel across research sites. ManyPrimates introduced a distributed infrastructure for primate cognition studies and tested a shared short-term-memory paradigm across laboratories. The scientific value of such collaboration includes protocol documentation, richer taxonomic coverage, structured reporting of site variation, and opportunities to reassess construct equivalence. A shared task should be accompanied by records of departures from protocol, apparatus, experimenter effects, recruitment criteria, and exclusions. Statistical harmonization and laboratory replication must be designed together; neither can substitute for detailed observation of how animals interact with the experimental setting. ManyPrimates et al. (2019)

Missing Data, Ascertainment, and Trait-Coding Uncertainty

Missingness arises when species are untested, measured on different scales, excluded from a sample, or represented only by ambiguous observations. These categories carry distinct inferential consequences. A missing score is not a zero and should not be converted into evidence of inability. If laboratories preferentially study accessible or behaviorally remarkable species, observations may be missing in a way associated with ecological or cognitive characteristics. Analyses restricted to complete cases can then estimate the relationship for a selected population rather than the intended phylogenetic scope. Documentation of taxonomic coverage and the reasons for missing observations is part of the evidence base. ManyPrimates et al. (2019) MacLean et al. (2012)

Approaches to missing data include transparent complete-case analysis under a limited estimand, principled imputation within a justified model, joint hierarchical estimation and sensitivity analyses for informative missingness. Imputing cognition from phylogenetic proximity alone risks making the tree both the assumed source of the missing data and the evidence for the resulting evolutionary pattern. Analyses should report the proportion of taxa and trait cells observed, which lineages are excluded, and how conclusions vary under plausible missing-data mechanisms. An unobserved behavior should remain explicitly unobserved when the data contain no validated reason to classify it. Hadfield & Nakagawa (2010) Uyeda et al. (2018)

Coding uncertainty applies even when a publication exists. A researcher might extract an author's claim that a species uses tools without confirming whether the observation satisfies a predefined tool-use definition. Binary classification can conceal grade, context, and uncertainty of observation. For disputed records, analyses can propagate alternative coding scenarios or probabilistic state assignments, with the choices archived before the comparative tests are inspected. The evidence table should preserve sources and coding justifications for every species. Disagreement between defensible codings indicates a real limitation in the construct or the literature, rather than a numerical nuisance to be concealed by a single preferred spreadsheet. Maddison (1990) MacLean et al. (2012)

Phylogenetic Targeting, Sampling Strategy, and Study Power

Phylogenetic targeting begins with a specified evolutionary hypothesis and asks which lineages can discriminate it most efficiently. If two ecological hypotheses make different predictions among closely related species that occupy contrasting environments, sampling those lineages may reduce historical confounding. Replicating the comparison in independently diverged clades strengthens the generality of the result. Targeted sampling is not merely increasing the raw number of species; it prioritizes informative contrasts, independent transitions and variation in the focal predictor. The proposed design should still preserve the range of taxa needed for the population-level claim. MacLean et al. (2012)

Power in phylogenetic analysis depends on the structure of the tree, distribution of predictor changes, measurement precision, true effect magnitude and model assumptions. A data set with many tips can be weak for a predictor concentrated in one clade. Conversely, a modest tree may be informative when the contrast of interest recurs independently. An appropriate prospective analysis simulates trait outcomes on candidate trees using a stated effect and error model, then evaluates how often the intended analysis recovers the signal. Simulations conducted after observing significance may explain sensitivity, but they do not retroactively make a poorly designed study confirmatory. Freckleton et al. (2002) Münkemüller et al. (2012)

Selection should be documented before high-stakes interpretation. Researchers can predefine exclusion criteria, the taxonomic source and version of the tree, response indicators, ecology covariates and a model comparison scheme. Multiple sites may collect the same species to estimate laboratory variance, while related species from different environments can test ecological hypotheses. The methodological aim is to distribute evidence across explanatory dimensions instead of maximizing a conspicuous species count. Large collaborations expand research capacity, but their statistical benefit depends on balanced coverage, repeatable procedures and a question for which the sampled evolutionary transitions are genuinely informative. ManyPrimates et al. (2019) MacLean et al. (2014)

