Correcting for Allometry in Taxonomic Geometric Morphometrics: A Comprehensive Guide for Accurate Species Delimitation

Abstract

This article provides a comprehensive framework for understanding and correcting allometric effects in taxonomic geometric morphometric studies. It covers foundational concepts of size and shape, explores methodological approaches for allometry correction, addresses common troubleshooting scenarios, and establishes validation protocols. Designed for researchers in evolutionary biology and systematics, this guide integrates theoretical principles with practical applications using widely adopted software tools to ensure accurate species identification and delimitation by isolating true taxonomic signal from size-related shape variation.

Understanding Allometry: From Core Concepts to Taxonomic Implications

Allometry, the study of size-related changes in morphology, is a foundational concept in evolutionary and developmental biology [1]. In taxonomic studies using geometric morphometrics (GMM), understanding and correcting for allometric variation is crucial for accurately identifying evolutionarily significant units and delineating taxa [2]. The morphological differences observed among populations or species often contain a substantial component that is correlated with, or driven by, differences in overall size. Failure to account for these allometric effects can lead to misinterpretations of phylogenetic relationships and taxonomic status. Currently, two predominant schools of thought guide methodological approaches to allometry: the Gould-Mosimann school and the Huxley-Jolicoeur school [1] [3] [4]. These frameworks differ in their fundamental definitions of allometry and their implementation in geometric morphometric analyses, yet both provide powerful tools for taxonomic research. This article provides a detailed comparison of these approaches and protocols for their application in taxonomic studies.

Theoretical Foundations: A Tale of Two Schools

The Gould-Mosimann School

The Gould-Mosimann framework defines allometry specifically as the covariation between shape and size [1] [3]. This conceptualization requires a clear separation between size and shape, following the criterion of geometric similarity, where shape is defined as "all geometric information that remains when location, scale, and rotational effects are filtered out from an object" [3]. In this school, allometry is quantified by analyzing how shape variables change in relation to a measure of overall size.

• Key Tenet: Shape and size are distinct biological constructs that can be separated analytically.
• Basis: Allometry results from the correlation between these separate entities.
• Taxonomic Implication: This approach allows taxonomists to ask, "After removing size variation, what shape differences remain that might characterize taxonomic groups?"

The Huxley-Jolicoeur School

The Huxley-Jolicoeur school characterizes allometry as the covariation among morphological traits that all contain size information [1] [3] [4]. This framework does not presuppose an a priori separation of size and shape, but rather considers the organismal form as an integrated whole. Allometric patterns emerge from the coordinated variation of multiple traits in response to size variation.

• Key Tenet: Organismal form is a unified entity; size and shape are interrelated components.
• Basis: Allometry manifests as the dominant axis of covariation among morphological features.
• Taxonomic Implication: This approach enables taxonomists to identify the primary axis of morphological integration that may reflect underlying growth processes or functional constraints.

Table 1: Conceptual Comparison of the Two Allometric Schools

Aspect	Gould-Mosimann School	Huxley-Jolicoeur School
Definition of Allometry	Covariation of shape with size	Covariation among morphological features containing size information
Size-Shape Relationship	Separate entities that covary	Integrated components of form
Analytical Implementation	Multivariate regression of shape on size	First principal component in form space
Morphospace Used	Shape tangent space	Conformation space (size-and-shape space)
Primary Output	Allometric slope (regression vector)	Allometric trajectory (PC1)
Taxonomic Application	Size correction to reveal non-allometric shape differences	Identification of major axes of morphological variation

Methodological Implementation in Geometric Morphometrics

Gould-Mosimann Protocol: Shape-Size Regression

The Gould-Mosimann approach is implemented through multivariate regression of shape coordinates on a size measure, typically centroid size [1] [3] [4].

Step-by-Step Protocol:

• Data Collection: Digitize landmarks from all specimens. For 2D data, ensure consistent orientation and scale during imaging [2].
• Procrustes Superimposition:
  • Perform Generalized Procrustes Analysis (GPA) to align specimens by translating, rotating, and scaling them to a common coordinate system.
  • Centroid size (the square root of the sum of squared distances of landmarks from their centroid) is computed as part of this process.
  • The resulting Procrustes coordinates represent shape variables.
• Size Measurement: Use centroid size as the size variable. Log-transform if necessary to linearize relationships.
• Multivariate Regression: Regress the Procrustes shape coordinates (dependent variables) on centroid size (independent variable) using multivariate regression techniques.
• Visualization: Visualize the allometric pattern as a deformation from the consensus configuration in the direction of the regression vector.

Huxley-Jolicoeur Protocol: Form Space PCA

The Huxley-Jolicoeur approach identifies allometry as the primary axis of variation in form space, where size remains incorporated [1] [4].

Step-by-Step Protocol:

• Data Collection: Collect landmark data as in Step 3.1.1.
• Form Space Superimposition:
  • Perform Procrustes superimposition without scaling (or use Boas coordinates) [4].
  • This preserves size variation in the data, resulting in "conformation" or "size-and-shape" space.
• Principal Component Analysis: Perform PCA on the form space coordinates.
• Allometric Vector Identification: The first principal component (PC1) typically represents the allometric vector, as size variation often accounts for the largest proportion of morphological variance.
• Validation: Confirm the allometric interpretation of PC1 by correlating PC1 scores with centroid size.
• Visualization: Visualize shape changes along the PC1 axis to interpret allometric trajectories.

Table 2: Comparison of Analytical Protocols

Protocol Step	Gould-Mosimann Approach	Huxley-Jolicoeur Approach
Data Preprocessing	Generalized Procrustes Analysis with scaling	Procrustes alignment without scaling OR use of Boas coordinates
Size Representation	External variable (centroid size)	Intrinsic to the data structure
Allometry Detection	Multivariate regression of shape on size	PCA on form space coordinates
Allometry Quantification	Regression vector and Goodall's F-test	PC1 loadings and variance explained
Statistical Testing	Permutation test for regression significance	Correlation of PC1 with size measures
Visualization	Predicted shapes along size gradient	Shape changes along PC1 axis

Experimental Design and Workflow for Taxonomic Studies

The following diagram illustrates the decision pathway for selecting and implementing allometric analyses in taxonomic geometric morphometrics:

Figure 1: Decision workflow for selecting appropriate allometric analysis methods in taxonomic geometric morphometric studies.

Table 3: Essential Research Reagent Solutions for Allometric Studies in Geometric Morphometrics

Tool/Resource	Type	Function in Allometric Analysis	Implementation Examples
Landmark Digitation Software	Software	Capture morphological coordinates from specimens	tpsDig2, MorphoJ, IMP suites
Procrustes Superimposition Algorithms	Computational Method	Remove non-shape variation (position, rotation) prior to Gould-Mosimann analysis	GPA in MorphoJ, geomorph R package
Centroid Size Calculation	Size Metric	Standardized measure of size independent of shape; used as independent variable in Gould-Mosimann approach	Computed during Procrustes analysis
Form Space Coordinates	Data Structure	Preserve size information for Huxley-Jolicoeur analyses; alternative to traditional shape space	Boas coordinates, Procrustes analysis without scaling
Multivariate Regression Algorithms	Statistical Tool	Quantify relationship between shape and size in Gould-Mosimann framework	procD.lm in geomorph, lm in R with Procrustes coordinates
Principal Component Analysis (PCA)	Multivariate Method	Identify major axes of variation in form space for Huxley-Jolicoeur approach	PCA in MorphoJ, R prcomp function
Permutation Testing Frameworks	Statistical Validation	Assess significance of allometric relationships non-parametrically	Residual randomization in geomorph, MorphoJ
Shape Visualization Tools	Graphical Output	Display allometric vectors as deformation grids or 3D models	Vector displacement diagrams, thin-plate splines

Taxonomic Applications: Correcting for Allometry in Practice

In taxonomic studies, the choice between allometric frameworks depends on the specific research question. The Gould-Mosimann approach is particularly valuable when the goal is to remove size variation to reveal shape differences potentially indicative of taxonomic boundaries [2]. For example, when comparing populations that differ substantially in body size, this method can determine whether shape differences are merely allometric consequences of size variation or represent independent evolutionary events.

Conversely, the Huxley-Jolicoeur approach provides insights into patterns of morphological integration that may reflect shared developmental or functional constraints within lineages. This can inform taxonomic decisions by revealing whether groups share common allometric trajectories, potentially indicating close evolutionary relationships, or exhibit divergent trajectories suggestive of independent lineages.

Both methods have demonstrated utility in mammalian taxonomy. Studies of marmot mandibles [2] and rat crania [4] have successfully employed these approaches to disentangle allometric components from taxonomic signal. The protocols outlined herein provide a rigorous framework for implementing these analyses in novel taxonomic contexts.

The Gould-Mosimann and Huxley-Jolicoeur schools offer complementary perspectives on allometry in geometric morphometrics. While the Gould-Mosimann approach provides a powerful framework for size correction in taxonomic studies, the Huxley-Jolicoeur approach reveals fundamental patterns of morphological integration. Taxonomists should select the approach most aligned with their specific research questions, and may benefit from implementing both frameworks to obtain a comprehensive understanding of morphological variation in their study systems. The protocols detailed herein provide a rigorous foundation for such analyses, supporting robust taxonomic decisions grounded in comprehensive morphological analysis.

In geometric morphometrics (GM), the precise quantification of biological form relies on the interdependent concepts of size, shape, and form. Shape is defined as the geometric properties of an object that are invariant to location, scale, and rotation, while size represents the scalar component that scale invariance removes. Form encompasses both size and shape, preserving their biological interplay [5]. This distinction is paramount in taxonomic studies, where isolating shape for phylogenetic inference or understanding how shape changes with size (allometry) are common objectives. Correcting for allometry—the relationship between shape and size—is particularly crucial in taxonomy to distinguish true taxonomic signals from size-dependent morphological variation [6]. The following sections detail the operationalization of these concepts, provide a protocol for allometry correction, and discuss the impact of data quality on taxonomic conclusions.

Operational Definitions and Their Quantitative Frameworks

Defining Size, Shape, and Form Mathematically

• Form: Form is the total morphological configuration, represented by the original landmark coordinates. In practice, forms are the raw data before any Procrustes superimposition.
• Size: The most common metric for size in GM is Centroid Size, calculated as the square root of the sum of squared distances between each landmark and the object's centroid [7]. This measure is statistically independent of shape under certain conditions and is used as a standard size variable in allometric studies.
• Shape: Shape is quantified through Procrustes-aligned coordinates. The process of Generalized Procrustes Analysis (GPA) standardizes configurations by translating them to a common origin, scaling them to unit Centroid Size, and rotating them to minimize the sum of squared distances between corresponding landmarks [5] [8]. The resulting Procrustes coordinates reside in a non-Euclidean tangent space where standard multivariate statistics can be applied.

Table 1: Core Concepts in Geometric Morphometrics

Concept	Mathematical Definition	Biological Interpretation	Role in Taxonomic Studies
Form	Original landmark coordinates	The complete morphological structure	Serves as the raw data; contains both size and shape information.
Size	Centroid Size (CS)	A geometric scale factor	Used to study allometry; can be a confounding variable in shape analysis.
Shape	Procrustes Aligned Coordinates	Configuration after removing location, scale, and rotation	The primary data for discriminating taxa after correcting for allometry.
Allometry	Regression of shape on size (e.g., logCS)	The pattern of shape change correlated with size change	Must be accounted for to avoid misinterpreting size-related shape changes as taxonomic signals.

Application Note: A Protocol for Correcting Allometry in Taxonomic Studies

Correcting for allometry ensures that shape differences used for taxonomic discrimination are not merely a byproduct of size variation. This protocol is adapted from methods used in fossil and modern taxa [7] [9].

Experimental Workflow

The following diagram outlines the logical workflow for processing specimens and correcting for allometric effects in a geometric morphometric study.

Detailed Methodology

Stage 1: Data Acquisition and Landmarking

• Specimen Imaging: Capture high-resolution 2D or 3D images using standardized equipment (e.g., digital cameras, micro-CT scanners). Maintain consistent specimen presentation to minimize methodological error [5].
• Landmark Digitization: Place homologous landmarks (Type I, II, or III) and semi-landmarks on curves and surfaces using software like TpsDig2 or MorphoDig. To reduce intra-observer error, the same trained individual should place all landmarks multiple times, using the mean configuration for analysis [7] [5].

Stage 2: Data Preprocessing and Allometry Correction

• Calculate Centroid Size: Compute the Centroid Size for each specimen from the raw landmark coordinates.
• Perform Generalized Procrustes Analysis (GPA): Align all landmark configurations using GPA. This produces Procrustes shape coordinates, which are the dependent variables for allometry correction.
• Correct for Allometry:
  • Perform a multivariate regression of the Procrustes shape coordinates on Centroid Size (often log-transformed) [7] [9].
  • The null hypothesis is that shape is independent of size. A significant regression indicates the presence of allometry.
  • Extract the residuals from this regression. These residuals represent the portion of shape variation that is not explained by size, i.e., the allometry-corrected shapes [7].

Stage 3: Downstream Taxonomic Analysis

• Use the allometry-corrected shape residuals in subsequent analyses such as Principal Component Analysis (PCA) to visualize shape variation, or Linear Discriminant Analysis (LDA) for taxonomic classification [8] [9].
• These analyses will now reflect shape differences that are more likely to be genetic or phylogenetic in origin, rather than consequences of size differences.

Research Reagent Solutions and Essential Materials

Table 2: Essential Toolkit for a Geometric Morphometrics Study

Item Category	Specific Examples	Function in Research
Imaging Equipment	Digital SLR camera, micro-CT scanner, 3D laser scanner	Creates high-fidelity 2D/3D digital representations of specimens for measurement.
Digitization Software	TpsDig2, MorphoJ, R (geomorph package)	Used to place landmarks and semi-landmarks on digital images.
Statistical Software	R (with geomorph, Morpho packages), PAST	Performs core GM analyses: Procrustes superimposition, regression, PCA, and visualization.
Landmark Types	Type I (homologous junctions), Type II (maxima of curvature), Semi-landmarks	Quantify the geometry of biological forms in a comparable way across specimens.

Critical Considerations for Robust Taxonomic Inference

Managing Measurement Error

Measurement error is a significant, though often underreported, confounder in GM. It can arise from various sources and, if unaccounted for, can be misinterpreted as biological signal [5].

Table 3: Sources and Mitigation of Measurement Error in GM

Error Source	Type	Impact on Data	Recommended Mitigation
Specimen Presentation	Methodological	Projection distortion can displace landmark positions.	Standardize imaging angle and distance for all specimens [5].
Imaging Device	Instrumental	Different lenses/scanners introduce unique distortions.	Use the same imaging equipment and settings throughout the study [5].
Inter-observer Error	Personal	Different operators place landmarks inconsistently.	Have a single, trained individual digitize all specimens [7] [5].
Intra-observer Error	Personal	The same operator is inconsistent over time.	Digitize each specimen multiple times and use the average configuration [7].

Advanced Topics and Future Directions

• Weighted Covariance Estimates: Standard Procrustes methods assume homogeneous noise across all landmarks. Advanced methods incorporate landmark-specific measurement covariance, providing greater statistical stability and efficiency, especially for semi-landmarks [10].
• Automated Phenotyping: To overcome observer bias and landmark definition limitations, novel automated methods like morphVQ are emerging. These techniques use descriptor learning to establish dense correspondence across entire surfaces, capturing more comprehensive morphological variation without manual landmarking [11].
• Out-of-Sample Classification: A common challenge is classifying new specimens not included in the original study. This requires projecting the new specimen's raw coordinates into the shape space of the training sample using a predefined template, rather than re-running the GPA with the entire dataset [8].

A rigorous understanding of size, shape, and form is the foundation of any taxonomic study using geometric morphometrics. By implementing a structured protocol that includes allometry correction and robust error mitigation strategies, researchers can ensure that the taxonomic signals they identify are biologically meaningful and not artifacts of size variation or methodological inconsistency. As the field evolves with automated methods and more sophisticated statistical tools, the ability to disentangle complex morphological patterns will continue to improve, leading to more refined and accurate taxonomic classifications.

In taxonomic studies using geometric morphometrics, the influence of allometry—the relationship between size and shape—is a critical consideration that can determine the validity of scientific conclusions. When comparing groups, failure to account for allometric effects can lead to spurious group differences, where observed morphological distinctions actually reflect underlying size variation rather than genuine taxonomic signals. The foundational concepts of allometry are rooted in two main schools of thought: the Gould-Mosimann school, which defines allometry as the covariation between size and shape, and the Huxley-Jolicoeur school, which characterizes allometry as covariation among morphological features that all contain size information [3]. In practical taxonomic terms, allometry matters because species or populations often differ in body size, and these size differences can drive associated shape changes that might be mistakenly interpreted as independent taxonomic characters. This application note provides a structured framework for identifying, quantifying, and correcting for allometric effects in taxonomic studies using geometric morphometrics, ensuring that reported group differences reflect genuine morphological distinctions rather than size-correlated variation.

Theoretical Framework: Concepts and Schools of Thought

Competing Paradigms in Allometric Studies

The interpretation of allometry in morphological research is guided by two distinct but complementary philosophical frameworks:

• Gould-Mosimann School: This approach rigorously separates size and shape according to the criterion of geometric similarity. It defines allometry specifically as the covariation of shape with size, typically implemented through multivariate regression of shape variables on a measure of size [4] [3]. This framework operates primarily in shape space, where size is external to the shape representation, making it particularly useful for questions about how shape depends on size independently of other factors.

• Huxley-Jolicoeur School: This paradigm characterizes allometry as the covariation among morphological features that all contain size information, without pre-separating size and shape components [3]. In this framework, allometric trajectories are characterized by the first principal component of morphological variation, implemented in geometric morphometrics using either Procrustes form space or conformation space (also known as size-and-shape space) [4] [3]. This approach is valuable when researchers wish to understand integrated size-shape variation without artificial separation.

Despite their philosophical differences, these frameworks are logically compatible and unlikely to yield contradictory results when applied appropriately to taxonomic questions [3]. The choice between them should be guided by specific research questions rather than perceived superiority.

Levels of Allometric Variation in Taxonomic Contexts

Allometric patterns can manifest at different biological levels, each with distinct implications for taxonomic interpretation:

• Ontogenetic Allometry: Shape changes correlated with size variation during growth; particularly relevant when comparing taxa at different developmental stages or with different growth trajectories [3].

• Static Allometry: Shape-size relationships within a single ontogenetic stage, typically adults from a population; most commonly applied in taxonomic studies comparing adult specimens across groups [3].

• Evolutionary Allometry: Shape changes correlated with size differences across evolutionary lineages; critical for understanding how size evolution has driven morphological diversification in taxonomic groups [3].

Each level requires specific analytical approaches, and confounding these levels can lead to misinterpretation of taxonomic patterns. For instance, evolutionary allometry might be obscured if analyses inadvertently include ontogenetic variation.

Methodological Approaches for Allometric Analysis

Core Analytical Techniques

Four primary methods have emerged for estimating allometric vectors from landmark data, each with particular strengths for taxonomic applications:

• Multivariate Regression of Shape on Size: This Gould-Mosimann approach involves regressing Procrustes shape coordinates onto centroid size (or log-transformed centroid size) to isolate the component of shape variation that is predicted by size [4] [3]. The regression vector represents the allometric trajectory, and the residuals provide size-corrected shape data for taxonomic comparisons.

• First Principal Component (PC1) of Shape: In this approach, principal component analysis is performed on shape coordinates, and the association between PC1 scores and size is evaluated [4]. A strong correlation suggests that the major axis of shape variation represents allometry, which should be accounted for in subsequent taxonomic analyses.

• PC1 in Conformation Space: This Huxley-Jolicoeur method analyzes landmark configurations in size-and-shape space (without size normalization) and uses the first principal component as the allometric vector [4] [3]. This approach captures integrated size-shape covariation without pre-separating these components.

• PC1 of Boas Coordinates: A recently proposed method that uses the first principal component of Boas coordinates (similar to conformation space) to estimate allometric vectors [4]. Simulations show this method performs very similarly to the conformation space approach, with marginal differences in performance.

Performance Comparison of Allometric Methods

Computer simulations comparing these four methods under controlled conditions provide guidance for selecting appropriate analytical approaches [4]:

Table 1: Performance Comparison of Allometric Methods Under Different Variation Patterns

Method	Isotropic Residual Variation	Anisotropic Residual Variation	Theoretical Framework
Multivariate Regression of Shape on Size	Consistently better performance	Consistently better performance	Gould-Mosimann
PC1 of Shape	Lower performance	Lower performance	Gould-Mosimann
PC1 in Conformation Space	Very close to simulated vectors	Very close to simulated vectors	Huxley-Jolicoeur
PC1 of Boas Coordinates	Very close to simulated vectors	Very close to simulated vectors	Huxley-Jolicoeur

These results suggest that multivariate regression generally provides the most accurate estimation of allometric vectors under various noise conditions, while conformation space and Boas coordinates methods also perform well [4].