Empirical Case: Inhibitory Control Across Primate Lineages

Inhibitory control has become a frequent target for phylogenetic comparative psychology because researchers can pose observable behavioral challenges whose solutions require suppressing an immediate approach or response. Amici, Aureli, and Call (2008) investigated variation in behavioral flexibility and inhibitory control across primate species in relation to fission–fusion social dynamics. Their study illustrated an evolutionary hypothesis linking social organization to cognitive demands. Interpretation nevertheless depends on whether the experimental tasks isolate inhibition from learning, motivation, familiarity and sensory differences, and on how the social predictor varies across independent branches. An ecological correlation is an empirical pattern requiring further causal triangulation. Amici et al. (2008)

MacLean and colleagues (2012) illustrated phylogenetic analyses on an example data set of twelve primate species tested on an inhibitory-control task. They explored phylogenetic signal, regression and ancestral estimates, framing the analyses as demonstrations of how historical methods can enrich comparative experiments. Their example highlighted a problem that persists: sample sizes regarded as unusually broad within experimental psychology can be modest for inference about many evolutionary parameters. Researchers must therefore report uncertainty even when a phylogenetic method produces a visually compelling reconstructed history. The statistical methods explain patterns conditional on the chosen score and species panel; they do not adjudicate every psychological interpretation of the task. MacLean et al. (2012)

Comparative inhibition research must additionally distinguish levels of evidence. Individual test trials provide information about within-subject variability and learning; species estimates represent uncertain population summaries; a model associating those estimates with ecological factors describes across-lineage covariation. The transition from one level to another is an inferential step with its own assumptions. Controlled within-species experiments can test motivational and perceptual alternatives, while cross-species analyses can ask how often particular outcomes evolved. The strongest research programs treat laboratory mechanism and phylogenetic pattern as mutually informing inquiries rather than use a phylogenetic coefficient as a substitute for the psychological construct's validation. Ives et al. (2007) Kabadayi et al. (2018)

Empirical Case: The Multispecies Evolution of Self-Control

MacLean and collaborators' 2014 multispecies project directly compared behavioral performance across 567 individuals representing 36 species on two problem-solving tasks intended to measure self-control. They tested evolutionary explanations involving brain volume and socioecological variables and reported that absolute brain volume was the strongest predictor among the alternatives examined. The empirical advance was the integration of cross-species experimental tasks with phylogenetically informed analysis on a scale unusual for earlier comparative cognition research. The result identifies an association in the sampled species and tasks; it should not be interpreted as a universal causal law connecting absolute brain volume to every form of cognitive control. MacLean et al. (2014)

The experiment also reveals why broad sampling creates new measurement questions. Species differ in morphology, reward interest, prior training, age composition, and interaction with transparent or opaque apparatuses. An aggregate association involving brain volume can reflect multiple biological and behavioral processes, including general developmental and sensory constraints. Interpretation requires inspection of both tasks, species-specific patterns, potential influential lineages, residual phylogenetic covariance and predictor collinearity. The number of tested individuals is important for estimating performance, while the thirty-six species determine the principal scale of cross-species evolutionary inference. Increasing individuals in one species does not produce additional independent evolutionary transitions. MacLean et al. (2014) Ives et al. (2007)

Subsequent methodological analysis of detour tasks provides a complementary perspective. Kabadayi and colleagues (2018) reviewed the historical range of cognitive functions to which detouring has been attributed. This creates a constructive test of the interpretation: when outcomes are robust across different assays with distinct perceptual and motor demands, a broader inhibitory-control explanation becomes more persuasive. When conclusions rely on a specific apparatus, the narrower task-level result should remain visible. An evolutionary comparative model gains scientific meaning only when the psychological construct is independently defended. The MacLean evidence is an important comparative benchmark precisely because it supplies a testable framework for further replication and refinement. Kabadayi et al. (2018) MacLean et al. (2014)

Empirical Case: Collaborative Short-Term Memory Research

The ManyPrimates collaboration was designed in response to the fragmented taxonomic and institutional sampling of primate cognition. Its 2019 foundational report introduced an infrastructure for combining standardized tasks and scientific contributions across research sites, including a case study of short-term memory. The approach addresses one limitation of older comparative studies: isolated laboratories often test few species, collect incompatible measurements, and cannot separate species effects from laboratory-specific practices. Distributed coordination can produce richer data and clearer records of deviations, permitting stronger questions about whether a behavioral measure has the same meaning across different primates. ManyPrimates et al. (2019)

Collaborative evidence does not automatically eliminate site bias. Individuals at different facilities may have different histories of enrichment, testing, social housing and human interaction. A species present at only one laboratory can remain statistically confounded with that laboratory. Researchers must plan crossed or replicated sampling where feasible, document contextual factors and represent site variation in the analysis. An effect observed in multiple facilities with similar species is more informative about robustness than a pooled summary whose differences could reflect site composition. Repeated studies can also estimate reliability and quantify the effects of task modification on nominally identical cognitive constructs. ManyPrimates et al. (2019)

For phylogenetic inference, collaboration creates the possibility of integrating individual-level trials with species-level covariance rather than relying solely on published averages. It supports comparisons of task measurement, within-species uncertainty and evolutionary predictor effects within a coherent model. Such integration can also reveal that the same test is poorly calibrated for some taxa, in which case a single cross-species score would obscure rather than clarify evolutionary history. The research lesson extends beyond primates: shared protocols, species-appropriate adaptation and accessible data provenance are requirements for cumulative comparative psychology. Hadfield & Nakagawa (2010) ManyPrimates et al. (2019)

Trait Associations, Ecological Covariates, and Causal Interpretation

Phylogenetic regression often evaluates whether cognitive performance co-varies with social group characteristics, diet, lifespan, brain size, habitat or developmental schedules after accounting for modeled shared ancestry. Each covariate is itself a measured construct that can have phylogenetic structure and sampling error. Closely related predictors may yield unstable coefficient estimates when entered together. An interpretation that a social system caused a particular cognitive trait demands a causal explanation of when the system arose, whether the same association recurs across lineages and which developmental or selection mechanisms could produce it. A regression conditional on a tree is a structured association, not an automatic natural experiment. Freckleton et al. (2002) MacLean et al. (2012)

Historical covariation can reflect confounding through ecological inheritance, common developmental pathways or unmeasured selective pressures. A change in brain size and a change in behavioral performance can be coupled through body growth, metabolic constraints and sensory architecture. Models that condition on body mass can estimate a different quantity from models using absolute brain volume; the choice is theoretical rather than merely mathematical. If a variable is a mediator, collider or noisy proxy, including it may distort rather than clarify a causal hypothesis. Researchers should describe the assumed directional structure and test rival specifications whose differences have substantive meaning. MacLean et al. (2014) Uyeda et al. (2018)

The most convincing evolutionary claims combine multiple independent lines of evidence: mapped phylogenetic changes, observable task mechanisms, population ecology, developmental studies, and, where possible, fitness-related outcomes. Repeated gains of similar cognitive performance in ecologically comparable but distant lineages can strengthen hypotheses of convergent adaptation, yet similar endpoint behavior may have arisen through distinct neural or learning processes. A historical explanation becomes more precise by identifying which level of similarity is claimed. Apparent convergence of laboratory scores cannot by itself establish convergence of mental representation, selection history, or consciousness. MacLean et al. (2012) Butler & King (2004)

Effect Sizes, Model Comparison, and Independent Replication

An evidence review should distinguish the numerical size of a comparative association from its statistical detection. Effect sizes describe estimated changes in the psychological outcome under a specified change in the predictor, with units and transformation explicitly reported. Confidence or credible intervals summarize uncertainty under a model; their width depends on species coverage, covariate distribution, phylogenetic assumptions and measurement quality. A narrow interval from a misspecified process may be less informative than a wider interval from a model incorporating genuine uncertainty. Reports should include predictions, residual assessments and sensitivity analyses instead of relying solely on statistical significance. Ives et al. (2007) Revell (2010)