Practical Experimental Protocol for Taxonomic Studies

Complete Workflow for Allometric Analysis in Taxonomy

Step-by-Step Implementation Guide

• Landmark Data Collection

  • Digitize two-dimensional or three-dimensional landmarks representing biologically homologous points across all specimens
  • Include semilandmarks for curves and surfaces as needed, following standard sliding protocols
  • Ensure adequate sample sizes (minimum 20-30 specimens per group, with larger samples for complex allometric patterns)

• Generalized Procrustes Analysis (GPA)

  • Translate all configurations to a common centroid
  • Scale configurations to unit centroid size (root summed squared distance of landmarks from their centroid)
  • Rotate configurations to minimize the summed squared differences between each configuration and the mean shape [12]
  • This step produces Procrustes shape coordinates for shape space analyses

• Centroid Size Calculation

  • Compute centroid size for each specimen as the square root of the sum of squared distances of all landmarks from their centroid [12]
  • Log-transform centroid size if allometric relationships appear nonlinear on initial plots

• Allometry Assessment

  • Perform multivariate regression of Procrustes shape coordinates on centroid size
  • Test significance using permutation tests (typically 1000-10000 permutations)
  • Calculate the percentage of shape variance explained by size (goodness-of-fit statistic)
  • For Huxley-Jolicoeur approaches, perform PCA in conformation space and examine PC1-size correlations

• Size Correction Procedures (if significant allometry detected)

  • Regression Residual Method: Use residuals from the multivariate regression of shape on size as size-corrected shape data [3]
  • Prediction-Based Method: Predict shape values at a common size (e.g., group mean size) using the allometric vector
  • Validate that size correction successfully removes the association between shape and size in the corrected data

• Taxonomic Comparisons

  • Proceed with standard taxonomic analyses (MANOVA, discriminant analysis, cluster analysis) on size-corrected data
  • Compare results with uncorrected analyses to determine how allometry affects taxonomic conclusions
  • Explicitly report whether and how allometry was addressed in all taxonomic interpretations

Research Reagent Solutions for Allometric Studies

Table 2: Essential Tools and Software for Allometric Analysis in Geometric Morphometrics

Tool/Software	Primary Function	Application in Allometric Studies	Availability
MorphoJ	Geometric morphometrics analysis	Multivariate regression of shape on size; PCA of shape variables; permutation tests	Free download
R (geomorph package)	Comprehensive morphometric analysis	Procrustes ANOVA; allometric trajectory comparisons; modularity tests	Open source
R (Morpho package)	Shape analysis and manipulation	Procrustes registration; PCA; regression diagnostics	Open source
tps系列软件	Landmark digitization and basic analysis	Data collection and preliminary visualization; semilandmark placement	Freeware
EVAN Toolbox	Paleontological and anthropological morphometrics	Allometric scaling visualization; comparative analyses	Free download
PAST	Paleontological statistics	Multivariate statistics including PCA and regression; basic shape analysis	Freeware

Case Study: Marmot Mandible Taxonomy with Allometric Considerations

A practical example from marmot mandible taxonomy illustrates the critical importance of allometric assessment in taxonomic studies. In comparisons of North American marmot species, researchers found that while mandibular shape was an accurate predictor of taxonomic affiliation, allometry in adults explained only a modest amount of within-species shape change [13]. However, there was a degree of divergence in allometric trajectories that seemed consistent with subgeneric separation, suggesting that allometric patterns themselves can provide taxonomically informative characters [13].

This case highlights two key insights:

• Failure to account for allometry could have led to overestimation of taxonomic differences if size variation differed systematically among groups
• Allometric trajectories themselves (not just size-corrected shape) may contain taxonomically relevant information, particularly for understanding evolutionary diversification

The Vancouver Island marmot emerged as the most distinctive species for mandibular shape, but allometric analysis helped confirm that this distinctiveness persisted after accounting for size variation, strengthening the taxonomic interpretation [13].

Interpretation Guidelines and Reporting Standards

Evaluating Allometric Effects in Taxonomic Contexts

When interpreting allometric analyses in taxonomic studies, several guidelines ensure robust conclusions:

• Magnitude Matters: Report both statistical significance (p-values) and biological significance (effect size) of allometric relationships. A statistically significant allometric relationship with minimal explanatory power (e.g., <5% shape variance) may not require correction in taxonomic analyses.

• Consistency Across Groups: Test whether allometric trajectories differ significantly among taxonomic groups using methods such as multivariate analysis of covariance (MANCOVA) or trajectory analysis [13]. Differing allometric patterns can themselves be taxonomically informative.

• Biological Plausibility: Consider whether observed allometric relationships make functional, developmental, or ecological sense. Unexpected allometric patterns may indicate data quality issues or particularly interesting biological phenomena worth highlighting.

Minimum Reporting Standards for Taxonomic Publications

To ensure reproducibility and proper interpretation, taxonomic studies using geometric morphometrics should report:

• Complete Methods Description: Specify which allometric framework and analytical methods were used, with software and specific functions
• Allometry Assessment Results: Report the proportion of shape variance explained by size, with statistical significance
• Correction Procedures: If size correction was applied, detail the specific method and validation of its effectiveness
• Comparative Results: Present both uncorrected and size-corrected taxonomic comparisons when allometry is substantial
• Visualization: Include deformation graphics or transformation grids showing the allometric vector and its magnitude

Proper assessment and accommodation of allometric effects represents a fundamental methodological imperative in taxonomic studies using geometric morphometrics. The approaches outlined in this application note provide a structured framework for distinguishing genuine taxonomic signals from spurious group differences arising from size variation. By integrating these protocols into taxonomic research workflows, scientists can produce more robust, biologically meaningful classifications that better reflect evolutionary relationships rather than artifacts of size variation. As geometric morphometrics continues to transform taxonomic practice [12], rigorous allometric analysis will remain essential for valid morphological comparisons across disparate taxa.

Allometry, the study of how organismal traits change with size, is a foundational concept in evolutionary biology and taxonomy [14]. In geometric morphometrics (GMM), which quantifies and analyzes shape variation, understanding and correcting for allometric variation is crucial for accurate taxonomic interpretation [3] [2]. When species or populations differ in size, observed shape differences may represent allometric consequences of size variation rather than independent evolutionary events [3]. This application note outlines protocols for distinguishing and analyzing three primary levels of allometric variation—static, ontogenetic, and evolutionary—within the context of taxonomic research using geometric morphometrics. A proper methodological approach allows researchers to test hypotheses about morphological evolution while controlling for confounding allometric effects [4].

Theoretical Foundations and Definitions

Concepts of Allometry

The term allometry, coined by Julian Huxley and Georges Tessier in 1936, originally described relative growth relationships where organ size scales with body size following a power law [14]. This relationship is expressed by the equation log y = α log x + log b, where α is the allometric coefficient indicating whether a trait shows positive (α > 1), negative (α < 1), or isometric (α = 1) scaling [14]. Two primary schools of thought have shaped allometric analysis: the Gould-Mosimann school defines allometry as covariation between size and shape, while the Huxley-Jolicoeur school characterizes it as covariation among morphological features that all contain size information [3] [4]. In geometric morphometrics, this translates to different analytical approaches using either shape space or form space [4].

Levels of Allometric Variation

Biological allometry manifests at three distinct levels, each with different implications for taxonomic research [14] [3]:

• Ontogenetic Allometry: Describes shape change correlated with size increase during growth within a single organism or species [14] [3]. This reflects developmental programs that coordinate trait growth.
• Static Allometry: Captures size-shape covariation among individuals measured at the same developmental stage (typically adults) within a population or species [14] [3]. This represents population-level variation in morphology.
• Evolutionary Allometry: Examines relationships between size and shape across different species or higher taxonomic groups, reflecting divergent evolutionary trajectories [14] [3].

Comparative Framework

Table 1: Characteristics of the Three Levels of Allometric Variation

Characteristic	Ontogenetic Allometry	Static Allometry	Evolutionary Allometry
Definition	Shape change during growth within individuals	Size-shape covariation among conspecifics at similar developmental stages	Size-shape relationships across species or higher taxa
Data Structure	Longitudinal or cross-sectional ontogenetic series	Single population sample at comparable developmental stage	Multiple species means
Biological Interpretation	Developmental programming and growth trajectories	Population-level morphological integration	Macroevolutionary patterns and adaptive divergence
Taxonomic Application	Identifying heterochronic shifts; developmental basis of morphological differences	Understanding intraspecific variation and population structure	Testing hypotheses of adaptive radiation and phylogenetic constraints
Primary Analytical Method	Multivariate regression of shape on size; Principal Component Analysis	Multivariate regression of shape on size; Principal Component Analysis	Regression of species mean shapes on mean sizes

Protocols for Allometric Analysis in Taxonomic Studies

General Workflow for Geometric Morphometric Allometry Analysis

The following diagram illustrates the core decision process and methodological workflow for conducting allometric analyses in taxonomic geometric morphometrics:

Data Collection and Processing Protocol

Protocol 1: Landmark Data Acquisition and Processing

• Objective: To collect and process landmark data suitable for allometric analysis in taxonomic studies.
• Materials: Specimens representing appropriate taxonomic and developmental series; imaging equipment; digitization software (tpsDig, MorphoJ, R)
• Procedure:
  • Sample Design: For ontogenetic allometry, select specimens covering complete developmental series. For static allometry, use adult specimens from a single population. For evolutionary allometry, include multiple species with adequate sample sizes [2].
  • Landmark Digitization: Capture 2D or 3D landmark coordinates using homologous anatomical points. Include sliding semi-landmarks for curves and surfaces where necessary [2].
  • Procrustes Superimposition: Perform Generalized Procrustes Analysis (GPA) to remove non-shape variation (position, orientation, scale) [4]. This aligns specimens by minimizing Procrustes distance among landmark configurations.
  • Size Calculation: Compute Centroid Size as a geometric measure of size. Centroid Size is the square root of the sum of squared distances of all landmarks from their centroid [3].
  • Measurement Error Assessment: Conduct replicate measurements on a subset of specimens to quantify and account for measurement error, which is particularly crucial for detecting subtle allometric patterns [2].

Level-Specific Analytical Protocols

Protocol 2: Analyzing Ontogenetic Allometry

• Objective: To characterize shape trajectories through growth and development.
• Applications: Identifying heterochronic processes in evolution; understanding developmental basis of taxonomic differences.
• Procedure:
  • Data Preparation: Assemble Procrustes-aligned coordinates and Centroid Size values for complete ontogenetic series.
  • Allometric Vector Estimation: Perform multivariate regression of shape coordinates (dependent variables) on Centroid Size (independent variable) [4]. The regression vector represents the ontogenetic allometric trajectory.
  • Visualization: Plot shape changes along the allometric vector using deformation grids or 3D models to illustrate ontogenetic shape transformation.
  • Comparison: For taxonomic applications, compare ontogenetic trajectories between species or populations using methods such as PCA of the allometric vectors or Procrustes ANOVA.

Protocol 3: Analyzing Static Allometry

• Objective: To quantify size-related shape variation within a population of adults.
• Applications: Understanding intraspecific variation; determining whether taxonomic differences reflect allometric scaling.
• Procedure:
  • Data Preparation: Use Procrustes-aligned coordinates and Centroid Size values from adult specimens only.
  • Regression Analysis: Perform multivariate regression of shape on Centroid Size [14] [4]. The proportion of shape variance explained by size (R²) indicates the strength of static allometry.
  • Alternative Approach: For the Huxley-Jolicoeur framework, conduct Principal Component Analysis in conformation space (size-and-shape space) where the first principal component often captures allometric variation [3] [4].
  • Taxonomic Application: Test whether supposed taxonomic boundaries correspond to allometric extremes or deviate from allometric expectations.

Protocol 4: Analyzing Evolutionary Allometry

• Objective: To examine size-shape relationships across species or higher taxa.
• Applications: Testing hypotheses of adaptive radiation; identifying phylogenetic constraints; reconstructing evolutionary patterns.
• Procedure:
  • Data Preparation: Calculate mean shape and mean Centroid Size for each species to avoid pseudoreplication.
  • Regression Analysis: Perform multivariate regression of species mean shapes on mean sizes [14].
  • Phylogenetic Control: Implement phylogenetic generalized least squares (PGLS) to account for non-independence due to shared evolutionary history.
  • Comparison with Other Levels: Contrast evolutionary allometry with static and ontogenetic allometries to infer how evolutionary patterns originate from developmental and population-level processes [15].

Protocol for Allometry Correction in Taxonomy

Protocol 5: Size Correction for Taxonomic Comparisons

• Objective: To remove allometric effects and reveal non-allometric shape differences for taxonomic discrimination.
• Applications: Identifying true taxonomic characters independent of size; improved species delimitation.
• Procedure:
  • Allometric Vector Estimation: Calculate the pooled within-group allometric vector using regression of shape on size across all specimens [3].
  • Residual Computation: Project each specimen orthogonally to the allometric vector to obtain size-corrected shape residuals [3].
  • Validation: Verify that size-corrected residuals show no correlation with Centroid Size.
  • Taxonomic Analysis: Use size-corrected shapes in subsequent discriminant analysis, clustering, or other taxonomic procedures.

Research Reagent Solutions and Essential Materials

Table 2: Essential Materials and Software for Allometric Analysis in Geometric Morphometrics

Category	Item	Specification/Function	Application Context
Data Acquisition	Imaging System	Micro-CT, laser scanner, or digital camera	2D/3D specimen imaging
	Specimen Collection	Representative developmental series and taxonomic samples	All allometry levels
Software Tools	MorphoJ	Integrated morphometrics analysis	General allometric analysis
	R packages (geomorph)	Comprehensive GMM analysis	Advanced and customized analyses
	tps Suite	Digitization and basic shape analysis	Landmark data collection
Analytical Components	Procrustes Algorithm	Removes non-shape variation	Data preprocessing
	Centroid Size	Geometric size measure	Size variable in analyses
	PCA & Regression	Multivariate statistical methods	Allometry vector extraction

Interpretation and Taxonomic Applications

The relationship between different allometric levels provides crucial insights for taxonomic research. A key finding from seminal studies indicates that phenotypic allometry may not accurately guide patterns of evolutionary change [15]. Specifically, Cheverud (1982) demonstrated that patterns of phenotypic, genetic, and environmental allometry can be dissimilar, with only environmental allometries showing ontogenetic allometric patterns [15]. This highlights the importance of not automatically assuming that static or ontogenetic allometries directly predict evolutionary patterns.

In practice, taxonomic decisions should consider the relationship between allometric levels:

• When evolutionary allometry aligns with static allometry within constituent species, differences between taxa may represent simple allometric scaling.
• When evolutionary allometry diverges from within-species patterns, genuine evolutionary shape changes independent of size may define taxonomic boundaries.
• Heterochronic processes (paedomorphosis, peramorphosis) can be identified by comparing ontogenetic trajectories across taxa.

These protocols provide a systematic approach for incorporating allometric analysis into taxonomic studies using geometric morphometrics, enabling more biologically informed interpretations of morphological differences among taxa.

In geometric morphometrics, the study of allometry—the pattern of covariation between the size and shape of an organism—is fundamental to understanding evolutionary and developmental processes [3]. The conceptual approach to quantifying this relationship largely falls into two historically distinct schools of thought: the Gould-Mosimann school, which defines allometry as the covariation of shape with size, and the Huxley-Jolicoeur school, which characterizes allometry as the covariation among morphological features that all contain size information [3] [4]. These philosophical differences have materialized in the implementation of different mathematical spaces for analysis: Procrustes form space and conformation space (also known as size-and-shape space) [3] [4].

For taxonomic studies aimed at correcting for allometric effects, the choice between these frameworks is not merely statistical but biological, influencing how size-related variation is interpreted and handled. This application note details the theoretical foundations, practical implementations, and taxonomic applications of these two spaces, providing researchers with a structured comparison to inform their methodological decisions.

Theoretical Foundations of Morphometric Spaces

Procrustes Form Space (Gould-Mosimann School)

The Gould-Mosimann school conceptually separates size and shape according to the criterion of geometric similarity [3] [4]. In this framework, "form" is defined as the combination of size and shape, and Procrustes form space is constructed by superimposing landmark configurations while optimizing for position and orientation, but not scaling them to a common size [3]. This space retains centroid size as an intrinsic property of each specimen's configuration.

• Allometry Definition: Allometry is explicitly defined as the covariation between shape (residing in Kendall's shape space) and a separate, external measure of size (typically centroid size) [3].
• Analysis Implementation: Allometric analysis is typically implemented via the multivariate regression of shape coordinates on centroid size [3] [4]. The residual shape variation after this regression is considered size-corrected shape.

Conformation Space (Huxley-Jolicoeur School)

The Huxley-Jolicoeur school does not presuppose a separation of size and shape, instead considering morphological "form" as a unified feature [3]. In this framework, conformation space (or size-and-shape space) is constructed by standardizing landmark configurations for position and orientation, but like form space, not for size [3] [4].

• Allometry Definition: Allometry is characterized as the dominant pattern of covariation among traits that all contain size information. The allometric trajectory is identified as the line of best fit through the data points in this space [3].
• Analysis Implementation: The primary allometric vector is typically characterized by the first principal component (PC1) of the data in conformation space [4]. This PC1 represents the major axis of morphological variation, which often correlates strongly with size.

Table 1: Conceptual Comparison of the Two Allometric Frameworks

Feature	Gould-Mosimann School (Procrustes Form Space)	Huxley-Jolicoeur School (Conformation Space)
Core Concept	Separation of size and shape via geometric similarity	Form as a unified entity; no prior size-shape separation
Definition of Allometry	Covariation between shape and size	Covariation among morphological traits containing size information
Primary Analytical Method	Multivariate regression of shape on size	First principal component (PC1) in conformation space
Size Representation	External variable (e.g., centroid size)	Intrinsic property embedded within the form data
Taxonomic Application	Size correction via regression residuals	Projection of data orthogonal to the allometric vector (PC1)

Visualizing the Spatial Relationship

The following diagram illustrates the conceptual relationship between conformation space, shape space, and the allometric vectors within them, as discussed in the theoretical frameworks [3] [4].

Performance Comparison and Methodological Considerations

Statistical Performance under Different Conditions

Computer simulation studies have compared the performance of methods derived from both frameworks under varying conditions of residual variation [4]. The results provide crucial guidance for selecting an appropriate method based on data characteristics.

Table 2: Performance Comparison of Allometric Methods Under Different Variation Types

Method	Underlying Framework	Isotropic Residual Variation	Anisotropic Residual Variation	Deterministic Allometry (No Noise)
Multivariate Regression of Shape on Size	Gould-Mosimann	Good performance	Consistently better performance	Logically consistent with other methods
PC1 of Shape	Gould-Mosimann	Less accurate than regression	Performance degraded	Logically consistent with other methods
PC1 of Conformation Space	Huxley-Jolicoeur	Very close to simulated vector	Very close to simulated vector	Logically consistent with other methods
PC1 of Boas Coordinates	Huxley-Jolicoeur	Very similar to conformation	Very similar to conformation	Logically consistent with other methods

Practical Implications for Taxonomic Studies

The choice between frameworks has direct implications for taxonomic research:

• Data Structure Considerations: When the allometric signal is strong and residual variation is relatively small or isotropic, both frameworks yield similar results [4]. With complex, anisotropic residual variation, the regression-based approach (Gould-Mosimann) generally performs better at recovering the true allometric vector [4].

• Biological Interpretation: The Gould-Mosimann approach is more intuitive when testing explicit hypotheses about size's effect on shape. The Huxley-Jolicoeur approach is advantageous when the researcher wishes to discover the dominant integrated pattern of variation without a priori size-shape separation [3].

• Size Correction Efficacy: For removing allometric effects to discern taxonomic signals, regression-based size correction effectively creates a shape subspace orthogonal to the allometric vector, while conformation-based approaches remove variation along the primary allometric trajectory [3].

Experimental Protocols for Taxonomic Applications

Protocol 1: Allometric Analysis Using Procrustes Form Space

This protocol implements the Gould-Mosimann approach for taxonomic studies where explicit size correction is required.

Step 1: Data Collection and Preparation

• Digitize homologous landmarks across all specimens in the taxonomic study.
• For 2D data, ensure consistent scale and orientation during imaging.
• For 3D data, use a digitizing device or CT/MRI reconstructions.

Step 2: Generalized Procrustes Analysis (GPA) Without Scaling

• Perform GPA to optimize landmark configurations for position and orientation ONLY.
• Do not scale configurations to unit centroid size.
• Retain the centered landmark coordinates (Procrustes form space coordinates) and the centroid size values.

Step 3: Multivariate Regression of Shape on Size

• Compute the multivariate regression of the Procrustes form coordinates on centroid size (or log-transformed centroid size).
• The regression coefficient matrix represents the allometric vector.
• Test the statistical significance of the allometry using a permutation test (typically 1000-10,000 permutations).

Step 4: Size Correction for Taxonomic Comparison

• Calculate the regression residuals, which represent shape variation independent of size.
• Use these size-corrected shapes for subsequent taxonomic analyses (e.g., discriminant analysis, MANOVA, clustering).