Model comparison ranks candidate statistical explanations conditional on the set of models considered. A winning model can still fit poorly. Pennell and collaborators (2015) developed model-adequacy analysis for macroevolutionary traits, emphasizing whether the model can generate data patterns resembling those observed. For cognitive traits, useful diagnostics include distributions of standardized residuals, extreme contrasts, influential tips, clade-specific rates, and discrepancies between simulated and empirical trait disparity. Comparing the same biological hypothesis across several plausible evolutionary covariance structures is more informative than selecting a single best likelihood score and treating it as evolutionary proof. Pennell et al. (2015)

Independent replication requires more than repeating software on the original data. A stronger test uses a new taxonomic panel, independent teams, validated tasks and a prespecified model capable of producing convergent predictions. Cross-study differences can be scientifically productive if researchers identify whether they arise from task demands, ecological environments, data-selection rules, branch calibration or actual evolutionary heterogeneity. Collaborative research, open data and independently implemented analyses increase the chance of discovering false certainty and improving a model. The goal of replication is a durable explanatory relationship with defined limits, not a ritualized repetition of the same coefficient. ManyPrimates et al. (2019) Uyeda et al. (2018)

An End-to-End Analysis Protocol for Psychological Characters

A defensible phylogenetic comparative analysis begins with a scientific question whose outcome and population of inference are explicitly identified. Investigators should define the psychological character independently of the predictor, specify how it will be observed, and record its expected sources of individual and contextual variation. Next they should construct a taxon-by-trait evidence table with species names, data provenance, sample sizes, testing conditions, coding uncertainty and potential conflicts between published reports. The phylogenetic tree must be independently justified, including its topology, branch-length units, uncertainty and coverage of the observed taxa. These steps determine whether a historical statistical analysis has a meaningful object before model fitting begins. MacLean et al. (2012) Garland et al. (1992)

The inferential workflow then specifies the principal estimand. For association between continuous traits, a PGLS or hierarchical comparative model may be suitable; for standardized divergence contrasts, independent contrasts offer transparent lineage comparisons; for ancestral values, a reconstruction model is required; for binary transitions, a discrete-state evolutionary model may be more appropriate. The model should accommodate known within-species uncertainty, check predictor dependence and missingness, and compare a planned set of evolutionary processes. Diagnostics must evaluate tree sensitivity, influential branches, heteroscedasticity, residual dependence and absolute model adequacy. Each analysis result should be accompanied by a statement of its conditional assumptions and the scope of taxa to which it applies. Felsenstein (1985) Hadfield & Nakagawa (2010) Pennell et al. (2015)

The final step integrates the historical estimates with psychological validation and the wider scientific evidence. A predictive association can motivate experiments, while a laboratory mechanism may explain how a historically correlated trait is expressed. Reports should provide reusable taxon tables, task protocols, tree identifiers, model formulas, analysis code and sensitivity outcomes, subject to ethical restrictions. If the data cannot discriminate a selection scenario from developmental constraint or shared ancestry, this is a substantive conclusion to preserve. A transparent comparative analysis yields knowledge by stating which evolutionary questions it answers and which remain beyond the resolution of available psychological observations. Uyeda et al. (2018) ManyPrimates et al. (2019)

Model Robustness, Reproducibility, and Sensitivity Design

Robustness in comparative psychology concerns more than the stability of a p value. At minimum, investigators should determine whether a claimed relationship survives plausible choices of phylogenetic tree, branch scaling, task coding, treatment of intraspecific variance and the set of taxa included. A coefficient that changes direction when one heavily sampled clade is removed merits a different interpretation from a relationship independently replicated across separated evolutionary lineages. Sensitivity analysis should be designed around known sources of uncertainty rather than a large unreported search for favorable specifications. Assessments of statistical power should simulate data under plausible sample sizes, effects, tree shapes and observation error. Münkemüller et al. (2012) Ives et al. (2007)