Step 5: Visualization

• Visualize the allometric pattern by reconstructing shapes along the regression vector (e.g., from -10% to +10% of centroid size range).
• Plot the regression scores against centroid size to illustrate the allometric relationship.

Protocol 2: Allometric Analysis Using Conformation Space

This protocol implements the Huxley-Jolicoeur approach, suitable for discovering integrated size-shape relationships in taxonomic groups.

Step 1: Data Collection and Preparation

• Follow the same data collection procedures as Protocol 1.

Step 2: Construct Conformation Space

• Standardize landmark configurations for position and orientation using Procrustes superimposition WITHOUT scaling.
• The resulting coordinates represent the conformation space (size-and-shape space).

Step 3: Principal Component Analysis in Conformation Space

• Perform PCA on the variance-covariance matrix of the conformation space coordinates.
• The first principal component (PC1) typically represents the primary allometric vector.

Step 4: Validate Allometric Interpretation

• Correlate PC1 scores with centroid size to confirm its interpretation as an allometric axis.
• If the correlation is strong (|r| > 0.7), PC1 can be confidently interpreted as capturing allometry.

Step 5: Taxonomic Comparisons Independent of Allometry

• For size-free taxonomic comparisons, project the data orthogonal to PC1.
• Alternatively, use subsequent principal components (PC2, PC3, etc.) for taxonomic discrimination, as these represent major axes of shape variation independent of the primary allometric trend.

Step 6: Visualization

• Visualize shape changes along PC1 to interpret the allometric trajectory.
• Plot specimens in the space of PC2 vs. PC3 to examine taxonomic clustering free of primary allometry.

Research Reagent Solutions and Essential Materials

Table 3: Essential Tools for Geometric Morphometric Allometry Studies

Category	Specific Tools/Software	Function in Allometric Analysis
Data Acquisition	3D digitizers (MicroScribe), CT/MRI scanners, high-resolution digital cameras	Capture landmark coordinates from biological specimens
Landmarking Software	tpsDig2, Landmark Editor, IDAV Landmark	Digitize 2D/3D landmark coordinates from images or scans
Morphometric Analysis	MorphoJ, geomorph R package, PAST	Perform Procrustes superimposition, regression analyses, and PCA
Statistical Programming	R (with shapes, vegan packages), MATLAB	Custom analyses, simulation studies, advanced visualization
Visualization	tpsRelw, EVAN Toolkit, MeshLab	Visualize shape changes and allometric trajectories as deformation grids

The theoretical distinction between Procrustes form space and conformation space manifests in practical differences for analyzing and correcting allometry in taxonomic studies. While both frameworks are logically consistent and unlikely to yield fundamentally contradictory results [3], their performance varies under different data conditions.

For most taxonomic applications focused on correcting for allometry, the regression-based approach using Procrustes form space is recommended, particularly when:

• The research question explicitly involves testing size effects on shape
• Sample sizes are sufficient for robust regression estimation (>30 specimens per group)
• Residual variation is complex or anisotropic

The conformation space approach is preferable when:

• The goal is to discover the dominant integrated pattern of form variation
• Studying taxa with strong size-shape integration
• Analyzing complex allometric patterns without a priori size-shape separation

Taxonomists should consider reporting analyses from both frameworks when feasible, as their convergence provides stronger evidence for true biological signals, while divergence may reveal interesting complexities in size-shape relationships within and among taxa.

Practical Protocols for Allometry Detection and Correction in Taxonomic Studies

In taxonomic geometric morphometric studies, accurately capturing and analyzing shape is paramount for understanding evolutionary relationships and patterns. The presence of allometry, the change in shape with size, presents a significant challenge, as it can confound taxonomic signals if not properly addressed [3]. This application note provides detailed protocols for the crucial first step in this process: the acquisition and preparation of morphological data using landmarks, outlines, and semi-landmarks. Proper execution of these foundational techniques ensures that subsequent allometry correction and shape analysis are based on reliable, high-quality data, ultimately leading to more robust taxonomic interpretations. The frameworks for understanding allometry are primarily divided into two schools of thought: the Gould-Mosimann school, which defines allometry as the covariation of shape with size, and the Huxley-Jolicoeur school, which views it as the covariation among morphological features all containing size information [3]. The choice of data acquisition strategy directly influences how these allometric effects can be quantified and removed.

The Scientist's Toolkit: Essential Materials and Software

Table 1: Research Reagent Solutions for Geometric Morphometric Data Acquisition

Item Name	Type	Primary Function	Example Use Case
Microscribe Digitizer	Hardware	Captures 3D coordinates of physical specimens	Precise landmark digitization on skulls [16]
Structured-Light Scanner (e.g., Artec Eva)	Hardware	Creates high-resolution 3D surface meshes	Non-contact scanning of fragile archaeological bones [17]
R Statistical Environment	Software	Core platform for statistical analysis and visualization	Performing Generalized Procrustes Analysis (GPA) and Principal Component Analysis (PCA) [16]
geomorph R package	Software	Comprehensive toolbox for geometric morphometrics	Implementing Procrustes alignment and allometry analysis [16]
Viewbox 4 Software	Software	Digitizes landmarks, curves, and surfaces on 3D models	Applying a standardized template of coordinate points to os coxae scans [17]
Coordinate Point Template	Data/Protocol	Defines homologous points and curves for a specific structure	Ensuring consistent and comparable data capture across multiple specimens and researchers [17]

Data Acquisition Protocols

Landmark and Semi-Landmark Digitization

The precise capture of morphological data is the foundation of any geometric morphometric study. The following protocol outlines the steps for digitizing a biological structure, such as a skull or os coxae (hip bone), using a combination of landmark types.

Materials:

• Specimens or their high-resolution 3D scans (e.g., from a structured-light scanner) [17]
• Digitizing hardware (e.g., Microscribe digitizer) or software (e.g., Viewbox 4) [16] [17]
• Pre-defined landmark template [17]

Procedure:

• Template Design: Establish a template that defines the number and location of all points. This includes:
  • Fixed Landmarks: Anatomically homologous points that can be precisely located across all specimens (e.g., suture intersections, apex of a process) [17].
  • Curve Semi-Landmarks: Points placed along homologous curves to capture their geometry. A high initial density is recommended [17].
  • Surface Semi-Landmarks: Points distributed across homologous surfaces to capture overall form. The protocol from [17] started with 425 surface points for an os coxae.
• Data Capture: Apply the template consistently to every specimen in the dataset. For physical specimens, use a digitizer like the Microscribe MX [16]. For 3D scans, digitize directly on the mesh using software like Viewbox 4 [17].
• Data Export: Save the resulting coordinate configuration for each specimen as a (k × m) matrix, where (k) is the number of points and (m=3) for the x, y, and z coordinates. Ensure point order is consistent across all specimens [17].

Determining Optimal Coordinate Density

Oversampling or undersampling a structure can reduce statistical power and analytical sensitivity. The following experimental protocol, adapted from [17], determines the minimal number of points needed to faithfully capture shape variation.

Experimental Workflow:

• Initial Oversampling: Begin by designing a preliminary template that substantially oversamples the structure. For example, the os coxae protocol used 609 total points [17].
• Subsample and Analyze: Apply this dense template to a representative subset of specimens (e.g., 5 individuals). Use an algorithm like Watanabe's Landmark Sampling to progressively reduce the number of points while monitoring the impact on the statistical assessment of shape, such as the ability to detect structural modularity [17].
• Identify Optimal Density: The optimal coordinate density is the point at which further reduction of points begins to significantly degrade or alter the statistical signal of interest. This ensures efficiency without loss of essential morphological information [17].

The following workflow diagram illustrates the key stages of data acquisition and preparation, from initial specimen handling to the final data ready for allometry analysis.

Data Preparation and Validation Protocols

Handling Missing Data and Damage

Specimen damage is a common issue in taxonomic and archaeological studies. Removing incomplete specimens sacrifices statistical power, so imputation is often preferable.

Materials:

• Dataset with missing landmarks in some specimens.
• Statistical software (R environment).

Procedure:

• Assess Level of Missingness: Determine the percentage of missing coordinate points across the dataset and their distribution.
• Choose Imputation Method: The optimal method depends on the extent of damage.
  • For minimal damage, parametric statistical methods like Partial Least Squares regression can be used. Note that these methods require a sufficiently large sample size relative to the dimensionality of the data and the number of missing points [17].
  • For larger areas of missing data, non-parametric "geometric" methods that interpolate based on the intact geometry of a reference specimen or sample may be more effective [17].
• Impute and Validate: Perform the imputation and, where possible, validate the accuracy of the reconstructed shapes.

Estimating Measurement Error

Before proceeding to allometry analysis, it is critical to quantify the precision of the digitization process to ensure observed variation is biological and not methodological.

Materials:

• A subset of specimens (e.g., one skull and one lower jaw).
• Digitizing equipment and template.

Procedure:

• Repeated Measures: Digitize the same subset of specimens multiple times (e.g., five times) at the beginning of the data collection process [16].
• Statistical Comparison: Perform a Procrustes ANOVA or similar analysis to compare the variance introduced by repeated digitization against the biological shape variance within the entire dataset.
• Acceptance Criterion: The measurement error should be small compared to the biological variation present in the dataset. High measurement error necessitates a review of the digitization protocol and template [16].

Table 2: Quantitative Data from Exemplar Geometric Morphometric Studies

Study Focus	Specimen Type	Sample Size	Landmark Strategy	Key Metric	Value / Outcome
Carnivore Skull Analysis [16]	316 adult skulls	86 breeds / taxa	53 landmarks on skull	Procrustes distance between mean shapes	Used to quantify morphological difference between breeds
Os Coxae Protocol Development [17]	29 archaeologically-recovered bones	2 collections	25 fixed landmarks, 159 curve, 425 surface semi-landmarks	Optimal point density	Determined via landmark sampling to avoid over/under-sampling
Tooth Mark Identification [18]	Experimentally-derived bone surface modifications	4 carnivore types	Outline analysis vs. semi-landmarks	Classification accuracy	Geometric Morphometrics: <40%; Computer Vision: ~80%

Pathway to Allometry Correction in Taxonomic Studies

The ultimate goal of meticulous data acquisition is to enable rigorous statistical analysis, with correcting for allometry being a central task in taxonomic studies. The prepared data undergoes Generalized Procrustes Analysis (GPA) to remove differences due to position, orientation, and size, projecting specimens into a linearized shape space [17]. Once aligned, the two main conceptual frameworks for allometry can be applied, each leading to a different size-correction technique, as illustrated below.

Implementation of Allometry Correction:

• Following the Gould-Mosimann Framework: Perform a multivariate regression of the Procrustes-aligned shape coordinates on a measure of size, typically centroid size (the square root of the sum of squared distances of all landmarks from the centroid) [16] [3]. The residuals from this regression represent size-corrected shape and can be used in subsequent taxonomic analyses like Principal Component Analysis (PCA) [16].
• Following the Huxley-Jolicoeur Framework: Analyze the data in Procrustes form space (which retains size information) or conformation space. The first principal component (PC1) of this data often represents the primary allometric trajectory. Size correction can be achieved by projecting data orthogonally to this axis or by analyzing subsequent principal components [3].

In taxonomic geometric morphometric studies, the accurate characterization of an organism's form is fundamental for distinguishing between species, understanding evolutionary relationships, and identifying evolutionary significant units. However, a fundamental challenge lies in the fact that the raw coordinates of morphological landmarks capture a composite of an organism's true shape, its size, and its orientation in space [19]. Procrustes Superimposition addresses this challenge by providing a robust statistical method for removing the effects of position, scale, and rotation from landmark data, thereby isolating pure shape information for subsequent comparison [19]. This separation is a critical prerequisite for the study of allometry—the pattern of covariation between shape and size—which, if unaccounted for, can confound taxonomic interpretations by mimicking or obscuring true phylogenetic signal [3] [13]. This application note details the protocols for performing Procrustes superimposition and framing it within essential allometric analyses, providing a structured workflow for taxonomic researchers.

Theoretical Foundations: Shape, Size, and Allometry

The Concept of Shape in Morphometrics

In geometric morphometrics, shape is formally defined as all the geometric information that remains when location, scale, and rotational effects are filtered out from an object [19]. The goal of Procrustes superimposition is to standardize specimens based on this definition, allowing for the direct comparison of their shapes.

The Role of Centroid Size

A key component of the Procrustes methodology is the calculation of Centroid Size, a measure of size that is statistically independent of shape under certain models of variation [3]. Centroid Size is calculated as the square root of the sum of squared distances of all landmarks from their centroid (center of gravity). It serves as the standard size metric in most geometric morphometric studies and is central to allometric analyses.

Schools of Allometric Thought

The approach to analyzing allometry depends on the conceptual framework, which can be broadly divided into two schools [3] [4]:

• The Gould-Mosimann School: This framework strictly separates size and shape. Allometry is defined as the covariation of shape with an external measure of size, typically analyzed through the multivariate regression of shape variables on Centroid Size.
• The Huxley-Jolicoeur School: This framework does not pre-separate size and shape but considers "form" (size-and-shape) as a single entity. Allometry is characterized as the primary axis of covariation among morphological traits, typically identified by the first principal component (PC1) in a form space, known as conformation space.

The following diagram illustrates the logical relationship between these concepts and their associated analytical spaces.

Protocols for Procrustes Superimposition and Allometry Analysis

Protocol 1: Generalized Procrustes Analysis (GPA)

This protocol standardizes a set of landmark configurations, producing shape variables for subsequent analysis [19].

• Objective: To remove differences in position, scale, and orientation from landmark data, creating a set of Procrustes shape coordinates.
• Materials: A 3D array (p x k x n) of landmark coordinates, where p is the number of landmarks, k is the dimensionality (2 or 3), and n is the number of specimens.
• Software: The gpagen function in the geomorph R package is used for this protocol.
• Procedure:
  • Translation: Translate each specimen so that its centroid (the mean of its landmark coordinates) is at the origin (0,0) of the coordinate system.
  • Scaling: Scale each specimen to unit Centroid Size.
  • Rotation: Rotate each specimen to minimize the total sum of squared distances between its landmarks and the corresponding landmarks of a consensus (mean) configuration. This is an iterative process that refines the consensus as specimens are aligned.
• Output:
  • coords: A (p x k x n) array of Procrustes shape coordinates.
  • Csize: A vector of Centroid Size for each specimen.
  • consensus: The Procrustes consensus (mean) configuration.

Protocol 2: Analyzing Allometry within the Gould-Mosimann Framework

This protocol tests for and characterizes the relationship between shape and size using multivariate regression [3] [4].

• Objective: To quantify static, ontogenetic, or evolutionary allometry by regressing shape on size.
• Prerequisites: Completion of Protocol 1 to obtain Procrustes shape coordinates and Centroid Size.
• Software: Standard functions in morphometric R packages (e.g., procD.lm in geomorph).
• Procedure:
  • Model Fitting: Perform a multivariate regression of the Procrustes shape coordinates (dependent variable) on log-transformed Centroid Size (independent variable). The model is: Shape ~ log(Centroid Size).
  • Significance Testing: Test the statistical significance of the regression using a permutation-based procedure (e.g., 1000 permutations) to obtain a p-value.
  • Visualization: Visualize the allometric trend as a deformation of the consensus configuration in the positive and negative directions along the regression vector.

Protocol 3: Analyzing Allometry within the Huxley-Jolicoeur Framework

This protocol identifies the major axis of form variation, which often corresponds to the allometric trajectory [3] [4].

• Objective: To identify the primary axis of covariation in form (size-and-shape) space.
• Prerequisites: Raw landmark coordinates or coordinates aligned in conformation space (GPA without scaling).
• Software: Principal Component Analysis (PCA) performed on coordinates from conformation space.
• Procedure:
  • Create Conformation Space: Perform a Procrustes superimposition that removes differences in location and orientation, but not size.
  • Principal Component Analysis: Perform a PCA on the coordinates from the conformation space.
  • Interpretation: The first principal component (PC1) often represents the allometric vector. The correlation between PC1 scores and Centroid Size should be checked to confirm this.
  • Visualization: Visualize the shape changes associated with the minimum and maximum scores along PC1.

The following workflow diagram integrates these protocols into a coherent research pipeline for taxonomic studies.

Quantitative Data and Method Comparison

The choice of method for studying allometry can impact results. The table below summarizes the core features of the two main approaches, while a performance comparison based on simulation studies highlights their statistical properties.

Table 1: Comparison of Allometric Frameworks in Geometric Morphometrics

Feature	Gould-Mosimann Framework	Huxley-Jolicoeur Framework
Core Definition	Covariation between shape and size	Covariation among morphological traits containing size information
Analytical Space	Shape tangent space	Conformation (size-and-shape) space
Primary Method	Multivariate regression of shape on size	First principal component (PC1)
Size Variable	External (e.g., Centroid Size)	Intrinsic to the analysis
Logical Basis	Separation of size and shape via geometric similarity	Line of best fit to form data

Table 2: Performance Comparison of Allometry Methods Based on Simulation Studies [4]

Method	Accuracy with Isotropic Noise	Accuracy with Anisotropic Noise	Logical Consistency (No Noise)
Regression of Shape on Size	High performance	High performance	Logically consistent
PC1 of Shape	Lower performance	Lower performance	Logically consistent
PC1 of Conformation/Boas Coordinates	Very high performance	Very high performance	Logically consistent

The Scientist's Toolkit: Essential Reagents and Software

A standardized set of tools is required to execute the protocols outlined in this document.

Table 3: Research Reagent Solutions for Geometric Morphometrics

Item	Function/Brief Explanation
Landmark Data	2D or 3D coordinates of biologically homologous points. The fundamental raw data for analysis.
R Statistical Software	Open-source environment for statistical computing and graphics. The primary platform for morphometric analysis.
geomorph R Package	A comprehensive package for performing geometric morphometric analyses, including GPA (gpagen), statistical tests, and visualization [19].
gpagen Function	The core function for performing Generalized Procrustes Analysis on landmark data, handling both fixed landmarks and sliding semilandmarks [19].
Momocs R Package	Another R package useful for outline and morphological analysis, providing an alternative toolkit for shape analysis.
StereoMorph Software	An R package for digitizing landmarks and curves from images, facilitating data acquisition.
TpsDig Software	A standalone Windows program for digitizing landmarks from image files.

Procrustes superimposition is the foundational step that enables the rigorous quantification and comparison of biological shape in taxonomic research. Isolating shape from size and orientation is not an end in itself but a critical prerequisite for unbiased investigation of allometry, which is a pervasive source of morphological variation. By applying the detailed protocols for Procrustes alignment and subsequent allometry analysis—either through multivariate regression on size or via the primary axis of form variation—researchers can effectively dissect the complex interplay between size and shape. This process is indispensable for making robust taxonomic decisions, identifying true phylogenetic signals distinct from allometric covariation, and advancing our understanding of evolutionary patterns and processes.

In geometric morphometrics, allometry—the study of the relationship between size and shape—remains an essential concept for understanding evolution and development [3]. For taxonomic studies, accurately assessing and correcting for allometry is crucial to isolate shape variation that is independent of size, thereby ensuring that taxonomic comparisons are not confounded by allometric scaling. The approach of using multivariate regression of shape on centroid size falls within the Gould-Mosimann school of allometry, which defines allometry specifically as the covariation of shape with size [3]. This method provides a powerful and direct way to quantify and test allometric relationships, making it a cornerstone technique for taxonomic research.

Theoretical Framework: Two Schools of Allometric Thought

Understanding the conceptual underpinnings is vital for choosing the correct analytical approach. The two primary schools of thought provide different, yet complementary, perspectives.

• Gould-Mosimann School: This framework strictly separates the concepts of size and shape. Within this school, allometry is explicitly defined as the covariation between shape (the geometric information remaining after removing location, scale, and rotation effects) and size (a scalar measure like centroid size) [3]. Analyzing allometry via multivariate regression of shape variables on centroid size is the direct implementation of this concept. This is often the most intuitive approach for taxonomic studies aiming to answer: "How much of the observed shape difference can be explained by size variation alone?"

• Huxley-Jolicoeur School: This school does not pre-separate size and shape but considers them together as "form." Allometry is characterized as the covariation among multiple morphological features that all contain size information [3]. In geometric morphometrics, this is often implemented by performing a Principal Component Analysis (PCA) in Procrustes form space (which retains size information) and interpreting the first principal component as the primary allometric trajectory [3]. While useful for describing multivariate growth, it is less direct for testing a specific size-shape relationship.

For the purpose of isolating allometric effects for taxonomic correction, the regression-based approach of the Gould-Mosimann school is typically the most appropriate and interpretable method.

Experimental Protocol: Multivariate Regression of Shape on Centroid Size

This protocol provides a step-by-step guide for assessing allometry using multivariate regression, from data collection to interpretation, specifically framed for taxonomic studies.

Data Acquisition and Landmarking

Function: To capture the geometric configuration of specimens for subsequent shape analysis.