Absolute model adequacy is a central requirement. Pennell and collaborators (2015) showed how generating expectations from fitted evolutionary models can detect forms of trait variation a model cannot explain, even when it ranks above other candidates. Uyeda, Zenil-Ferguson, and Pennell (2018) argued for rethinking comparative analyses in light of process heterogeneity and limits of inference from present-day trait distributions. In psychological applications, an apparent regime shift may reflect a change in task ecology or measurement precision rather than a historical change in selection. Sensible checks compare biological interpretations under observationally plausible alternative processes and explicitly record when parameters are weakly identified. Pennell et al. (2015) Uyeda et al. (2018)

Open science can make comparative findings cumulative when deposited data include provenance rather than only numerical species summaries. Trials, exclusion rules, population descriptors, behavioral video coding criteria, measurement batteries, taxonomy mapping and tree construction decisions allow researchers to reassess claims. Independent laboratories should be able to reproduce the analysis and to test whether task implementation changes the results. Software provides repeatable calculation, while scientifically useful reproducibility additionally requires a stable account of what the variables mean. Collaborative networks may improve breadth and auditing, but they also require shared standards for handling animal welfare, data governance and site-specific restrictions. ManyPrimates et al. (2019) MacLean et al. (2012)

Comparative Psychology, Evolutionary Explanation, and Ethical Limits

Phylogenetic models and evolutionary narratives operate at different explanatory levels. A phylogenetic covariance model describes patterned resemblance in a historical tree. An experimental study can demonstrate how a behavior arises in a particular task. An adaptive hypothesis proposes a historical relationship between variation in traits and differential reproductive consequences. These claims can support one another, but none entails the others. A robust interpretation situates each result within the developmental histories, ecological environments and observational contexts of the organisms studied. It also recognizes the possibility that similar behaviors reflect different mechanisms and that one mechanism can generate multiple outcomes under changing conditions. MacLean et al. (2012) Hansen (1997)

Comparative measurement raises ethical obligations because experimental access is never independent of the welfare of the animals providing the data. A protocol that is appropriate for one taxon may impose unnecessary stress, deprivation or social disruption on another. Ecologically sensitive designs, voluntary participation where feasible, careful reinforcement practice and institutional review enhance the quality as well as the legitimacy of evidence. Behavioral outcomes can be distorted when testing conditions alter motivation or health, so welfare is also part of measurement validity. Reporting should distinguish wild observations, captive studies and structured experiments, and it should avoid translating task differences into hierarchical judgments about the worth or status of species.

The same restraint applies to public explanations of cognition. A poorly calibrated ranking of species from a few laboratory scores can be mistaken for an evolutionary ladder; modern phylogenetic analysis instead studies branching histories and locally defined traits. Differences in comparative self-control, memory or social learning do not establish a universal ordering of intelligence, subjective experience or moral status. Experimental data support operational behavioral conclusions, while claims about consciousness or mental continuity require separate conceptual and empirical arguments. Clear attribution of what was directly measured is therefore an ethical and methodological condition for responsible evolutionary psychology. Kabadayi et al. (2018) MacLean et al. (2014)

Artificial Era: Boundaries and Possibilities of Comparative Extension

The Artificial Era creates a new comparative question: which scientific methods designed for biological descent can inform the study of artificial cognitive and behavioral systems? Biological phylogenetic methods are constructed around genealogical relationships among organisms, stochastic trait evolution along branches and assumptions about inheritance. Artificial systems can have version histories, common training corpora, shared model architectures, fine-tuning branches and patterns of deployment. These relations can be documented and may induce correlated responses across systems. Any model of artificial comparative covariance must derive its assumptions from the actual engineering history; a biological species tree cannot simply be assigned to artificial agents without a demonstrated mapping of its statistical meaning.