• Digital Image Acquisition: Capture high-resolution, standardized 2D or 3D images of all specimens. Ensure consistent orientation, scale, and lighting. For 2D studies, enface photographs are standard [20].
• Landmark and Semilandmark Digitization: Place landmarks: anatomical points that are biologically homologous across all specimens (e.g., the tip of a rostrum, the base of a spine). These are classified by Bookstein's typology [20]. To capture outlines and curves, place semilandmarks. These are points that lack strict homology but are necessary to quantify the geometry of contours and surfaces [20].
• Automated Landmarking (Optional but Recommended): For large datasets or to improve reproducibility, consider using automated tools like FaceDig or points placed by tools in the TPS series [20]. These AI-powered tools can place landmarks with human-level precision and output files compatible with standard software like TpsDig2, promoting consistency and saving time [20]. Always visually inspect automated landmark placements for potential errors.

Procrustes Superimposition and Size Extraction

Function: To remove non-shape differences (position, orientation, scale) and extract a measure of isometric size.

• Generalized Procrustes Analysis (GPA): Perform GPA on the raw landmark coordinates. This procedure optimally translates, rotates, and scales all specimens to a common unit centroid size, minimizing the Procrustes distance among configurations. The resulting coordinates reside in a curved, non-Euclidean shape space [3] [21].
• Shape Variables for Analysis: The Procrustes-aligned coordinates are the shape variables used as the dependent variable in the regression. For statistical analysis that requires linear space, they are typically projected into a linear tangent space, which is a close approximation of shape space near the consensus [21].
• Centroid Size Calculation: Centroid size (CS) is computed as the square root of the sum of squared distances of all landmarks of a specimen from their centroid [3]. It is the standardized, isometric size measure used as the independent variable in the allometric regression.

Multivariate Regression and Statistical Testing

Function: To quantify the relationship between shape (dependent variable) and centroid size (independent variable).

• Model Execution: Perform a multivariate multiple regression in morphometric software (e.g., MorphoJ) where the matrix of Procrustes shape coordinates is regressed onto the vector of centroid sizes [22].
• Statistical Significance Testing: Test the null hypothesis that there is no relationship between shape and size. This is typically done using a Goodall's F-test or a permutation test (e.g., 10,000 permutations) against the residual randomization. A significant p-value (e.g., p < 0.05) indicates the presence of a statistically significant allometric signal [3].
• Effect Strength Quantification: Calculate the proportion of total shape variance explained by size. This is provided by the multivariate regression R² value. An R² of 0.15, for example, means that 15% of the total shape variation in the sample is attributable to allometry.

Visualization and Interpretation

Function: To understand the biological meaning of the allometric relationship.

• Visualizing the Allometric Vector: The regression model yields an allometric vector in shape space. This vector is visualized as a shape change using:
  • Deformation Grids: Thin-plate spline (TPS) deformation grids show the shape transformation predicted by the regression model from the smallest to the largest centroid size in the dataset [21].
  • Vector Diagrams: Diagrams showing the direction and magnitude of landmark displacement along the allometric trajectory.
• Taxonomic Context: In a mixed-species sample, a significant allometric relationship could indicate that observed taxonomic differences are partly driven by size differences (evolutionary allometry). Correcting for this allometry (see Section 5) helps reveal non-allometric shape differences that may have greater taxonomic significance.

The following workflow diagram summarizes the core analytical pipeline described in this protocol:

The Scientist's Toolkit: Essential Reagents & Software

Table 1: Key Software and Analytical Tools for Allometric Assessment.

Tool Name	Function/Brief Explanation	Relevance to Allometry Protocol
TPS Software Series (e.g., TpsDig2) [21] [20]	Free, widely-used software for digitizing landmarks and semilandmarks on 2D images.	Used in the initial Data Acquisition phase to collect raw coordinate data.
MorphoJ [22]	An integrated, user-friendly software package for geometric morphometric analysis.	Executes Procrustes superimposition, multivariate regression, and statistical testing. Essential for the core analysis.
FaceDig [20]	An open-source, AI-powered tool for automated landmark placement on 2D facial photographs.	Standardizes and accelerates the landmarking process, reducing human error and time investment in data acquisition.
R package 'geomorph'	A powerful R-based package for comprehensive morphometric analysis.	Provides a flexible, script-based environment to perform all steps, including GPA, regression, and permutation tests.
Thin-Plate Spline (TPS) [21]	A geometric metaphor and algorithm for visualizing shape change as a smooth deformation.	The primary method for visualizing the allometric vector from the regression as a biological shape transformation.

Application: Correcting for Allometry in Taxonomic Studies

Once allometry is quantified, its effects can be removed to analyze size-free shape variation, which is critical for taxonomy.

Protocol for Allometry Correction

Function: To compute shape residuals that are independent of size, allowing for fair taxonomic comparisons.

• Obtain Regression Residuals: After running the multivariate regression, extract the regression residuals. These residuals represent the portion of each specimen's shape that is not predictable by its size—the size-corrected shapes.
• Analyze Corrected Shapes: Use the residuals as the new shape data for subsequent taxonomic analyses, such as Canonical Variate Analysis (CVA), Linear Discriminant Analysis (LDA), or between-group PCA [22]. These analyses will now reveal shape differences between taxa that are independent of allometric scaling.
• Interpretation with Caution: Compare the results of analyses performed on original shapes versus size-corrected shapes. If taxonomic separation decreases after correction, it indicates that allometry was a major component of the perceived taxonomic difference. Remaining differences are more likely due to other evolutionary or developmental factors.

The following diagram illustrates the logical decision-making process for interpreting allometry within a taxonomic framework:

Table 2: Key Quantitative Outputs from the Multivariate Regression of Shape on Centroid Size.

Output Metric	Description & Interpretation	Relevance to Taxonomy
Centroid Size (CS)	A measure of isometric size for each specimen. Used as the predictor variable.	Allows for the examination of size overlap or disparity between putative taxonomic groups.
Procrustes Distance	The geometric difference in shape between specimens after superimposition.	The raw shape variation that the analysis seeks to partition into allometric and non-allometric components.
Regression R²	The proportion of total shape variance explained by size.	A key metric: a high R² indicates allometry is a strong force, and correction is critical for unbiased taxonomy.
p-value	The statistical significance of the shape-size relationship (from permutation test).	A non-significant result suggests that correcting for allometry may be unnecessary for the dataset.
Allometric Vector	The multivariate direction of shape change associated with increasing size.	Describes the specific morphological transformation (e.g., relative elongation, widening) linked to size increase.

Allometry, defined as the size-related changes in morphological traits, represents a fundamental challenge in taxonomic geometric morphometric studies. When characterizing species differences based on shape, the confounding effects of size variation can obscure true taxonomic signals if not properly addressed [3]. The need for allometry correction arises from the biological reality that organisms change shape predictably as they grow, and different species may follow distinct allometric trajectories. In taxonomic research, this is particularly crucial when comparing specimens across developmental stages or when size differences reflect ecological rather than taxonomic variation [23].

The theoretical foundation for allometry correction rests on two historical schools of thought: the Gould-Mosimann school defines allometry as the covariation of shape with size, typically implemented through multivariate regression of shape variables on size measures. In contrast, the Huxley-Jolicoeur school characterizes allometry as the covariation among morphological features that all contain size information, implemented by analyzing allometric trajectories along the first principal component in morphospace [3]. Understanding this distinction is essential for selecting appropriate correction methods in taxonomic research, as each approach carries different implications for how size and shape relationships are conceptualized and analyzed.

Theoretical Foundations of Allometry Correction

Concepts and Definitions

The statistical foundation of allometry correction rests on several key concepts and definitions that form the vocabulary of geometric morphometric analysis:

• Form: The complete geometric configuration of morphological structures, encompassing both size and shape components [3]
• Shape: The geometric properties of a morphological structure that remain after accounting for differences in position, scale, and rotation [3]
• Size: Typically quantified as centroid size, calculated as the square root of the sum of squared distances between landmarks and their centroid [24]
• Allometric trajectory: The characteristic path of shape change associated with size increase, which may be shared across taxa or unique to specific groups [3]

In taxonomic contexts, researchers must distinguish between different levels of allometry: ontogenetic allometry (shape change through growth), static allometry (shape-size covariation within a single developmental stage), and evolutionary allometry (divergence in allometric patterns across taxa) [3]. Each level requires consideration when designing taxonomic studies, as confounding these levels can lead to misinterpretation of taxonomic signals.

Statistical Frameworks for Allometry

Two primary statistical frameworks implement the conceptual schools of allometric thought:

Table 1: Comparison of Allometric Frameworks

Aspect	Gould-Mosimann Framework	Huxley-Jolicoeur Framework
Core concept	Allometry as shape-size covariation	Allometry as covariation among size-informative traits
Implementation	Multivariate regression of shape on size	Principal component analysis in form space
Size-shape relationship	Explicit separation of size and shape	Unified treatment of morphological form
Typical application	Size correction via regression residuals	Characterization of allometric trajectories
Taxonomic utility	Removing size effects for shape comparison	Understanding evolutionary allometry patterns

The Gould-Mosimann approach is implemented through Procrustes-based geometric morphometrics, where shape variables (Procrustes coordinates) are regressed on centroid size, and the residuals form the size-corrected shape data [3]. The Huxley-Jolicoeur approach operates in Procrustes form space or conformation space, where the first principal component often captures the allometric trajectory [3]. For most taxonomic applications focused on discriminating species based on shape characteristics, the Gould-Mosimann framework provides more straightforward implementation and interpretation.

Methodological Protocols for Allometry Correction

Data Acquisition and Preparation

The initial phase of allometry correction requires careful data collection and processing to ensure meaningful results:

• Landmark digitization: Place homologous landmarks consistently across all specimens using standardized protocols. For 2D analyses, this typically involves 15-20 landmarks capturing functionally and taxonomically relevant structures [24]

• Procrustes superimposition: Normalize landmark configurations through Generalized Procrustes Analysis (GPA) to remove differences in position, orientation, and scale using the geomorph package in R [24]

• Size calculation: Compute centroid size for each specimen as the square root of the sum of squared distances from each landmark to the centroid of the configuration [24]

• Data screening: Examine distributions of centroid size and Procrustes distances to identify outliers that might indicate data quality issues

Consistent data collection is particularly important in taxonomic studies where subtle shape differences may characterize species boundaries. All specimens should be complete, fully articulated, and preserved consistently to minimize non-biological sources of variation [24].

Testing for Allometric Effects

Before applying correction methods, researchers must quantitatively establish the presence and nature of allometric patterns in their dataset:

This initial test determines whether a significant relationship exists between shape and size across the entire dataset [25]. A significant result (typically p < 0.05) indicates that allometry represents a substantial source of shape variation that may require correction for taxonomic comparisons.

To test whether allometric patterns are consistent across groups (e.g., species or populations):

A non-significant result in this model comparison suggests that groups share a common allometric trajectory, making size correction appropriate. A significant result indicates divergent allometries, complicating size correction approaches [25] [23].

Burnaby's Approach and Regression Residual Methods

Two primary methodological approaches exist for correcting allometric effects in morphometric data:

Burnaby's approach projects data into a space orthogonal to the size-related vector of variation. Originally developed for traditional morphometrics, it has been adapted for geometric morphometrics by using the first principal component as the size-related vector when curvature is the dominant source of variation [24]. This approach is particularly effective when a single vector captures the majority of size-related shape change.

The regression residual method (more commonly used in contemporary geometric morphometrics) involves calculating residuals from a multivariate regression of shape variables on size:

This approach removes the component of shape variation predictable from size while retaining the mean shape configuration, making it suitable for subsequent taxonomic analyses [23]. The resulting residuals represent size-corrected shape variables that can be used in Procrustes ANOVA, discriminant analysis, or other multivariate procedures to test taxonomic hypotheses.

The following workflow diagram illustrates the key decision points in selecting and applying allometry correction methods:

Research Reagent Solutions and Materials

Table 2: Essential Research Tools for Allometry Correction Studies

Tool/Resource	Function	Application Context
TPSDig2	Landmark digitization	Collecting 2D coordinate data from specimen images
R statistical environment	Data analysis platform	Performing statistical tests and corrections
geomorph package	Geometric morphometrics implementation	Procrustes analysis, allometry tests, and visualization
Morpho package	Supplemental morphometric analyses	Additional shape analysis procedures
PCA (Principal Component Analysis)	Dimension reduction	Visualizing allometric trajectories and shape variation
Procrustes ANOVA	Hypothesis testing	Evaluating group differences in shape
Centroid size	Size metric	Quantifying biological size independent of shape

These tools collectively provide researchers with a comprehensive toolkit for implementing allometry correction protocols in taxonomic studies. The R ecosystem, specifically the geomorph and Morpho packages, offers specialized functions for each step of the allometry correction pipeline, from initial data input through final visualization of corrected shapes [24] [25].

Case Studies and Applications

Fossil Fish Body Curvature Correction

A compelling application of allometry correction methods comes from paleontological studies of fossil fishes, where postmortem body curvature introduces substantial error into morphometric data. Researchers working with exceptionally preserved gonorynchiform fossils from the Las Hoyas deposits (Early Cretaceous, Spain) tested two correction approaches on the species Rubiesichthys gregalis and Gordichthys conquensis [24].

The study employed an Index of Curvature (IC) calculated as the ratio between the curved length along the vertebral column and the straight-line distance between terminal points. Researchers compared a regression-unbending method (multivariate regression of Procrustes data against IC) with a TPS unbending function (mathematical straightening of specimens based on landmark configurations) [24]. The regression approach successfully removed curvature effects while preserving biologically meaningful shape variation, demonstrating the utility of allometry correction methods even in fossil specimens where additional taphonomic effects complicate morphological analyses.

Grasshopper Lineage Comparisons

In a study of pronotum shape variation between genetic lineages of grasshoppers, researchers faced the question of whether to correct for allometric effects before assessing phenotypic differentiation [25]. Initial analysis revealed a weak but significant allometric effect (R² = 0.041, p = 0.0004), prompting further investigation into whether lineages shared common allometric patterns.

The researchers tested models with common versus unique allometries and found no significant interaction between size and lineage group (p = 0.292), indicating parallel allometric trajectories [25]. This justified the application of a common size correction using regression residuals, enabling direct comparison of shape differences between lineages independent of size variation. This case illustrates the importance of testing allometric homogeneity assumptions before applying corrections in taxonomic studies.

Lake Fish Population Differentiation

A study of fish populations across different lakes encountered complex allometry correction challenges due to heterogeneous allometric slopes between populations [23]. Some lakes contained only recently introduced juveniles, creating a situation where size variation reflected different age distributions rather than taxonomic differences.

When researchers initially applied a standard regression residual correction assuming common allometry, results showed significant shape differences between lakes. However, when they tested a model with heterogeneous slopes (size × lake interaction), they found significant differences in allometric trajectories [23]. This indicated that shape differences varied with size—what differentiated lakes at small sizes did not necessarily hold at larger sizes—making simple allometry correction inappropriate. Instead, the researchers focused on comparing allometric patterns themselves rather than attempting to remove size effects, highlighting the importance of diagnostic testing before correction.

Implementation Considerations and Best Practices

Diagnostic Procedures and Validation

Effective implementation of allometry correction requires thorough diagnostic procedures to validate methodological choices:

• Visualize allometric trajectories: Plot principal components against size to identify group-specific patterns
• Check model assumptions: Examine residuals from allometric regressions for homogeneity of variance
• Validate correction effectiveness: Confirm that corrected shapes show no residual correlation with size
• Assess biological interpretability: Ensure that corrected shapes maintain anatomical feasibility

Researchers should document the variance explained by allometric factors before correction and verify that correction procedures do not remove non-allometric shape variation of taxonomic interest. In practice, reporting both corrected and uncorrected results can provide a more comprehensive understanding of morphological patterns.

Limitations and Alternative Approaches

Allometry correction methods carry important limitations that researchers must consider:

• Common allometry assumption: Correction methods assume groups share allometric patterns, which may not hold for distantly related taxa [23]
• Information loss: Removing size-related shape variation may eliminate biologically meaningful taxonomic signals
• Size range effects: Corrections are most reliable within the observed size range, with limited extrapolation capability
• Interaction with other factors: Allometric patterns may vary with sex, environment, or other covariates

When heterogeneous allometries preclude standard correction approaches, alternative strategies include:

• Analyzing allometric trajectories as taxonomic characters themselves
• Comparing shapes at equivalent sizes using size-matched subsamples
• Utilizing ontogenetic sequences rather than static adult comparisons
• Implementing more complex models that explicitly parameterize group-specific allometries

The choice between approaches should be guided by research questions, sample characteristics, and diagnostic results rather than automatic application of standard protocols.

Allometry correction represents an essential methodological component in taxonomic geometric morphometric studies, where disentangling size-related shape variation from taxonomic signals is crucial for robust species discrimination and characterization. The regression residual method, rooted in the Gould-Mosimann school of allometry, provides the most widely applicable approach for size correction when diagnostic tests validate the assumption of common allometric trajectories across groups.

Successful implementation requires careful attention to data quality, thorough diagnostic testing of allometric patterns, and appropriate interpretation of corrected shapes in light of biological context. As the case studies illustrate, there is no one-size-fits-all solution to allometry correction—researchers must select and validate methods based on their specific taxonomic questions and dataset characteristics. By following the protocols and considerations outlined here, researchers can enhance the validity and interpretability of their taxonomic conclusions based on geometric morphometric data.

In geometric morphometrics, allometry—the relationship between shape and size—is a crucial factor to account for in taxonomic research. Failure to correct for allometric effects can confound taxonomic interpretations, as shape differences due to growth or size variation may be misinterpreted as phylogenetic signals. Two principal schools of thought guide allometric studies: the Gould-Mosimann school, which defines allometry as the covariation between shape and size, and the Huxley-Jolicoeur school, which characterizes allometry as covariation among morphological traits that all contain size information [3] [4]. This protocol provides detailed methodologies for implementing allometry correction in taxonomic studies using two prominent software packages: MorphoJ and geomorph.

Table 1: Key Concepts in Allometry Correction

Concept	Definition	Taxonomic Relevance
Allometry	The relationship between organismal size and shape	Can confound taxonomic discrimination if unaccounted for
Size Correction	Statistical removal of size-related shape variation	Isolates taxonomic signal from allometric effects
Static Allometry	Allometric patterns within a single developmental stage	Primary focus for taxonomic studies of adult specimens
Ontogenetic Allometry	Shape change trajectories throughout growth	Important for taxonomic studies including immature specimens
Geometric Morphometrics	Quantitative analysis of form based on Cartesian landmark coordinates	Provides powerful tools for quantifying and comparing shapes

MorphoJ

MorphoJ is an integrated program package for geometric morphometric analysis of both 2D and 3D landmark data [26]. The software is written in pure Java and is freely available for use in education and research [26].

Installation Procedure:

• Download the appropriate self-contained package for your operating system (Windows, Mac OS from 10.11 "El Capitan" onward, or Ubuntu Linux) from the official MorphoJ page [22]
• For Mac OS users: After installing from the .dmg file, navigate to Applications, right-click "MorphoJ.app," select "Open," choose "Cancel" in the first dialog box, then repeat the right-click and "Open" procedure, this time selecting "Open" [22]
• The program requires no external Java runtime environment, as it uses the Eclipse Temurin OpenJDK [26]

geomorph

geomorph is an R package that provides a comprehensive toolkit for performing all stages of geometric morphometric shape analysis within the R statistical computing environment [27]. It supports the analysis of landmark data from points, curves, and surfaces.

Installation Procedure:

• Install R (version 4.4 or higher) from the Comprehensive R Archive Network (CRAN)
• Launch R and execute: install.packages("geomorph")
• Load the package for use: library(geomorph)

Table 2: Software Comparison for Allometry Correction

Feature	MorphoJ	geomorph
User Interface	Graphical user interface (GUI)	Command-line in R
Data Dimensionality	2D and 3D landmark data	2D and 3D landmark data
Allometry Methods	Regression-based (Gould-Mosimann)	Multiple frameworks
Size Correction	Multivariate regression of shape on size	Regression and projection methods
Statistical Framework	Integrated methods	Customizable analyses
Visualization	Built-in graphics	R-based plotting
Citation	Klingenberg, 2011 [26]	Adams et al. [27]

Theoretical Framework for Allometry Correction

Two principal methodological frameworks exist for analyzing allometry in geometric morphometrics, each with distinct implications for taxonomic studies:

Gould-Mosimann Framework (Size-Shape Covariation)

This approach strictly separates size and shape according to the criterion of geometric similarity [4]. Allometry is defined as the covariation between shape and size, typically analyzed through multivariate regression of shape variables on a measure of size (usually centroid size) [3]. This method is particularly appropriate when the research question requires explicit separation of size and shape components.

Huxley-Jolicoeur Framework (Multivariate Trait Covariation)

This approach characterizes allometry as the covariation among morphological features that all contain size information [3] [4]. Allometric trajectories are represented by the first principal component in either Procrustes form space or conformation space (size-and-shape space). This framework is valuable when investigating integrated growth patterns or when the size-shape distinction is theoretically undesirable.