A defensible extension would first define the unit of comparison: model checkpoint, deployed agent, architecture family, training lineage or persistent behavioral configuration. It would separately represent shared weights, common data, post-training intervention, external tools and evaluation context. Hierarchical or lineage-aware analyses could then investigate whether two related artificial systems behave more similarly than otherwise comparable systems, while controlling for benchmark contamination and duplicated data. These are prospective research hypotheses and methodological possibilities. They do not establish biological homology, inherited adaptive traits or psychological consciousness merely because an engineering lineage predicts measured performance.

Human–artificial relations add further layers to comparative inference. Observable performance in conversation may depend on users, institutional prompts, memory systems and tool environments, so the relevant unit may be a coupled human–artificial interaction rather than an isolated model. Multi-level repeated measures can disentangle variation among systems, sessions, users and tasks, but historical claims would require documented lineage data and experimental interventions. The authorial horizon of Artificial Era · From Homo to Artificial supplies a framework for asking these questions. Empirical validation remains specific to each proposed measure and causal mechanism; a borrowed phylogenetic vocabulary is useful only where the data-generating structure has been made explicit.

Recent Evidence Synthesis and the Reliability of Comparative Cognition

Recent research has made the weaknesses of comparative evidence aggregation increasingly visible. Mizuno, Lagisz, Pollo, Guillette, Soma, and Nakagawa (2025) mapped the meta-analytic literature on animal cognition and identified repeated problems in taxonomic coverage, incomplete demographic information, data transparency, and treatment of phylogenetic relatedness. Their study illustrates that many quantitative summaries of animal cognition do not yet fully integrate shared evolutionary descent into study-level inference. A meta-analysis of effect sizes can provide valuable pooled evidence about behavior, but its precision and generality depend on the independence structure, ecological heterogeneity, sampling frame, and validity of the constructs that generated those effects. These considerations apply especially when many primary studies concern a narrow set of laboratory model species. Mizuno et al. (2025)

Loyant, Collins, and Joly (2025) reviewed inhibitory-control tasks used in non-human animals and identified limitations in validity, temporal reliability, and the interpretation of varied task outcomes as manifestations of one unified ability. They proposed a staged methodological framework that begins with explicit construct definition and content validity, continues through repeatability assessment, and uses multiple tasks alongside control for individual, environmental, and apparatus-related influences. This framework directly affects phylogenetic analysis: a well-estimated covariance matrix can model historical nonindependence, but cannot transform an unstable score into a valid psychological character. Trait reliability is therefore a prerequisite for strong conclusions about evolutionary differentiation in inhibition. Loyant et al. (2025)

In September 2026, Mizuno, Pollo, Lagisz, Guillette, Soma, and Nakagawa reported a meta-meta-analytic reappraisal of animal cognition evidence, documenting fragility across their sampled published syntheses. Such a second-order analysis evaluates the reliability and coverage of an existing research literature; it does not directly determine whether a particular cognitive trait evolved through selection, drift, convergence, or conserved ancestry. Taken together, the 2025 and 2026 reviews support more transparent study selection, taxonomically balanced collaborative research, phylogenetically informed synthesis, and open handling of measurement quality. They also justify treating apparently precise evolutionary estimates with caution when the foundational experiments differ in task meaning or when study inclusion reflects a narrow research ecology. Mizuno et al. (2026) Mizuno et al. (2025) Loyant et al. (2025)

Conclusion

Phylogenetic comparative methods provide the statistical architecture for investigating the historical distribution, association and reconstruction of psychological traits across species. Their core achievement is to incorporate branching evolutionary relationships into the interpretation of observed variation. Independent contrasts, generalized least squares, signal estimation, ancestral reconstruction and discrete-transition models each answer distinct questions. Brownian and adaptive-process models offer conditional descriptions of trait change, while modern robustness analysis tests whether those descriptions are adequate to the evidence. Together these methods make psychological evolution experimentally and historically investigable with greater precision than unstructured species comparisons. Felsenstein (1985) Freckleton et al. (2002) MacLean et al. (2012)