Allometry Analysis Frameworks

Step-by-Step Protocols

MorphoJ Protocol for Allometry Correction

4.1.1 Data Preparation and Import

• Prepare landmark data in TPS or NTS format
• Launch MorphoJ and select File > Load Project or File > Load Data
• For new projects, select File > New Project and follow the data import wizard
• Assign specimens to groups if conducting taxonomic comparisons

4.1.2 Procrustes Superimposition

• Navigate to Preprocessing > Procrustes superimposition
• Select the appropriate symmetry option for your data (bilaterally symmetric or asymmetric)
• Execute the superimposition and check for outliers using the built-in diagnostics

4.1.3 Allometry Analysis Using Regression Method

• Select Covariance > Regression from the main menu
• Choose the Procrustes coordinates as the dependent variable
• Select centroid size (or log-transformed centroid size) as the independent variable
• Execute the analysis and examine the regression statistics

4.1.4 Size Correction for Taxonomic Comparisons

• In the regression results window, select Save Residuals to obtain size-corrected shapes
• Use these residuals for subsequent taxonomic analyses such as Canonical Variate Analysis (CVA)
• Perform CVA via Covariance > CVA using the size-corrected data
• Examine the CVA plot to assess taxonomic discrimination free from allometric effects

geomorph Protocol for Allometry Correction

4.2.1 Data Preparation and Import

4.2.2 Procrustes Superimposition

4.2.3 Allometry Analysis Using Regression Method

4.2.4 Size Correction and Taxonomic Analysis

Allometry Correction Workflow

Research Reagent Solutions

Table 3: Essential Research Reagents for Geometric Morphometric Studies

Reagent/Resource	Function/Purpose	Implementation Examples
Landmark Data	Raw morphological data	2D or 3D coordinate data from specimens
Centroid Size	Size measurement	Square root of the sum of squared distances of landmarks from centroid
Procrustes Coordinates	Size-standardized shape variables	Superimposed landmark configurations
Taxonomic Groupings	A priori classification	Species, population, or subspecies identifiers
R Statistical Environment	Analysis platform	Installation of R and necessary packages
MorphoJ Software	GUI-based morphometrics	Standalone application for geometric morphometrics
geomorph R Package	Programmatic morphometrics	Comprehensive morphometric analysis in R

Troubleshooting and Methodological Considerations

Common Issues and Solutions

• Outlier Detection: Both MorphoJ and geomorph provide tools for identifying outliers in landmark data. Always check for outliers before proceeding with allometry analysis.
• Group Allometry vs. Global Allometry: In taxonomic studies, consider whether to analyze allometry within groups or across the entire dataset. Pooled within-group regression is often appropriate when groups have different size ranges.
• Size Measurement: While centroid size is standard, consider log-transformation if allometric relationships are nonlinear.

Methodological Recommendations for Taxonomic Studies

• Always test for allometry before conducting taxonomic discrimination analyses
• Use the same allometry correction method across comparative analyses within a study
• Report both uncorrected and size-corrected results to demonstrate the impact of allometry correction
• Consider biological justification when choosing between allometric frameworks
• Validate taxonomic discrimination with cross-validation methods where possible

The implementation of these protocols will enable robust taxonomic comparisons free from confounding allometric effects, strengthening morphological systematic studies through rigorous statistical control of size-related shape variation.

Allometry, the study of the relationship between size and shape, is a fundamental consideration in taxonomic geometric morphometric studies [3]. When comparing species, failing to account for allometric effects can confound true taxonomic differences with shape changes that are simple consequences of size variation [3] [28]. This application note presents a detailed protocol for correcting mandibular allometry in marmot species, based on a comprehensive study of North American marmots [13]. The methodology demonstrates how to disentangle size-related shape variation from genuine taxonomic differences, providing a framework for more accurate taxonomic assessments in geometric morphometrics research. The Vancouver Island marmot (Marmota vancouverensis) serves as a key case study, representing a distinctive insular population whose mandibular morphology suggests a long history of reduced variation and potential founder effects [13].

Theoretical Framework: Concepts of Allometry in Geometric Morphometrics

Schools of Allometric Thought

Two principal schools of thought inform allometric analysis in geometric morphometrics [3]:

• Gould-Mosimann School: Defines allometry as the covariation of shape with size, typically implemented through multivariate regression of shape variables on a size measure [3].
• Huxley-Jolicoeur School: Characterizes allometry as covariation among morphological features that all contain size information, implemented through principal component analysis in Procrustes form space or conformation space [3].

For taxonomic studies comparing closely related species, the Gould-Mosimann approach provides a more direct framework for isolating size-related shape variation from taxonomic differences [13] [3].

Levels of Allometry Relevant to Taxonomic Studies

• Static Allometry: Covariation between size and shape among adults within a population [3]
• Evolutionary Allometry: Shape changes associated with size differences across evolutionary lineages [3]
• Ontogenetic Allometry: Shape changes through growth (typically not addressed in adult-only taxonomic comparisons) [3]

Table 1: Allometry Concepts and Their Implementation in Geometric Morphometrics

Concept	Definition	Implementation in GM	Taxonomic Application
Static Allometry	Size-shape covariation within adult population	Regression of shape on size	Assessing intraspecific variation
Evolutionary Allometry	Size-shape covariation across species	Phylogenetic PCA or regression	Understanding macroevolutionary patterns
Allometric Trajectory	Pattern of shape change with size	Vector in shape space	Comparing developmental patterns across taxa
Size Correction	Removing allometric effects	Residuals from shape-size regression	Isolating non-allometric taxonomic differences

Experimental Protocol: Mandibular Allometry Correction in Marmots

Sample Preparation and Data Collection

Materials and Equipment:

• Adult marmot skulls with complete dentition (n ≥ 20 per species recommended) [13]
• High-resolution digital camera (e.g., Canon EOS Rebel T3i) with fixed focal length lens [29]
• Tripod and standardized lighting setup
• TPSDig2 software for landmark digitization [29]
• R statistical environment with geomorph, Morpho, and vegan packages

Landmark Scheme (adapted from Cardini 2023 and canine studies [13] [29]):

• Type I Landmarks (anatologically defined): 4-8 landmarks on mandible
• Semilandmarks: 2 curves along mandibular outline
• Landmark Configuration: Coronion, condylion, gonion, infradentale [29]

Detailed Workflow

The following diagram illustrates the complete workflow for correcting mandibular allometry in taxonomic studies:

Statistical Analysis Protocol

Step 1: Procrustes Superimposition

• Perform Generalized Procrustes Analysis (GPA) to remove non-shape variation (position, orientation, scale) [13] [30]
• Extract Procrustes coordinates and Centroid Size (CS) for all specimens [30]

Step 2: Assessing Allometric Effect

• Perform multivariate regression of Procrustes coordinates on ln(Centroid Size) [3] [30]
• Test significance using Procrustes ANOVA with 1000+ permutations [13]
• Calculate percentage of shape variance explained by size (R²) [29]

Step 3: Size Correction

• Extract residuals from the shape-size regression [3]
• Use residuals for subsequent taxonomic analyses [13]
• Alternative: Use within-group centering when comparing multiple groups [28]

Step 4: Taxonomic Comparisons

• Perform MANOVA on size-corrected shape residuals to test species differences [13]
• Conduct discriminant analysis to assess classification accuracy [13]
• Calculate Procrustes distances between species means [13]

Table 2: Key Statistical Tests for Allometry Correction in Taxonomic Studies

Analysis Type	Purpose	Implementation	Interpretation
Procrustes ANOVA	Test size-shape relationship	Permutation test (1000+ iterations)	Significant p-value indicates allometry present
Multivariate Regression	Quantify allometric effect	Shape coordinates ~ ln(Centroid Size)	R² indicates strength of allometry
MANOVA	Test species differences	On size-corrected residuals	Significant p-value indicates taxonomic differences
Discriminant Analysis	Classification accuracy	Cross-validated classification	Percentage correct indicates distinctness
Procrustes Distance	Magnitude of difference	Between group means	Larger distances indicate greater divergence

Case Study Application: North American Marmots

Experimental Findings from Marmot Mandibles

The protocol was applied to compare mandibular morphology across North American marmot species, with particular focus on the Vancouver Island marmot (VAN) [13]:

• Allometric Effect: Modest but significant allometry explained a portion of within-species shape variation [13]
• Taxonomic Discrimination: After size correction, shape accurately predicted taxonomic affiliation with minimal misclassification [13]
• Insular Divergence: VAN displayed the most distinctive mandibular shape and reduced morphological variation consistent with founder effects [13]
• Allometric Trajectories: Divergence in allometric trajectories was consistent with subgeneric separation [13]

Vancouver Island Marmot: Special Considerations

For the Vancouver Island marmot population, additional analyses were conducted [13]:

• Variance Comparison: Compared magnitude of variance in mandibular size and shape between VAN and its sister species on the mainland
• Founder Effect Assessment: Evaluated reduced variation as evidence of population history
• Subsampling Experiments: Assessed sensitivity of results to small sample sizes using randomized selection

Research Reagent Solutions

Table 3: Essential Materials and Software for Mandibular Allometry Studies

Item	Specification	Application	Notes
Digital Camera	18+ MP, fixed 50mm lens	Standardized image acquisition	Ensure perpendicular orientation to mandibular plane [29]
TPSDig2	Version 2.31+	Landmark digitization	Free software for landmark collection [29]
R Statistical Environment	Version 4.0+	Statistical analysis	Open-source platform for morphometrics [13]
geomorph Package	Version 4.0+	GM analyses	Procrustes ANOVA, regression, modularity tests [13]
Morpho Package	Version 2.10+	GM utilities	PCA, CVA, outlier detection [13]
Specimen Mount	Stable tripod system	Standardization	Fixed distance (40cm) for all specimens [29]

Technical Considerations and Limitations

Methodological Challenges

• Sample Size Sensitivity: Results should be tested against heterogeneous sample size using subsampling and randomized selection experiments [13]
• Landmark Precision: Use "clean" datasets without very low precision landmarks and outliers [13]
• Size Correction Controversy: Debate exists regarding whether to correct for size when it may be biologically meaningful [28]
• Intermediate Outcome Problem: Treatment effects may be mediated through size changes [28]

Integration with Other Evidence

While geometric morphometrics is powerful for taxonomic research, findings must be corroborated with an integrative approach combining multiple lines of evidence [13]:

• Molecular data (genetic distances) [31]
• Ecological and behavioral observations [13]
• Fossil evidence where available
• Biomechanical analyses [32]

Correcting for mandibular allometry is essential for accurate taxonomic comparisons in marmots and other mammalian groups. The protocol outlined here provides a robust framework for distinguishing genuine taxonomic differences from size-related shape variation. The case study of North American marmots demonstrates that while allometry explains a modest amount of within-species shape variation, substantial taxonomic signal remains after size correction. The distinctive mandibular morphology of the Vancouver Island marmot highlights the value of this approach for understanding evolutionary patterns in insular populations. This methodology can be adapted for taxonomic studies across diverse mammalian groups, particularly where allometric effects might confound phylogenetic interpretations.

Application Note: Concepts and Workflow for Allometry Visualization

Theoretical Framework of Allometric Analysis

Understanding allometry is fundamental for taxonomic geometric morphometric studies, as it refers to the size-related changes of morphological traits [3]. In evolutionary biology, two primary schools of thought guide allometric analysis. The Gould-Mosimann school defines allometry as the covariation of shape with size, typically implemented through multivariate regression of shape variables on a measure of size. Conversely, the Huxley-Jolicoeur school characterizes allometry as the covariation among morphological features that all contain size information, where allometric trajectories are represented by the first principal component as a line of best fit to the data points [3]. These conceptual approaches manifest at different biological levels: ontogenetic allometry (changes during growth), static allometry (variation within a single ontogenetic stage, typically adults), and evolutionary allometry (divergence among species or clades) [3].

Experimental Workflow for Allometry Visualization

The following diagram outlines the comprehensive workflow for analyzing and visualizing allometric trajectories in taxonomic studies, integrating both conceptual approaches and practical analytical steps:

Key Quantitative Frameworks in Allometry Studies

Table 1: Allometry Analysis Approaches in Geometric Morphometrics

Analysis Approach	Statistical Method	Morphospace Used	Key Output	Taxonomic Application
Gould-Mosimann Framework	Multivariate regression of shape on size	Procrustes shape space	Regression vectors showing shape covariation with size	Testing size-shape dependence across taxa
Huxley-Jolicoeur Framework	Principal Component Analysis (PCA)	Procrustes form space or conformation space	PC1 as primary allometric trajectory	Characterizing multivariate growth patterns
Common Allometry Model	ANCOVA with common slope	Shape space	Common allometric vector	Testing shared evolutionary constraints
Unique Allometry Model	Homogeneity of slopes tests	Shape space	Group-specific vectors	Identifying divergent evolutionary trajectories

Protocols for Allometric Trajectory Analysis

Protocol 1: Data Acquisition and Preprocessing

Purpose: To acquire and prepare high-quality landmark data for allometric analysis in taxonomic studies.

Materials and Reagents:

• High-resolution specimens (physical or digital)
• Imaging equipment (microscope, camera, or scanner)
• Digitization software (tpsDig, MorphoJ, or equivalent)
• Statistical computing environment (R with geomorph package)

Procedure:

• Specimen Selection: Select specimens representing taxonomic groups of interest, ensuring adequate sample size (minimum n=15-20 per group recommended) [2].
• Landmark Digitization:
  • Place Type I (anatomical), Type II (mathematical), and Type III (sliding semilandmark) landmarks consistently across all specimens
  • Record 2D or 3D coordinates using appropriate digitization software
  • For 2D analyses, ensure consistent orientation and magnification [2]
• Data Quality Control:
  • Assess measurement error through replicate digitization
  • Identify and address outliers using Procrustes distance metrics
  • Test for statistical power using preliminary PCA [2]
• File Organization: Export landmark coordinates in standard format (TPS, NTS, or equivalent) for subsequent analysis.

Protocol 2: Procrustes Superimposition and Shape Variable Extraction

Purpose: To remove non-shape variation (position, orientation, scale) and extract pure shape variables for allometric analysis.

Materials:

• Landmark coordinate data
• R statistical environment with geomorph package [2] [33]
• Alternative: MorphoJ, PAST, or other geometric morphometrics software

Procedure:

• Generalized Procrustes Analysis (GPA):
  • Center each specimen to a common origin
  • Scale to unit centroid size
  • Rotate to minimize Procrustes distances among specimens
• Shape Variable Generation:
  • Extract Procrustes coordinates as shape variables
  • Calculate centroid size as a measure of isometric size
  • Assess superimposition quality using Procrustes residuals
• Shape Space Validation:
  • Confirm that major shape variation is captured in preliminary PCA
  • Verify that technical artifacts do not dominate biological signal

Protocol 3: Allometric Trajectory Analysis and Visualization

Purpose: To quantify and visualize allometric patterns among taxonomic groups.

Materials:

• Procrustes shape coordinates
• Centroid size values
• R with geomorph, ggplot2, and other visualization packages

Procedure:

• Preliminary Shape Visualization:
  • Perform PCA on Procrustes shape coordinates
  • Visualize group separation in morphospace using scatterplots
  • Create wireframe graphs to illustrate shape changes along PCs
• Multivariate Allometry Test:
  • Perform multivariate regression of shape on size: procD.lm(shape ~ size) in geomorph
  • Assess significance using permutation tests (1000+ permutations)
  • Calculate and report effect size (R²) for allometry
• Allometric Trajectory Comparison:
  • Test for common allometric trajectory using trajectory.analysis()
  • Compare vector correlation angles between groups
  • Test for differences in trajectory lengths using ANOVA
• Visualization Generation:
  • Create regression plots of shape vs. size with group discrimination
  • Generate deformation grids or vectors illustrating allometric shape changes
  • Plot group trajectories in principal component space

The following diagram illustrates the core analytical pipeline for comparing allometric trajectories across taxonomic groups:

Research Reagent Solutions for Allometry Studies

Table 2: Essential Research Tools for Allometric Analysis in Geometric Morphometrics

Tool Category	Specific Software/Package	Primary Function	Application in Allometry Studies
Statistical Programming	R with geomorph package [33]	Comprehensive GM analysis	Procrustes ANOVA, trajectory analysis, phylogenetic correction
Landmark Digitization	tpsDig Suite	2D/3D landmark placement	Coordinate acquisition from specimen images
Graphical Visualization	ggplot2 (R) [34]	Publication-quality graphs	Creating scatterplots, regression visuals, and trajectory plots
Shape Visualization	MorphoJ	User-friendly GM analysis	PCA, regression, deformation grid visualization
3D Data Processing	MeshLab	3D surface processing	Handling 3D scan data and surface models
Phylogenetic Analysis	phytools (R)	Phylogenetic comparative methods	Phylogenetic correction of allometric analyses

Data Interpretation and Taxonomic Applications

Protocol 4: Statistical Interpretation and Hypothesis Testing

Purpose: To correctly interpret statistical outputs and make biologically meaningful taxonomic inferences from allometric analyses.

Materials:

• Statistical outputs from Protocols 2-3
• Phylogenetic information (if available)
• Taxonomic framework for group definitions

Procedure:

• Allometry Significance Assessment:
  • Interpret permutation p-values for multivariate regression
  • Calculate and report effect sizes (R² values) for allometric relationships
  • Distinguish statistical significance from biological importance
• Trajectory Comparison Interpretation:
  • Interpret vector correlation angles: smaller angles indicate more parallel trajectories
  • Assess trajectory magnitude differences: indicate rate of shape change per unit size
  • Evaluate statistical support for trajectory differences (p-values)
• Taxonomic Decision Framework:
  • Integrate allometric results with other data (molecular, ecological)
  • Consider phylogenetic context when interpreting allometric patterns
  • Evaluate whether allometric differences support taxonomic distinctions

Key Quantitative Outputs and Their Interpretation

Table 3: Interpreting Statistical Results in Allometric Analyses

Statistical Output	Interpretation	Taxonomic Significance
Multivariate R² (allometry)	Proportion of shape variance explained by size	High R² indicates strong size-dependence; may complicate taxonomic discrimination
Vector Correlation Angle	Similarity of allometric trajectories between groups	Small angles suggest shared developmental constraints; large angles indicate divergent evolution
Trajectory Magnitude Difference	Relative rate of shape change per unit size	Different magnitudes suggest heterochronic evolution or differential constraint
Common vs. Unique Allometry (Interaction p-value)	Test of homogeneity of allometric slopes	Significant interaction supports taxonomic distinction based on allometric pattern
Phylogenetic Signal (K-mult)	Degree of phylogenetic constraint in shape/size	High K suggests phylogenetically structured variation; low K suggests ecological adaptation

Advanced Applications and Integration

Protocol 5: Phylogenetically Informed Allometric Analysis

Purpose: To incorporate phylogenetic relationships into allometric analyses for more evolutionarily meaningful comparisons.

Materials:

• Phylogenetic tree of study taxa
• R with geomorph and phytools packages
• Shape and size data from previous protocols

Procedure:

• Phylogenetic Signal Assessment:
  • Calculate K-mult statistic for shape and size using physignal()
  • Test significance via permutation
  • Interpret results: K > 1 indicates stronger phylogenetic signal than Brownian motion expectation [33]
• Phylogenetic Comparative Analysis:
  • Implement phylogenetic Generalized Least Squares (pGLS) for allometry
  • Account for phylogenetic non-independence in regression models
  • Compare phylogenetic and non-phylogenetic models
• Phylogenetic Visualization:
  • Map shape data onto phylogenetic trees
  • Visualize allometric trajectories in phylogenetic context
  • Reconstruct ancestral shapes and allometric patterns

Implementation Considerations for Taxonomic Studies

Effective application of allometric visualization in taxonomy requires attention to several practical considerations. First, sample size adequacy must be ensured, with minimum recommendations of 15-20 specimens per group, though power analysis should guide specific study designs [2]. Second, measurement error should be quantified through replicate digitization and included in error assessments. Third, the choice between common and unique allometry models has profound implications for taxonomic interpretations—common allometry suggests shared developmental constraints, while unique allometry provides evidence for evolutionary divergence [3] [33].

Recent studies of Euarchontoglires endocranial shape demonstrate the taxonomic value of these approaches, showing how allometric trajectory analysis can reveal fundamental differences in how shape and size covary among major clades, with some groups like platyrrhines showing strong size-shape relationships while rodents exhibit remarkable diversification despite weak allometric constraints [33]. These patterns provide crucial evidence for understanding the evolutionary processes underlying taxonomic diversity.

Solving Common Challenges in Allometry Correction for Complex Taxonomic Groups

In taxonomic geometric morphometric studies, allometry—the pattern of how organismal shape changes with size—provides fundamental insights into evolutionary and developmental processes [14]. However, a pervasive methodological challenge arises when allometry levels are confounded, such as when analyses inadvertently combine specimens from different ontogenetic stages (e.g., juveniles and adults) or from populations with distinct evolutionary trajectories [3] [35]. Such confounding introduces non-independence in data that violates the assumptions of standard statistical models, potentially leading to biased allometric estimates and erroneous taxonomic conclusions [36] [35].

This protocol outlines a structured framework for identifying and statistically addressing confounded allometry levels. We emphasize practical solutions using the R statistical environment and the geomorph package [37] [27], which provide robust tools for diagnosing confounding and implementing mixed models that can attribute variation to its correct source [36]. By applying these methods, researchers can improve the accuracy of their allometric corrections and strengthen the validity of subsequent taxonomic inferences.

Theoretical Background: Schools of Allometric Thought

Understanding how to address confounding requires grounding in the two primary conceptual frameworks for studying allometry, which are implemented differently in morphometric analyses.

Table 1: Two Primary Schools of Allometric Thought

School of Thought	Core Definition of Allometry	Typical Analytical Approach in GMM	Implication for Confounding
Gould-Mosimann School	Covariation between shape and size as separate concepts [3] [4].	Multivariate regression of shape variables (e.g., Procrustes coordinates) on a size measure (e.g., centroid size) [3] [4].	Confounding creates multiple, correlated size predictors, violating regression assumptions of independence.
Huxley-Jolicoeur School	Covariation among morphological features that all contain size information [3] [4].	Finding the major axis of covariation in a form space (size not removed), often via the first principal component (PC1) of form [3] [4].	Confounding introduces multiple, distinct axes of covariation, which may be inaccurately summarized by a single PC.

The Gould-Mosimann school's size-shape regression is most common in geometric morphometrics. However, when data contain mixed ontogenetic stages or populations, the single, universal size variable this approach requires may not exist, creating a fundamental problem for analysis [3].

Diagnosing Confounded Allometry Levels

Before applying corrective measures, researchers must diagnose potential confounding. The following workflow provides a systematic diagnostic approach.

Figure 1: A diagnostic workflow for detecting confounded allometry levels in a morphometric dataset. PCA: Principal Component Analysis.

Visual Diagnostics

• Allometric Plots: The most straightforward diagnostic is a plot of shape (e.g., the first principal component of shape or a regression score) against a size measure (e.g., centroid size), with points colored by their putative group (e.g., population or ontogenetic stage). Visibly distinct allometric slopes or intercepts for different groups suggest confounding [37] [35].
• Principal Component Analysis (PCA): A PCA of the shape data where points are colored by size can reveal whether larger specimens from one group cluster with smaller specimens from another, indicating that size variation is not independent of group identity [2].

Statistical Diagnostics

Formal statistical tests are essential to confirm visual diagnoses. Using the procD.lm function in the geomorph R package, one can test for significant differences in allometric slopes among groups.

A significant interaction term (log(Csize):Population) provides statistical evidence that allometric slopes differ among groups, confirming confounding [37].

Statistical Frameworks for Addressing Confounding

Once confounding is diagnosed, researchers can employ several statistical frameworks to account for it.

Table 2: Statistical Frameworks for Addressing Confounded Allometry

Framework	Core Principle	Ideal Use Case	Implementation in R (geomorph)
Generalized Linear Mixed Models (GLMM)	Attributes variation to fixed (e.g., population) and random effects (e.g., individual variation, distortion), modeling heterogeneous residual variation [36].	Datasets containing distorted specimens or hierarchical data structure where not all confounding factors are of direct interest [36].	Implemented via procD.lm with random effects specified, or using lme4 for complex designs.
Phylogenetic Comparative Methods	Accounts for the non-independence of species due to shared evolutionary history, which can confound evolutionary allometry if ignored [35].	Interspecific (evolutionary) allometry studies where species are the data points and a phylogeny is available [35].	procD.pgls function in geomorph for phylogenetic generalized least squares.
Model-Based Variance Structures	Explicitly models heteroscedasticity (non-constant variance) using exponential or power-of-the-mean variance functions, rather than assuming homogeneous variance [38].	Ontogenetic allometry studies where the amount of shape variation around the allometric line changes predictably with size or differs by group [38].	Can be implemented using the nlme package or the gnls function, with a defined variance structure.

Detailed Experimental Protocol

This section provides a step-by-step protocol for implementing a GLMM-based approach, which is particularly powerful for handling the non-biologic variation introduced by fossil distortion or mixed populations [36].

Workflow for a GLMM Analysis

The following workflow outlines the key steps for a GLMM analysis designed to address confounding, from data preparation to the interpretation of size-corrected shapes.

Figure 2: A generalized workflow for correcting allometry using a GLMM framework. GPA: Generalized Procrustes Analysis.

Step-by-Step Instructions

• Data Preparation and Procrustes Fitting

  • Input: 2D or 3D landmark coordinates.
  • Action: Perform a Generalized Procrustes Analysis (GPA) to superimpose specimens, removing differences due to position, orientation, and scale. This yields Procrustes shape coordinates and Centroid Size.
  • R Code:

• Define and Code Confounding Factors

  • Identify all known or potential grouping factors (e.g., Population, OntogeneticStage, PreservationQuality).
  • Code these factors as categorical variables in your data frame. For mixed models, decide which factors are fixed effects (of direct interest, e.g., Population) and which are random effects (sources of variation not of direct interest, e.g., IndividualID for repeated measures, or DistortionLevel).

• Fit the Generalized Linear Mixed Model (GLMM)

  • Use a function capable of handling the complex error structures of morphometric data. The procD.lm function in geomorph is highly flexible.
  • R Code for a complex model:

• Model Validation

  • Check the model's residuals for homogeneity of variance and normality. Plot residuals versus fitted values and against all predictors.
  • R Code:

• Extract Allometry-Free Shapes for Taxonomy

  • Once an adequate model is fitted, the residuals represent shape variation that is not explained by the model's predictors (e.g., size and population). These residuals are your "allometry-free" shapes.
  • R Code:

The Scientist's Toolkit: Essential Research Reagents

Table 3: Key Software and Statistical Tools for Allometry Correction

Tool / Reagent	Type	Primary Function in Protocol	Key Reference / Source
R Statistical Environment	Software	The core platform for all statistical analyses and graphical visualizations.	R Project
geomorph R Package	Software Library	Performs GPA, allometric regressions (procD.lm, procD.allometry), phylogenetic comparisons, and model diagnostics. [37] [27]	[37] [27]
nlme R Package	Software Library	Fits linear and nonlinear mixed effects models with various variance structures, useful for heteroscedastic data. [38]	[38]
Centroid Size	Morphometric Variable	A standardized, geometric measure of size, calculated as the square root of the sum of squared distances of all landmarks from their centroid. Serves as the primary size proxy in allometric regressions.	Standard in GMM
Phylogenetic Tree	Data Structure	A hypothesis of evolutionary relationships required for phylogenetic comparative methods (e.g., PGLS) to avoid confounding due to common ancestry. [35]	[35]

Confounded allometry levels present a significant obstacle in taxonomic morphometrics, but they can be effectively addressed through careful diagnostic practices and modern statistical modeling. The framework outlined here—emphasizing visualization, hypothesis testing, and the application of GLMMs and phylogenetic methods—empowers researchers to disentangle complex sources of variation. By formally incorporating potential confounders like population structure or ontogenetic stage into statistical models, scientists can extract more reliable allometry-free shape residuals, thereby solidifying the foundation for robust taxonomic decisions and evolutionary inferences.

In taxonomic studies using geometric morphometrics (GM), accurately correcting for allometry—the relationship between shape and size—is crucial for identifying true taxonomic signals. However, this endeavor is frequently complicated by small sample sizes, a common limitation when studying rare, cryptic, or fossil species. Small samples can lead to increased error in estimating population mean shape and variance, potentially confounding allometric corrections and obscuring the morphological differences vital for taxonomic discrimination [39]. This application note outlines robust methodological strategies and protocols for conducting reliable allometric analyses within the constraints of limited data, ensuring that taxonomic conclusions are both valid and reproducible.

The Impact of Sample Size on Morphometric Analyses

The challenges of small sample sizes are not merely theoretical. Empirical research demonstrates that reducing sample size directly impacts key morphometric parameters. A 2024 study on bat skulls found that smaller samples led to increased shape variance and less stable estimates of mean shape [39]. While measures of centroid size may remain relatively stable, the estimation of shape is particularly sensitive to limited sampling [39].

Table 1: Impact of Decreasing Sample Size on Shape Estimates (Empirical Data from Bat Skulls)

Sample Size Scenario	Effect on Mean Shape Estimate	Effect on Shape Variance	Reliability for Taxonomic Discrimination
Large Sample (n > 70)	Stable and accurate	Estimated with high precision	High
Moderate Sample (n ~ 30)	Increased estimation error	Moderate increase	Moderate
Small Sample (n < 20)	High error and instability	Greatly inflated, unreliable	Low

These findings underscore that in small-sample contexts, standard allometric corrections can be misled by inaccurate shape estimates, potentially resulting in flawed taxonomic interpretations.

Robust Methodological Framework

Core Concepts in Allometry Correction

Two primary schools of thought inform the study of allometry in geometric morphometrics, a distinction that remains relevant for choosing analytical methods [3] [4]:

• The Gould-Mosimann School defines allometry as the covariation between size and shape. In this framework, size is treated as an external variable, and allometry is typically analyzed using the multivariate regression of shape variables on a size measure like centroid size [3] [4].
• The Huxley-Jolicoeur School characterizes allometry as the covariation among morphological features that all contain size information. Here, the first principal component (PC1) in a space that includes size information (such as Procrustes form space) is often interpreted as the allometric vector [3] [4].

Performance of Allometric Methods under Limited Data

Simulation studies comparing the performance of different methods under various conditions provide critical guidance for small-sample research. A 2022 performance comparison found that while all methods are logically consistent in the absence of noise, they differ when dealing with the residual variation typical of real biological data [4].

Table 2: Performance Comparison of Allometry Analysis Methods

Method	Conceptual School	Key Strength	Performance with Isotropic Noise	Recommendation for Small Samples
Regression of Shape on Size	Gould-Mosimann	Directly tests shape-size relationship	Good [4]	Recommended; directly addresses the allometry question.
PC1 of Shape	Gould-Mosimann	Captures major axis of shape variation	Weaker than regression [4]	Use with caution; PC1 may not reflect allometry.
PC1 of Conformation/Size-and-Shape Space	Huxley-Jolicoeur	Characterizes covariation including size	Very good, close to simulated truth [4]	Recommended; robust performance.
PC1 of Boas Coordinates	Huxley-Jolicoeur	Similar to conformation space	Very good, almost identical to conformation [4]	Recommended; robust performance.

For small-sample studies, the multivariate regression of shape on size and the PC1 in conformation space (or with Boas coordinates) are particularly recommended due to their robust performance and logical consistency within their respective frameworks [4].

Experimental Protocols for Small-Sample Studies

Comprehensive Workflow for Robust Analysis

The following workflow integrates multiple strategies to enhance the robustness of allometry correction in taxonomic studies with limited data. This protocol is designed to be iterative, with results from preliminary analyses informing final decisions.

Detailed Protocol Steps

Step 1: Assess Measurement Error (Prior to Allometric Analysis)

Purpose: To ensure that observed shape variation is biological and not an artifact of data collection, which is critical when small effects are being sought.

• Procedure: Collect replicate landmark configurations for a subset of specimens. Perform a Procrustes ANOVA to partition variance into components of individual variation and measurement error [2].
• Small-Sample Adaptation: Even with a few specimens (e.g., 3-5), replicated digitization can provide an estimate of measurement error. A high error-to-individual variance ratio signals that conclusions are unreliable.

Step 2: Identify and Evaluate Outliers

Purpose: To prevent a single aberrant specimen from disproportionately influencing results in a small sample.

• Procedure: Calculate Procrustes distances of all specimens from the mean shape. Visually inspect specimens with the largest distances to determine if they are true biological outliers or data collection errors [2].
• Small-Sample Adaptation: Use conservative criteria for outlier exclusion; removal should be justified by biological or technical reasons, not statistical convenience.

Step 3: Evaluate Statistical Power

Purpose: To transparently acknowledge the analytical limitations of the dataset.

• Procedure: If a specific effect size is of interest (e.g., a known shape difference between species), a pilot analysis or a simulated power analysis can be conducted.
• Small-Sample Adaptation: Report statistical power alongside results. In the absence of formal power analysis, explicitly state that findings are preliminary and require validation with larger samples [2] [39].

Step 4: Conduct Preliminary Allometric Analyses

Purpose: To determine if and how allometry manifests within the limited dataset.

• Procedure:
  • Perform a multivariate regression of shape on centroid size (log-transformed) and assess significance via permutation [3] [4].
  • In parallel, perform a Principal Component Analysis (PCA) in conformation space (size-and-shape space) and examine the correlation of PC1 with centroid size [4].
• Small-Sample Adaptation: Use a high number of permutations (e.g., 10,000) to improve the reliability of p-value estimates for the regression.

Step 5: Apply Multiple Correction Methods and Compare

Purpose: To ensure taxonomic conclusions are not dependent on a single, potentially unstable, allometry correction technique.

• Procedure:
  • Regression-based correction: Use the residuals from the multivariate regression of shape on size as size-corrected shape variables [4].
  • Vector subtraction: If a strong allometric vector is identified, project the data orthogonally to this vector to remove its influence.
• Small-Sample Adaptation: Compare the taxonomic signal (e.g., group separation in morphospace) in both the raw data and the data corrected by different methods. A robust signal will persist across correction approaches.

The Scientist's Toolkit: Essential Research Reagents & Materials

Table 3: Essential Materials and Software for Robust Small-Sample GM

Item Name	Function/Application	Small-Sample Consideration
High-Resolution 2D/3D Digitizer	(e.g., DSLR camera, micro-CT scanner) Captures morphological data.	Maximizes information from few specimens; reduces measurement error [39].
tpsDig2 Software	Digitizes landmarks and semi-landmarks on 2D images.	Standardized digitization is critical to minimize added noise [39].
R Programming Language	Flexible statistical computing environment.	Enables implementation of specialized methods and permutation tests.
geomorph R Package	Comprehensive GM analysis toolkit.	Performs Procrustes ANOVA, allometry regression, and permutation tests [39].
Generalized Procrustes Analysis (GPA)	Algorithm to align specimens into shape space.	Foundational step; ensures differences are not due to position or orientation [40].
Centroid Size	A geometric measure of size (square root of sum of squared distances from landmarks to centroid).	The standard size metric for allometric regression in GM [3] [4].

Navigating allometry correction in geometric morphometric taxonomy with limited data is challenging but not intractable. By implementing a rigorous workflow that includes measurement error analysis, a focus on effect sizes, and the application of multiple robust allometric methods like regression-on-size and PC1-in-conformation-space, researchers can strengthen the validity of their taxonomic inferences. The protocols outlined here provide a conservative and transparent framework for making the most reliable conclusions possible from small samples, while clearly acknowledging their inherent limitations.

Dealing with Non-Linear Allometric Relationships

Allometry, the study of how organismal traits change with size, is a foundational concept in evolutionary biology and taxonomy. In geometric morphometrics, which quantitatively analyzes shape variation, correcting for allometry is essential for isolating taxonomic signals from size-related shape changes. The standard power-law model of allometry, while useful, often fails to capture the complexity of biological growth and variation, leading to a growing focus on non-linear allometric relationships. This protocol outlines the conceptual frameworks and detailed methodologies for detecting, analyzing, and correcting for these non-linear patterns within taxonomic studies, providing a crucial tool for unbiased systematic and phylogenetic research.

Conceptual Framework of Allometric Schools

Two primary schools of thought guide allometric studies in geometric morphometrics, each with implications for understanding non-linearity.

The Gould–Mosimann School

This framework strictly separates size and shape. Allometry is defined as the covariation of shape with size, where size is an external variable. In geometric morphometrics, this is typically implemented as a multivariate regression of shape variables (e.g., Procrustes coordinates) on a measure of size, such as centroid size [3] [4]. This approach is inherently well-suited for detecting non-linearity, as the regression model can be extended beyond a linear fit.

The Huxley–Jolicoeur School

This school defines allometry as the covariation among morphological features that all contain size information, without a prior separation of size and shape. Allometric trajectories are characterized by the first principal component (PC1) in a form space (size-and-shape space) [3] [4]. Non-linearity in this context may manifest as curvature in the data cloud within the conformation space, which a straight-line PC1 model might not adequately capture.

Quantitative Models and Data Presentation

The Traditional Power-Law and Its Limitations

The foundational allometric model is the power function: ( Y = \beta M^\alpha ), where:

• ( M ) is a measure of size (e.g., body mass).
• ( Y ) is a trait of interest (e.g., organ mass).
• ( \beta ) is the proportionality constant.
• ( \alpha ) is the allometric scaling exponent [41].

This model becomes a straight line after logarithmic transformation: ( \log(Y) = \log(\beta) + \alpha \log(M) ), allowing for linear regression. However, the assumption of a single, universal exponent (( \alpha )) across the entire size range of a species or clade is often biologically unrealistic and can introduce substantial bias if violated [42] [43].

The Emergence of Multi-Scaling Allometry

Recent research challenges the "uni-scaling" assumption, proposing "multi-scaling allometry" where the scaling exponent ( \alpha ) varies depending on the centile of the trait distribution [43]. This is formalized by fitting the power-law to the qth centile curve, yielding an exponent ( \alpha(q) ) that is a function of the centile index.

Table 1: Comparison of Uni-Scaling vs. Multi-Scaling Allometry

Feature	Uni-Scaling Allometry	Multi-Scaling Allometry
Core Assumption	A single scaling exponent ((\alpha)) applies to the entire population.	The scaling exponent ((\alpha(q))) can vary across different centiles of the population.
Model	( B = C \cdot A^{\alpha} )	( B = C(q) \cdot A^{\alpha(q)} ) for the ( q )th centile
Interpretation	All individuals follow the same growth strategy.	Individuals in different parts of the size distribution may employ distinct growth strategies.
Analysis Method	Standard linear regression on log-transformed data.	Quantile regression on log-transformed data.

This multi-scaling framework has been validated in diverse systems, demonstrating that the height-weight scaling exponent in children varies significantly with age and centile, and that the brain-body size relationship in mammals shows similar multi-scaling properties [43].

Experimental Protocols for Non-Linear Allometry

The following diagram outlines the core workflow for analyzing non-linear allometric relationships in a geometric morphometric context.

Detailed Protocol for Analysis and Correction

Protocol 1: Detection of Non-Linear Allometry using Regression & Residuals

This protocol tests the assumption of linearity in the shape-size relationship.

• Data Preparation: Perform a Generalized Procrustes Analysis (GPA) on the raw landmark coordinates to align specimens into shape space [4] [44]. Calculate Centroid Size (the square root of the sum of squared distances of all landmarks from their centroid) for each specimen as the size variable [3].
• Model Fitting: Perform a multivariate regression of Procrustes shape coordinates on Centroid Size. This is a standard function in morphometric packages (e.g., procD.lm in the geomorph R package) [4] [44].
• Assessing Linearity: Visually inspect a plot of regression scores (or Procrustes distance to mean shape) against Centroid Size. A linear relationship will appear as a straight line, while a non-linear one will show clear curvature.
• Test for Quadratic Term: Statistically test for non-linearity by adding a quadratic term (Centroid Size²) to the regression model. A significant effect of the quadratic term indicates a deviation from linearity. Alternatively, plot the regression residuals against Centroid Size; a non-random pattern (e.g., U-shaped) suggests a poor fit of the linear model.

Protocol 2: Implementing Multi-Scaling Allometry with Quantile Regression

This protocol characterizes allometry across different quantiles of the size distribution, ideal for detecting heterogeneous growth strategies.

• Data Transformation: Reduce the multidimensional shape data to a single, meaningful scalar variable that captures the primary allometric trend. This can be the regression score from a linear model of shape on size, or the score from the PC1 of the shape data if it is highly correlated with size.
• Quantile Regression: Apply quantile regression to model the relationship between the allometric shape score (Y) and Centroid Size (X). Unlike ordinary regression which models the mean, quantile regression fits models for specific quantiles (e.g., 0.1, 0.5, 0.9). This is implemented in R using the quantreg package [43].
• Exponent Comparison: For each quantile ( q ), the slope of the regression line ( \alpha(q) ) is the allometric exponent for that quantile. Compare the exponents across quantiles. If ( \alpha(q) ) is constant, the relationship is uni-scaling. If ( \alpha(q) ) varies systematically with ( q ), it provides strong evidence for multi-scaling allometry [43].
• Biological Interpretation: Interpret the variation in exponents. For example, in a taxonomic context, different exponents at high and low quantiles could indicate that larger species within a group follow a different allometric trajectory than smaller species, which has implications for evolutionary diversification.

Protocol 3: Size Correction Based on Non-Linear Models

Once non-linearity is established, correct for its effects before taxonomic analysis.