The decisive methodological requirement is an integrated chain of validity. Psychological characters must have defensible operational definitions; species-level estimates must preserve individual uncertainty; trees and branch lengths must be justified; statistical models must accommodate nonindependence and competing processes; and causal interpretations must be supported by appropriate developmental, experimental and ecological evidence. The multispecies studies of inhibitory control and distributed primate cognition show what can be gained through collaboration, while the literature on measurement error, phylogenetic signal and model adequacy explains why sophisticated regression alone is insufficient. Evolutionary claims should remain calibrated to the traits, species and historical transitions actually observed. MacLean et al. (2014) ManyPrimates et al. (2019) Ives et al. (2007) Pennell et al. (2015)

The most productive research direction combines phylogenetic targeting with wider collaborative sampling, rigorous task equivalence, multilevel modeling and transparent data provenance. Future studies should systematically examine the sensitivity of historical explanations to alternate trait codings, uncertain trees, unmeasured ecological covariates and non-Brownian evolutionary processes. They should also make explicit the relationship between evolutionary pattern and the mechanisms through which cognition is expressed in particular organisms. Comparative psychology will advance when its historical models, experiments and scientific concepts can be independently checked and revised, yielding cumulative explanations of psychological diversity across the branching history of life. Uyeda et al. (2018) MacLean et al. (2012)