• Model Selection: Choose the best-fitting model for the allometric relationship (e.g., a quadratic model or a model based on quantile regression at the median).
• Compute Allometric Residuals: Calculate the residuals from the selected non-linear model. These residuals represent the shape variation that is not explained by the non-linear allometric relationship with size.
• Create Size-Corrected Shapes: Add the residuals back to the predicted shape from the overall mean shape. This creates a set of size-corrected shapes where the non-linear allometric effect has been removed. The formula is: Size-Corrected Shape = Grand Mean Shape + Model Residuals
• Validation: Confirm that the correlation between the corrected shapes and size is no longer significant, or is substantially reduced. Subsequent taxonomic analyses (e.g., PCA, discriminant analysis) should be performed on these size-corrected shapes.

The Scientist's Toolkit

Research Reagent Solutions

Table 2: Essential Tools for Analyzing Non-Linear Allometry

Tool / Reagent	Function / Application
R Statistical Environment	The primary platform for statistical computing and analysis in morphometrics. It is free, open-source, and has extensive packages for morphometrics [45].
geomorph R Package	A comprehensive package for geometric morphometric analyses. It contains functions for GPA (gpagen), multivariate regression (procD.lm), and other allometry-related tests [45] [44].
quantreg R Package	Essential for implementing multi-scaling allometry. It provides functions for fitting quantile regression models [43].
gmShiny Web Application	A user-friendly, web-based interface for many functions in geomorph. It is particularly useful for visualizing allometric patterns and for researchers less comfortable with coding [46].
3D Landmark Digitizing Software	Software like IDAV Landmark Editor is used to collect the primary 3D coordinate data from specimens, which forms the raw data for all subsequent analyses.
Centroid Size	A standardized, geometrically unbiased measure of size calculated from landmark coordinates. It serves as the independent variable in allometric regressions [3] [4].

Logical Relationship of Methodologies

The decision to apply a linear or non-linear allometric correction depends on the outcome of initial diagnostic tests. The following diagram illustrates this logical workflow.

## Application Notes & Protocols

In taxonomic geometric morphometric (GM) studies, a foundational goal is to identify and characterize shape differences that reflect evolutionary relationships and distinctions among groups. However, size-related shape variation, known as allometry, often confounds these analyses. When allometric trajectories differ between the structures of an organism—a phenomenon termed differential allometry—the complexity of the problem increases. Failing to account for this can lead to misinterpreting allometrically induced shape changes as genuine taxonomic signals.

These Application Notes provide a structured framework for diagnosing, analyzing, and accounting for differential allometry across modular structures. The protocols are designed for researchers conducting taxonomic studies using geometric morphometrics, enabling them to isolate allometric effects and reveal underlying taxonomic shape variation.

## 2. Theoretical Framework: Key Concepts

• Allometry: The pattern of covariation between shape and size. Static allometry specifically refers to this pattern among individuals within a single population or species at the same developmental stage [3].
• Differential Allometry: Occurs when different anatomical parts or modules within a single organism exhibit distinct allometric trajectories. This can be due to divergent functional demands, developmental timing, or evolutionary pressures [47] [48].
• Modularity: The concept that an organism's structure is composed of semi-independent units, or modules, which are highly integrated internally but relatively independent of one another [49]. Common examples include the distinction between the cranium and mandible or between the blade and haft of a stone tool [47].
• Integration: The degree of covariation among different parts of a structure. High integration suggests that parts evolve in a coordinated manner, while low integration allows for more independent evolution [49].

The following table summarizes the schools of thought in allometric studies, which inform the analytical approaches.

Table 1: Schools of Thought in Allometric Analysis

School of Thought	Core Definition of Allometry	Typical Analytical Approach in GM
Gould–Mosimann School	Covariation of shape with size [3].	Multivariate regression of shape variables (Procrustes coordinates) on a measure of size (e.g., centroid size) [3].
Huxley–Jolicoeur School	Covariation among morphological features all containing size information [3].	Principal Component Analysis (PCA) in Procrustes form space or conformation space; the first PC often captures allometric variation [3].

## 3. Workflow for Analyzing Differential Allometry

The following diagram outlines the core workflow for a study investigating modularity and differential allometry, from data acquisition to final interpretation.

## 4. The Scientist's Toolkit: Essential Research Reagents & Software

Table 2: Essential Tools for Geometric Morphometric Analysis of Allometry and Modularity

Tool / Reagent	Type	Primary Function in Analysis	Example Use Case
MorphoJ	Software Package	Integrated program for geometric morphometrics; performs Procrustes ANOVA, regression, modularity tests, and CVA [22].	Testing predefined modular hypotheses and performing pooled within-group regression for allometry correction [22].
R package 'geomorph'	Software Package (R)	Comprehensive package for GM; functions for GPA, modularity tests, phylogenetic analyses, and allometry [50] [2].	Conducting a full pipeline from GPA to advanced modularity and integration tests in a customizable environment.
TPSDig2	Software Utility	Digitizes landmarks from 2D image files [50].	Creating landmark data files (TPS format) from images of specimens for later analysis in MorphoJ or R.
Structured-Light 3D Scanner	Hardware	Creates high-resolution 3D models of specimens for 3D landmarking [17].	Capturing the complex 3D geometry of an astragalus bone or cranium for a comprehensive shape analysis [51].
Centroid Size	Morphometric Variable	The square root of the sum of squared distances of all landmarks from their centroid; the standard measure of size in GM [3].	Serving as the independent variable (proxy for size) in multivariate regression of Procrustes coordinates to assess allometry [51].

## 5. Detailed Experimental Protocols

### Protocol 1: Data Collection and Preparation

Objective: To acquire high-quality landmark data that accurately captures the morphology of the structures under study.

• Landmarking:

  • Homologous Landmarks: Digitize Type I (discrete anatomical points), Type II (maxima of curvature), and Type III (extremal points) landmarks on 2D images or 3D models of all specimens [17].
  • Semi-landmarks: For curves and surfaces where homologous points are scarce, place semi-landmarks to capture outline and surface geometry. These must be slid to minimize bending energy against a reference specimen during Procrustes alignment [17].
  • Data Export: Save the final landmark configurations in a standard format (e.g., TPS, NTS) compatible with downstream software like MorphoJ or R.

• Procrustes Superimposition:

  • Perform a Generalized Procrustes Analysis (GPA). This procedure removes the non-shape effects of position, orientation, and scale by:
    1. Centering each configuration on its centroid (0,0,0).
    2. Scaling each to a common unit size (Centroid Size = 1).
    3. Rotating configurations to minimize the sum of squared distances between corresponding landmarks [17] [3].
  • The output is a set of Procrustes coordinates (the shape variables) and centroid size for each specimen, which are used in all subsequent analyses.

### Protocol 2: Testing for Morphological Modularity

Objective: To statistically test whether a hypothesized division of a structure into parts (modules) is supported by the pattern of shape covariation.

• Define a Priori Hypotheses: Based on anatomy, function, or development, define candidate modular patterns. For example, for a Clovis point, the hypothesis might be a two-module structure: Blade vs. Haft [47]. For a spider body, it might be Prosoma vs. Opisthosoma [48].

• Perform Modularity Test:

  • In MorphoJ: Use the "Integration and Modularity" menu. Select the landmark partitions corresponding to your hypothesis. The software will calculate the covariance ratio coefficient (CR). A CR value significantly greater than 1 indicates that the null hypothesis of complete integration is rejected, supporting modularity [47].
  • In R/geomorph: Use the modularity.test() function, providing the GPA-aligned data and the landmark partition. The function provides a test statistic and a p-value based on a permutation procedure.

• Interpretation: A significant result confirms that the hypothesized modules are more integrated within themselves than with each other, validating their treatment as semi-independent units for allometric analysis.

### Protocol 3: Analyzing Global and Modular Allometry

Objective: To quantify the allometric relationship between shape and size both for the entire structure and for individual modules.

• Global Allometry:

  • Perform a multivariate regression of Procrustes coordinates on log-transformed centroid size.
  • The significance of the regression is tested with a permutation test (typically 10,000 permutations). A significant p-value indicates that allometry is a major source of shape variation in the sample [3] [51].
  • The strength of the relationship can be assessed with an R² value.

• Differential Allometry Analysis:

  • For each module defined in Protocol 2, extract the relevant subset of Procrustes coordinates.
  • Perform a separate multivariate regression of each module's shape on global log-transformed centroid size (or a relevant proxy for overall body size).
  • Compare allometric vectors: Statistically compare the direction and magnitude of the allometric vectors for each module. This can be done using methods such as the dot product between vectors or by testing for homogeneity of slopes in a multivariate context [48].

Table 3: Statistical Methods for Allometric and Modularity Analysis

Analysis Goal	Statistical Test	Interpretation of Key Result	Example from Literature
Global Allometry	Multivariate Regression of shape on size (e.g., in MorphoJ or geomorph)	A significant p-value (p < 0.05) indicates size is a significant predictor of shape (allometry is present).	Astragalus shape in ruminants showed significant allometry (p-value = 0.001), with larger species having more robust bones [51].
Modularity	Covariance Ratio Test (CR)	A CR value significantly > 1 with a p-value < 0.05 supports the hypothesized modular structure.	Clovis points showed significant modularity (CR > 1, p < 0.05) when partitioned into blade and haft modules [47].
Difference in Allometric Slopes	Multivariate Analysis of Covariance (MANCOVA) or Vector Correlation	A significant interaction term between size and module identity indicates slopes are different (differential allometry).	Analysis of the spider Donacosa merlini revealed sex-differential shape allometry, indicating different male and female allometric trajectories [48].

### Protocol 4: Correcting for Allometry in Taxonomic Comparisons

Objective: To remove allometric shape variation so that residual, size-independent shape variation can be used for taxonomic discrimination.

• Pooled Within-Group Regression (Recommended):

  • This method calculates the allometric trend within each taxon (e.g., species) and then removes it. This is preferable if taxa have parallel but offset allometric trajectories.
  • In MorphoJ: Use the "Regression" function, selecting the "Pooled within-group" option. This uses the common slope within pre-defined groups to calculate allometric residuals [22].
  • The resulting regression residuals are the allometry-corrected shape variables. These residuals should be used in subsequent taxonomic analyses like Canonical Variate Analysis (CVA) or Discriminant Analysis [50].

• Using Size-Corrected Data in Taxonomy:

  • Perform a Canonical Variate Analysis (CVA) on the allometry-corrected Procrustes residuals.
  • Visually inspect the CVA plot to see if group separation improves after allometry correction compared to the analysis of raw shape data.
  • Calculate Mahalanobis distances between group means based on the corrected data. Significant distances between groups confirm taxonomic distinctiveness after controlling for allometric effects [50].

## 6. Concluding Remarks

Accounting for differential allometry is not merely a statistical exercise but a necessary step for robust taxonomic inference. By systematically testing for modularity and analyzing allometric patterns within and across modules, researchers can dissect complex morphological structures and avoid misattributing allometry-driven shape changes to taxonomic differences. The protocols outlined here, leveraging powerful and accessible software tools, provide a clear roadmap for integrating these considerations into standard geometric morphometric workflows, thereby strengthening the validity of taxonomic conclusions.

In geometric morphometric (GM) studies, particularly those focused on correcting for allometry in taxonomic research, validation techniques are paramount for ensuring the reliability and generalizability of findings. Allometry, which refers to the size-related changes of morphological traits, remains an essential concept for studying evolution and development [3]. The core challenge in this field is to develop statistical models that accurately capture true biological signals rather than overfitting to the specific sample collected. Cross-validation and resampling methods provide robust frameworks for assessing how well these models will perform on new, unseen data, which is crucial for making valid taxonomic distinctions.

The historical development of allometry concepts has led to two main schools of thought: the Gould-Mosimann school, which defines allometry as the covariation of shape with size, and the Huxley-Jolicoeur school, which characterizes allometry as the covariation among morphological features that all contain size information [3] [4]. In practical GM applications, researchers must validate models developed under either framework to ensure their predictive accuracy extends beyond their immediate sample to the broader population or taxonomic group of interest. This is especially critical when sample sizes are limited, as is often the case with fossil records or rare species where the completeness of the fossil record is a major conditioning factor [52].

Core Concepts of Resampling and Cross-Validation

Theoretical Foundations

Resampling techniques in statistics involve repeatedly drawing samples from a training dataset to create multiple simulated datasets. These methods allow researchers to approximate how a statistic would vary across different sampling scenarios, providing insights into the stability and reliability of models without requiring additional new data. In the context of geometric morphometrics, these techniques are particularly valuable for overcoming limitations associated with small sample sizes, which are common in paleontological and taxonomic studies [52].

Cross-validation represents a specific resampling approach used to evaluate how the results of a statistical analysis will generalize to an independent dataset. It is primarily used in settings where the goal is prediction and the researcher wants to estimate how accurately a predictive model will perform in practice. The fundamental principle involves partitioning a sample of data into complementary subsets, performing the analysis on one subset (called the training set), and validating the analysis on the other subset (called the validation set or testing set). This process helps researchers detect overfitting, which occurs when a model describes random error or noise instead of the underlying relationship [53] [54].

Implementation in Morphometric Context

In geometric morphometric studies, these validation techniques take on special significance due to the high-dimensional nature of shape data. When analyzing allometry, researchers often use multivariate regression of shape on size, and they need to verify that the observed patterns reflect true biological relationships rather than sample-specific idiosyncrasies [4]. Cross-validation provides a means to assess whether allometric trajectories identified in one sample would likely be recovered in new samples from the same population or taxonomic group.

The application of these methods is particularly important when using complex machine learning approaches for classification tasks in taxonomic research. Studies have demonstrated that algorithms such as Support Vector Machines (SVM) and Random Forests (RF) can achieve high classification accuracy in morphometric analyses, but their performance must be properly validated using rigorous resampling techniques to ensure taxonomic conclusions are reliable [53]. Without such validation, there is a risk of developing models that appear effective for the specific dataset but fail to generalize to new specimens.

Key Resampling Techniques for Allometry Correction

K-Fold Cross-Validation

K-fold cross-validation is one of the most widely used resampling methods in geometric morphometrics. In this approach, the original sample is randomly partitioned into k equal-sized subsamples. Of the k subsamples, a single subsample is retained as the validation data for testing the model, and the remaining k-1 subsamples are used as training data. The cross-validation process is then repeated k times (the "folds"), with each of the k subsamples used exactly once as the validation data. The k results can then be averaged to produce a single estimation [53].

The advantage of k-fold cross-validation is that all observations are used for both training and validation, and each observation is used for validation exactly once. This approach is particularly useful for evaluating allometric models in taxonomic studies because it provides a more robust estimate of model performance than a single train-test split, especially with moderate sample sizes. For example, in a study on mapping complex gully systems, researchers found that k-fold cross-validation provided more reliable performance estimates for both SVM and Random Forest algorithms compared to other validation approaches [53].

Table 1: Comparison of Cross-Validation Approaches in Geometric Morphometrics

Method	Key Features	Optimal Use Cases	Limitations
K-Fold Cross-Validation	Divides data into k subsets; uses each subset once for validation	Moderate sample sizes; comparing multiple models	Computational intensity increases with k
Leave-One-Out Cross-Validation (LOOCV)	Special case of k-fold where k equals sample size	Very small sample sizes	High variance with large samples; computationally expensive
Stratified K-Fold	Maintains class proportions in each fold	Taxonomic studies with imbalanced classes	More complex implementation
Repeated K-Fold	Repeats k-fold multiple times with different random splits	Obtaining more robust performance estimates	Further increases computational requirements

Bootstrapping Methods

Bootstrapping is another fundamental resampling technique with particular relevance to geometric morphometric studies of allometry. This approach involves repeatedly drawing samples with replacement from the original dataset, typically creating numerous "bootstrap samples" of the same size as the original dataset. Each bootstrap sample is used to fit a model, and the variability across these models provides insight into the stability of parameter estimates [52].

In the context of allometry correction, bootstrapping can be used to assess the reliability of allometric vectors identified through multivariate regression of shape on size. This is particularly important when making taxonomic distinctions based on size-corrected shape data, as it helps researchers distinguish between consistent biological patterns and sampling artifacts. One study comparing classification efficacy found that bootstrapping resampling techniques provided valuable insights into the performance of machine learning algorithms for morphological classification tasks [53].

Bootstrapping has distinct advantages in morphometric applications because it can provide estimates of confidence intervals for parameters such as allometric slopes and regression coefficients, which are crucial for testing hypotheses about differences in allometric patterns among taxonomic groups. Unlike traditional parametric approaches that rely on assumptions about the underlying distribution of shape variables, bootstrapping makes fewer distributional assumptions, making it particularly suitable for the complex multivariate distributions often encountered in geometric morphometric data [52].

Advanced Resampling Approaches

Beyond basic k-fold cross-validation and bootstrapping, several more specialized resampling techniques have been developed to address specific challenges in morphometric data analysis. The .632 bootstrap estimator is one such approach that was developed to correct for the optimistic bias in the apparent error rate of predictive models. This method combines estimates from bootstrap samples with the original training error to provide a more balanced estimate of model performance [52].

Stratified resampling approaches are particularly valuable in taxonomic studies where class imbalances exist. For example, when comparing allometric patterns across multiple species, some species may be represented by many more specimens than others. Stratified approaches ensure that each resampling iteration maintains the original proportion of classes, preventing biased performance estimates that might favor better-represented groups [55].

Monte Carlo cross-validation represents another advanced approach where the data are repeatedly randomly split into training and validation sets, with the split ratio typically not following the k-fold structure. This method is especially useful for assessing the stability of allometric patterns identified through geometric morphometric analyses, as it allows researchers to examine how consistent their findings are across many different partitions of the data [54].

Table 2: Resampling Techniques for Addressing Common Challenges in Allometry Studies

Challenge	Recommended Technique	Rationale	Implementation Considerations
Small sample sizes	Leave-One-Out Cross-Validation	Maximizes training set size in each iteration	High computational cost; can have high variance
Class imbalance in taxonomic groups	Stratified resampling	Preserves class proportions in training/validation sets	Requires careful programming implementation
Assessment of parameter stability	Bootstrapping	Provides confidence intervals for allometric parameters	May require 1000+ iterations for stable estimates
Model selection	Repeated k-fold cross-validation	Reduces variability in performance estimates	Computational intensity scales with repetitions
High-dimensional shape data	Nested cross-validation	Prevents optimistically biased performance estimates	Complex implementation but necessary for reliable results

Experimental Protocols for Validation in Allometry Studies

Protocol 1: K-Fold Cross-Validation for Allometric Regression

Purpose: To validate multivariate regression models of shape on size used for allometry correction in taxonomic studies.

Materials and Software: Morphometric software (e.g., MorphoJ [56], R with geomorph package), landmark data, centroid size values.

Procedure:

• Data Preparation: Perform Generalized Procrustes Analysis (GPA) on raw landmark coordinates to obtain shape variables. Calculate centroid size for each specimen.
• Stratification: If working with multiple taxonomic groups, stratify the data to ensure each fold contains representative specimens from all groups.
• Fold Creation: Randomly partition the dataset into k folds (typically k=5 or k=10). For smaller samples (n<50), consider using leave-one-out cross-validation (k=n).
• Iterative Modeling: For each fold i (where i=1 to k): a. Set aside fold i as the validation set. b. Use the remaining k-1 folds as the training set. c. Perform multivariate regression of shape coordinates on centroid size using the training set. d. Apply the resulting regression model to the validation set to predict shapes. e. Calculate the Procrustes distance between predicted and actual shapes for validation specimens.
• Performance Assessment: Compute the mean Procrustes distance across all folds as an overall measure of prediction error.
• Model Interpretation: If prediction error is acceptably low, perform the multivariate regression on the complete dataset to obtain the final allometric model for size correction.

Troubleshooting Tips:

• High variation in cross-validation error across folds may indicate insufficient sample size or heterogeneous allometric patterns.
• If performance is consistently poor, consider whether the relationship between shape and size may be nonlinear.
• Ensure that landmarks are compatible across all specimens and that missing data have been properly handled [4] [56].

Protocol 2: Bootstrap Validation of Allometric Group Differences

Purpose: To assess the reliability of taxonomic distinctions based on allometrically corrected shape data.

Materials and Software: Shape data after Procrustes superimposition, taxonomic classifications, statistical computing environment with resampling capabilities.

Procedure:

• Initial Analysis: Perform multivariate regression of shape on size using the complete dataset. Save the residuals as size-corrected shape variables.
• Test Statistic Calculation: Compute the observed value of the test statistic of interest (e.g., F-statistic for group differences, canonical variate scores between taxa).
• Bootstrap Resampling: a. For each bootstrap iteration (typically 1000-10000 iterations): i. Sample n specimens with replacement from the original dataset, where n is the original sample size. ii. Perform the same multivariate regression on the bootstrap sample. iii. Calculate the test statistic using the size-corrected shapes from the bootstrap sample. b. Store the test statistic from each bootstrap iteration.
• Confidence Interval Construction: a. Sort the bootstrap test statistics. b. For a 95% confidence interval, use the 2.5th and 97.5th percentiles of the bootstrap distribution.
• Bias Correction: Calculate the bias in the estimate as the difference between the mean of the bootstrap statistics and the original observed statistic.
• Interpretation: If the confidence interval for between-group differences excludes zero, conclude that the taxonomic distinction is statistically reliable after accounting for allometric effects.