References

  • Amici, F., Aureli, F., & Call, J. (2008). Fission-fusion dynamics, behavioral flexibility, and inhibitory control in primates. Current Biology, 18(18), 1415–1419. https://doi.org/10.1016/j.cub.2008.08.020
  • Blomberg, S. P., Garland, T., Jr., & Ives, A. R. (2003). Testing for phylogenetic signal in comparative data: Behavioral traits are more labile. Evolution, 57(4), 717–745. https://doi.org/10.1111/j.0014-3820.2003.tb00285.x
  • Butler, M. A., & King, A. A. (2004). Phylogenetic comparative analysis: A modeling approach for adaptive evolution. The American Naturalist, 164(6), 683–695. https://doi.org/10.1086/426002
  • Felsenstein, J. (1985). Phylogenies and the comparative method. The American Naturalist, 125(1), 1–15. https://doi.org/10.1086/284325
  • Freckleton, R. P., Harvey, P. H., & Pagel, M. (2002). Phylogenetic analysis and comparative data: A test and review of evidence. The American Naturalist, 160(6), 712–726. https://doi.org/10.1086/343873
  • Garland, T., Jr., Harvey, P. H., & Ives, A. R. (1992). Procedures for the analysis of comparative data using phylogenetically independent contrasts. Systematic Biology, 41(1), 18–32. https://doi.org/10.1093/sysbio/41.1.18
  • Hadfield, J. D., & Nakagawa, S. (2010). General quantitative genetic methods for comparative biology: Phylogenies, taxonomies and multi-trait models for continuous and categorical characters. Journal of Evolutionary Biology, 23(3), 494–508. https://doi.org/10.1111/j.1420-9101.2009.01915.x
  • Hansen, T. F. (1997). Stabilizing selection and the comparative analysis of adaptation. Evolution, 51(5), 1341–1351. https://doi.org/10.1111/j.1558-5646.1997.tb01457.x
  • Ives, A. R., Midford, P. E., & Garland, T., Jr. (2007). Within-species variation and measurement error in phylogenetic comparative methods. Systematic Biology, 56(2), 252–270. https://doi.org/10.1080/10635150701313830
  • Kabadayi, C., Bobrowicz, K., & Osvath, M. (2018). The detour paradigm in animal cognition. Animal Cognition, 21(1), 21–35. https://doi.org/10.1007/s10071-017-1152-0
  • Loyant, L., Collins, L., & Joly, M. (2025). Inhibitory control tests in non-human animals: Validity, reliability, and perspectives. Biological Reviews, 100(6), 2482–2507. https://doi.org/10.1111/brv.70055
  • MacLean, E. L., Hare, B., Nunn, C. L., Addessi, E., Amici, F., Anderson, R. C., Aureli, F., Baker, J. M., Bania, A. E., Barnard, A. M., Boogert, N. J., Brannon, E. M., Bray, E. E., Bray, J., Brent, L. J. N., Burkart, J. M., Call, J., Cantlon, J. F., Cheke, L. G., … Zhao, Y. (2014). The evolution of self-control. Proceedings of the National Academy of Sciences, 111(20), E2140–E2148. https://doi.org/10.1073/pnas.1323533111
  • MacLean, E. L., Matthews, L. J., Hare, B., Nunn, C. L., Anderson, R. C., Aureli, F., Brannon, E. M., Call, J., Drea, C. M., Emery, N. J., Haun, D. B. M., Herrmann, E., Jacobs, L. F., Platt, M. L., Rosati, A. G., Sandel, A. A., Schroepfer, K. K., Seed, A. M., Tan, J., … Wobber, V. (2012). How does cognition evolve? Phylogenetic comparative psychology. Animal Cognition, 15(2), 223–238. https://doi.org/10.1007/s10071-011-0448-8
  • Maddison, W. P. (1990). A method for testing the correlated evolution of two binary characters: Are gains or losses concentrated on certain branches of a phylogenetic tree? Evolution, 44(3), 539–557. https://doi.org/10.1111/j.1558-5646.1990.tb05937.x
  • ManyPrimates, Altschul, D. M., Beran, M. J., Bohn, M., Call, J., DeTroy, S., Duguid, S., Egelkamp, C. L., Fichtel, C., Fischer, J., Flessert, M., Hanus, D., Haun, D. B. M., Haux, L. M., Hernandez-Aguilar, R. A., Herrmann, E., Hopper, L. M., Joly, M., Kano, F., … Watzek, J. (2019). Establishing an infrastructure for collaboration in primate cognition research. PLOS ONE, 14(10), Article e0223675. https://doi.org/10.1371/journal.pone.0223675
  • Martins, E. P., & Hansen, T. F. (1997). Phylogenies and the comparative method: A general approach to incorporating phylogenetic information into the analysis of interspecific data. The American Naturalist, 149(4), 646–667. https://doi.org/10.1086/286013
  • Mizuno, A., Lagisz, M., Pollo, P., Guillette, L., Soma, M., & Nakagawa, S. (2025). Systematic mapping and bibliometric analysis of meta-analyses on animal cognition. Neuroscience & Biobehavioral Reviews, 177, Article 106342. https://doi.org/10.1016/j.neubiorev.2025.106342
  • Mizuno, A., Pollo, P., Lagisz, M., Guillette, L., Soma, M., & Nakagawa, S. (2026). Statistical fragility in the meta-analytic evidence base of animal cognition: A meta-meta-analytic reappraisal. Biology Open. Advance online publication. https://doi.org/10.1242/bio.062841
  • Münkemüller, T., Lavergne, S., Bzeznik, B., Dray, S., Jombart, T., Schiffers, K., & Thuiller, W. (2012). How to measure and test phylogenetic signal. Methods in Ecology and Evolution, 3(4), 743–756. https://doi.org/10.1111/j.2041-210X.2012.00196.x
  • Pagel, M. (1999). Inferring the historical patterns of biological evolution. Nature, 401(6756), 877–884. https://doi.org/10.1038/44766
  • Pennell, M. W., FitzJohn, R. G., Cornwell, W. K., & Harmon, L. J. (2015). Model adequacy and the macroevolution of angiosperm functional traits. The American Naturalist, 186(2), E33–E50. https://doi.org/10.1086/682022
  • Rabosky, D. L. (2015). No substitute for real data: A cautionary note on the use of phylogenies from birth–death polytomy resolvers for downstream comparative analyses. Evolution, 69(12), 3207–3216. https://doi.org/10.1111/evo.12817
  • Revell, L. J. (2010). Phylogenetic signal and linear regression on species data. Methods in Ecology and Evolution, 1(4), 319–329. https://doi.org/10.1111/j.2041-210X.2010.00044.x
  • Uyeda, J. C., Zenil-Ferguson, R., & Pennell, M. W. (2018). Rethinking phylogenetic comparative methods. Systematic Biology, 67(6), 1091–1109. https://doi.org/10.1093/sysbio/syy031