Analytical Considerations:

• The bootstrap can be applied to various test statistics relevant to taxonomic studies, including Mahalanobis distances between groups, classification accuracy, or discriminant function coefficients.
• When sample sizes are very small, consider using the bootstrap with caution and supplementing with other validation approaches [52] [53].

Applications in Taxonomic Geometric Morphometrics

Case Study: Validating Allometry Correction in Fossil Taxa

The application of cross-validation techniques is particularly critical when working with fossil specimens, where sample sizes are often limited and preservation may be incomplete. In one landmark study testing the reliability of geometric morphometric methods to identify carnivore agency based on tooth marks, researchers highlighted how previous generalizations of high accuracy were compromised by biased replication and exclusion of non-oval tooth pits [18]. By applying rigorous validation techniques, they demonstrated that earlier claims of high classification accuracy (>90%) were overstated, and that more realistic accuracy rates were in the range of 40% when using geometric morphometric approaches.

This case study illustrates the importance of proper validation in taxonomic applications of geometric morphometrics. When researchers used computer vision approaches with appropriate cross-validation, they achieved substantially higher classification accuracy (79-81%), providing more reliable taxonomic identifications [18]. The implementation of k-fold cross-validation in this context ensured that the performance estimates reflected true predictive accuracy rather than overfitting to specific tooth mark specimens.

Integrating Multiple Validation Approaches

For comprehensive validation in taxonomic studies correcting for allometry, researchers should consider implementing multiple complementary validation techniques. A sequential validation approach might begin with k-fold cross-validation to obtain initial performance estimates, followed by bootstrap resampling to assess the stability of allometric parameters, and concluding with external validation on independently collected data when available.

This multi-faceted approach is particularly valuable when making taxonomic decisions based on allometrically corrected shape data, as it provides multiple lines of evidence regarding the reliability of the patterns observed. Research has shown that relying on a single validation method can sometimes provide misleading results, especially when working with the high-dimensional data structures common in geometric morphometrics [53].

Table 3: Validation Outcomes for Different Morphometric Analysis Types

Analysis Type	Primary Validation Method	Typical Performance Metrics	Acceptance Thresholds
Allometric regression	K-fold cross-validation	Mean Procrustes prediction error	Context-dependent; compare to null models
Taxonomic classification	Stratified cross-validation	Classification accuracy, precision, recall	>70% for complex shapes; >90% for distinctive morphologies
Allometric trajectory comparison	Bootstrap confidence intervals	Overlap of confidence intervals	Non-overlapping 95% CIs suggest significant differences
Shape prediction	Leave-one-out cross-validation	Procrustes distance between predicted and observed	Lower than biological variation within groups

The Scientist's Toolkit: Research Reagent Solutions

Table 4: Essential Tools for Validation in Geometric Morphometric Studies

Tool/Software	Primary Function	Specific Application in Allometry Studies
MorphoJ [56]	General morphometric analysis	Regression of shape on size; calculation of residuals for size correction
R (geomorph package)	Comprehensive morphometric analysis	Procrustes ANOVA; advanced allometric analyses; integration with resampling methods
Custom R/Python Scripts	Implementation of resampling	Creating custom cross-validation routines tailored to specific research designs
Peripheral Software	Supplementary analyses	Visualization of allometric trajectories; shape changes associated with size
GIS Applications [53]	Spatial analysis	Integration of ecological variables that may interact with allometric patterns

Validation through cross-validation and resampling methods represents an essential component of rigorous geometric morphometric studies, particularly those focused on correcting for allometry in taxonomic research. These techniques provide the necessary framework for distinguishing between biological patterns that generalize to new samples and statistical artifacts that reflect only the specific sample collected. As geometric morphometrics continues to evolve with increasingly complex analytical approaches, the importance of proper validation will only grow.

The protocols and applications outlined in this article provide researchers with practical guidance for implementing these validation techniques in their own work. By adopting these methods as standard practice, the field can move toward more reliable taxonomic distinctions based on allometrically corrected shape data, leading to more robust conclusions about evolutionary patterns and processes. Future methodological developments will likely focus on more efficient implementation of these validation approaches for increasingly large and complex morphometric datasets.

Optimizing Landmark Schemes for Allometry Studies

In taxonomic geometric morphometric (GM) studies, allometry—the pattern of covariation between an organism's shape and its size—is a fundamental factor that must be characterized and corrected to isolate true taxonomic signal from size-related variation [3] [4]. The confounded nature of these effects means that failing to properly account for allometry can lead to spurious taxonomic conclusions. The foundational step upon which all subsequent allometric analyses depend is the design and application of a robust landmark scheme. This document provides detailed application notes and protocols for optimizing these landmark schemes, framed within the broader objective of correcting for allometry in taxonomic studies. We synthesize current methodologies from two primary schools of allometric thought: the Gould-Mosimann school, which defines allometry as the covariation of shape with size, and the Huxley-Jolicoeur school, which characterizes allometry as the covariation among morphological features that all contain size information [3] [4]. The protocols herein are designed for researchers, scientists, and professionals engaged in the morphometric analysis of taxonomic groups.

Core Concepts and Definitions

• Allometry: The pattern of change in morphological traits (e.g., shape) correlated with changes in size. This can be studied at ontogenetic, static, or evolutionary levels [3].
• Geometric Morphometrics (GM): A set of methodologies based on the statistical analysis of Cartesian landmark coordinates that preserve the geometric relationships among structures throughout the analysis [3].
• Landmark: A discrete point that corresponds between organisms and can be precisely defined based on anatomical, morphological, or mathematical criteria.
• Procrustes Superimposition: A procedure that removes the effects of variation in position, orientation, and scale from raw landmark coordinates, resulting in Procrustes shape coordinates [57] [4].
• Centroid Size (CS): A measure of size, calculated as the square root of the sum of squared distances of all landmarks from their centroid [3] [57]. It is the standard size metric in GM.
• Form: The totality of morphological information (size and shape together) [57].
• Conformation Space (Size-and-Shape Space): The space of landmark configurations after standardization for position and orientation, but not for size [3] [4].
• Boas Coordinates: Landmark coordinates that have been standardized for position and orientation, but not scaled to unit Centroid Size, thus preserving form information [57] [4].

Methodological Frameworks for Allometry Analysis

The choice of analytical framework is critical and depends on the research question and the concept of allometry being applied. The following table summarizes the two main schools of thought and their corresponding implementation in geometric morphometrics.

Table 1: Key Frameworks for Analyzing Allometry in Geometric Morphometrics

Conceptual School	Core Definition of Allometry	Morphometric Implementation	Primary Analytical Space
Gould-Mosimann School	Covariation between shape and size [3]	Multivariate regression of shape coordinates on Centroid Size (or log CS) [4]	Kendall’s Shape Space / Shape Tangent Space
Huxley-Jolicoeur School	Covariation among morphological traits, all containing size information [3]	First principal component (PC1) of coordinates in Conformation Space or of Boas Coordinates [57] [4]	Conformation Space (Size-and-Shape Space)

A recent performance comparison using computer simulations has shown that while all methods are logically consistent, the multivariate regression of shape on size generally performs well in the presence of isotropic residual variation [4]. Furthermore, the PC1 of Conformation Space and Boas Coordinates were found to be very similar and close to the true simulated allometric vectors under a variety of conditions [4]. For studies focused specifically on growth allometry, analyses using Boas coordinates can provide a more direct and biologically interpretable picture of the growth pattern, often revealing a "centric allometry" where landmarks displace radially outward from the centroid at varying rates [57].

Application Notes: Optimizing Landmark Schemes

The validity of any allometric correction is contingent on the quality of the initial landmark data. An optimal landmark scheme must capture the morphology relevant to both taxonomy and allometry.

Landmark Types and Selection Criteria

Landmarks are traditionally classified as:

• Type I: Anatomically discrete points (e.g., foramina, suture intersections).
• Type II: Points of maximum curvature or other local geometric extrema.
• Type III: Points defined by a geometric constraint relative to other landmarks (e.g., the point furthest from another landmark).

For allometry-corrected taxonomic studies, the scheme should:

• Prioritize Homology: Every landmark must have a clear biological correspondence across all specimens in the analysis. Type I landmarks are most reliable for this.
• Capture the Form Comprehensively: The landmark set must adequately represent the entire anatomical structure of interest to avoid introducing bias. This often requires a combination of all landmark types.
• Ensure Repeatability: Landmark definitions must be unambiguous to ensure high intra- and inter-observer repeatability.
• Cover Functional and Growth Centers: Include landmarks in regions known to be important for the organism's function and that exhibit differential growth, as these are critical for capturing allometric patterns.

Protocol 1: Designing and Collecting a Landmark Scheme for Allometry Studies

This protocol outlines the steps for establishing a robust landmark scheme from initial design to data collection.

Table 2: Essential Research Reagent Solutions for Geometric Morphometrics

Item/Category	Specific Examples	Function in Research
Imaging Equipment	High-resolution scanner, Micro-CT scanner, Digital camera with macro lens	Captures high-fidelity 2D or 3D digital images of specimens for landmark digitization.
Digitization Software	tpsDig2, MorphoJ, IMP (Integrated Morphometrics Package)	Software used to place and record the Cartesian coordinates of landmarks on digital images.
GM Analysis Software	R (geomorph, Morpho), MorphoJ, PAST	Performs Procrustes superimposition, statistical analysis, and visualization of shape and allometry.
Size Variable	Centroid Size	A standardized, geometrically derived measure of size calculated from all landmarks, used as the independent variable in allometric regressions [3].
Statistical Model	Multivariate Regression (Shape ~ Size)	The primary statistical model for quantifying the relationship between shape variation (dependent variable) and size (independent variable) [4].

Workflow:

• Define Taxonomic and Allometric Objectives: Clearly state the taxonomic groups under investigation and the specific allometric hypotheses to be tested (e.g., "Do species X and Y share a common allometric trajectory?").
• Conduct Preliminary Literature Review: Identify established landmark schemes for the studied organism or closely related taxa. Note landmarks used in previous allometric studies.
• Pilot Landmarking: On a small, representative subset of specimens (covering the full size and taxonomic range), draft a preliminary landmark scheme.
• Assess Repeatability: Have multiple trained observers (or the same observer on different days) digitize the pilot scheme. Quantify measurement error using Procrustes ANOVA [4].
• Refine the Scheme: Based on repeatability analysis, remove ambiguous landmarks and refine definitions. Ensure the final scheme adequately covers the morphology.
• Formal Data Collection: Digitize the finalized landmark scheme across the entire sample dataset.

The following workflow diagram summarizes the logical structure for designing and implementing an optimized landmark scheme.

Protocols for Allometric Analysis and Correction

Once a robust landmark dataset is acquired, the following protocols can be applied to analyze and correct for allometric effects.

Protocol 2: Estimating Allometric Vectors via Regression and PCA

This protocol details the steps for characterizing the allometric pattern using the primary methods identified in the literature.

Workflow:

• Data Preprocessing: Perform a Generalized Procrustes Analysis (GPA) on the raw landmark data to obtain Procrustes shape coordinates for analyses in shape space [4].
• Calculate Size: Compute Centroid Size (CS) for each specimen from the raw landmark coordinates [3].
• Gould-Mosimann Approach (Regression): a. Perform a multivariate regression of the Procrustes shape coordinates on Centroid Size (or log CS). b. The regression vector (the set of partial warp scores from the regression) represents the allometric vector [4]. c. The predicted values from this regression represent the shape component correlated with size. The residuals are the size-corrected shapes.
• Huxley-Jolicoeur Approach (PCA in Form Space): a. Generate Boas Coordinates by performing a Procrustes superimposition that standardizes for position and orientation but omits the scaling step [57] [4]. b. Perform a Principal Component Analysis (PCA) on the covariance matrix of the Boas Coordinates. c. The first principal component (PC1) often represents the primary allometric vector [57] [4].

Table 3: Comparison of Allometry Estimation Methods Based on Simulation Studies [4]

Method	Underlying Concept	Performance under Isotropic Noise	Performance under Anisotropic Noise	Key Advantage
Regression of Shape on Size	Gould-Mosimann	Excellent performance [4]	Good performance [4]	Directly tests and models the effect of size on shape.
PC1 of Shape	Gould-Mosimann	Inferior to regression [4]	Can be misleading if PC1 is not related to size	Not recommended if the goal is specifically to study allometry.
PC1 of Conformation/Boas Coordinates	Huxley-Jolicoeur	Very good, close to true vector [4]	Very good, close to true vector [4]	Intuitively captures the main axis of form variation, which is often allometry.

Protocol 3: Correcting for Allometry in Taxonomic Comparisons

After estimating the allometric vector, this protocol outlines the procedure for removing its effect to examine size-free taxonomic differences.

Workflow:

• Choose a Correction Method:
  • Regression Residuals: Use the residuals from the multivariate regression of shape on size (Protocol 2, Step 3). These residuals are, by definition, uncorrelated with size and represent the non-allometric component of shape variation [4].
  • Vector Subtraction: Project the data orthogonally to the estimated allometric vector. This can be done by subtracting the component of shape that lies along the allometric vector (e.g., the regression vector or Boas PC1 vector) from each specimen's shape coordinates.
• Validate the Correction: Check that the corrected shape data (residuals) no longer have a significant correlation with Centroid Size. This can be done via a follow-up multivariate regression of the residuals on size, which should be non-significant.
• Analyze Taxonomy on Corrected Data: Perform subsequent taxonomic analyses (e.g., MANOVA, CVA, clustering) on the allometry-corrected shape data to interpret pure taxonomic signal.

The following diagram illustrates the decision pathway for selecting and applying an allometry correction method.

Optimizing landmark schemes is the critical first step in robust allometric analysis for taxonomic morphometrics. By employing a landmarking protocol that emphasizes homology, coverage, and repeatability, researchers ensure their data is fit for purpose. Subsequent application of either the Gould-Mosimann (regression-based) or Huxley-Jolicoeur (PCA-based) frameworks allows for the precise estimation and correction of allometric effects. The choice between these methods depends on the specific research question, but simulation studies indicate that the regression of shape on size and the PC1 of Boas coordinates are both highly effective [4]. Correcting for allometry using these protocols allows researchers to isolate and analyze the true taxonomic signal, leading to more accurate and biologically meaningful conclusions in systematic and evolutionary studies.

Evaluating Correction Efficacy and Comparative Frameworks in Taxonomic Research

In taxonomic geometric morphometric studies, allometry—the pattern of covariation between organismal shape and size—presents both a challenge and an opportunity. When unaccounted for, allometric variation can confound taxonomic signals, leading to inaccurate classifications and misinterpretations of evolutionary patterns. Consequently, validating allometry correction is not merely a statistical exercise but a fundamental requirement for ensuring the biological validity of morphometric analyses. This protocol paper establishes a comprehensive framework for verifying that allometry correction methods successfully isolate true taxonomic signals from size-related shape variation within the context of geometric morphometrics.

The need for rigorous validation protocols stems from the complex nature of morphological data. As highlighted by Viscosi and Cardini, geometric morphometrics preserves the geometric information of shape differences, increases statistical power, and enables visualization of patterns, but its effectiveness depends entirely on appropriate application and validation of methods such as allometry correction [58]. Within this framework, we present standardized approaches for assessing the performance of allometry correction techniques, ensuring that researchers can confidently discriminate taxonomic groups based on shape characteristics independent of size.

Theoretical Foundations of Allometry

Concepts and Definitions

Allometry refers to the size-related changes of morphological traits and remains an essential concept for the study of evolution and development [3]. In geometric morphometrics, two primary schools of thought conceptualize allometry differently:

• Gould-Mosimann School: Defines allometry as the covariation of shape with size. This perspective is implemented through multivariate regression of shape variables on a measure of size [3].
• Huxley-Jolicoeur School: Characterizes allometry as the covariation among morphological features that all contain size information, typically represented by the first principal component as a line of best fit to the data points [3].

These conceptual differences inform distinct analytical approaches, yet both frameworks aim to understand how organismal form changes with size—a fundamental relationship in taxonomic studies.

Levels of Allometric Variation

Biological studies recognize several distinct levels of allometric variation, each with implications for taxonomic research:

• Ontogenetic allometry: Shape changes associated with growth and development
• Static allometry: Covariation among individuals at the same developmental stage
• Evolutionary allometry: Divergence in allometric patterns across taxa
• Fluctuating asymmetry allometry: Size-related asymmetry in bilateral organisms [3]

Taxonomic studies must carefully consider which level of allometry is relevant to their specific research questions, as confounding these levels can lead to erroneous interpretations.

Statistical Framework for Allometry Assessment

Core Statistical Methods

Table 1: Statistical Methods for Allometry Assessment

Method	Implementation	Application Context	Key Outputs
Multivariate Regression	procD.lm in geomorph R package	Testing shape~size relationship	Regression coefficients, p-values, effect sizes
Common Allometric Component (CAC)	two.b.pls in geomorph	Extracting primary allometric axis	CAC scores, effect size of covariation
Regression Score (RegScore)	plotAllometry in geomorph	Visualizing allometric trends	Regression scores, shape predictions
Prediction Line (PredLine)	plotAllometry in geomorph	Group allometry comparisons	Fitted values, principal components
Quantile Regression	quantreg R package	Multi-scaling allometry assessment	Scaling exponents across percentiles

Advanced Allometry Concepts

Recent research has challenged the traditional assumption of "uni-scaling" allometry, where a single power-law equation describes the relationship between traits. The emerging concept of "multi-scaling allometry" recognizes that scaling exponents may vary across different percentiles of trait distributions [43]. This approach, implemented through quantile regression, reveals that individuals in different segments of the distribution may follow distinct growth strategies—a crucial consideration for taxonomic studies where allometric patterns might differ among closely related species.

Furthermore, the statistical implementation of allometry models requires careful consideration of error structures. Traditional approaches often assume normally distributed errors, but real morphological data may exhibit heteroscedasticity or complex variance patterns that require alternative error distributions, such as logistic or normal mixture distributions, to achieve reliable fits [59].

Experimental Protocols for Validation

Workflow for Allometry Correction Validation

Allometry Correction Validation Workflow

Protocol 1: Initial Allometry Assessment

Purpose: To determine whether significant allometry exists in the dataset and quantify its strength.

Procedure:

• Perform Procrustes Superimposition: Align landmark configurations to remove non-shape variation [58]
• Calculate Centroid Size: Compute the square root of the sum of squared distances of all landmarks from their centroid [3]
• Multivariate Regression: Execute shape ~ size regression using ProcD.lm

• Effect Size Calculation: Compute the multivariate coefficient of determination (R²) to quantify allometric strength
• Significance Testing: Use permutation procedures (recommended: 999-9999 iterations) to assess statistical significance

Validation Metrics:

• Effect size (R²) with confidence intervals
• Permutation p-value
• Visualization of allometric trend via PredLine or RegScore methods [60]

Protocol 2: Allometry Correction Methods

Purpose: To remove allometric effects from shape data while preserving taxonomic signals.

Procedure:

• Model Selection: Choose appropriate allometry correction model based on research question:
  • Common Allometry: shape ~ size (assumes same allometry across groups)
  • Group-Specific Allometry: shape ~ size * group (allows different allometries)

• Residuals Extraction: Obtain size-adjusted shapes as regression residuals

• Alternative Approach - CAC Adjustment: Use the Common Allometric Component method when a unified allometric axis is appropriate [3]

Validation Check: Confirm that corrected shapes no longer correlate significantly with size.

Protocol 3: Validation of Correction Effectiveness

Purpose: To verify that allometry correction successfully removed size effects while preserving taxonomic information.

Procedure:

• Post-Correction Allometry Test:
  • Re-run multivariate regression of corrected shapes on size
  • Confirm non-significant relationship (p > 0.05 after correction)

• Taxonomic Signal Preservation Assessment:

  • Perform discriminant analysis on original and corrected data
  • Compare classification success rates
  • Evaluate group separation in morphospace (PCA)

• Effect Size Comparison:

  • Calculate partial η² or R² for size effect before and after correction
  • Document reduction in size-related variance

• Multi-Scaling Assessment (Advanced):

  • Implement quantile regression to test for residual allometry across distribution
  • Compare scaling exponents at different percentiles [43]

Table 2: Validation Metrics and Interpretation

Validation Metric	Target Outcome	Interpretation
Size-Shape Correlation (post-correction)	p > 0.05, R² < 0.01	Successful allometry removal
Group Discrimination Accuracy	Unchanged or improved	Taxonomic signal preserved
Effect Size Reduction	>80% reduction in size effect	Effective correction
Morphospace Structure	Maintained relative positions	Biological meaningfulness preserved

The Scientist's Toolkit

Essential Software and Reagents

Table 3: Research Reagent Solutions for Allometry Correction

Tool/Reagent	Function/Purpose	Implementation Notes
geomorph R package	Comprehensive GM analysis	Primary platform for ProcD.lm, GPA, allometry functions [37] [60]
Rphylipar	Phylogenetic correction	Integration of phylogenetic independent contrasts
tpsDig2	Landmark digitization	Precise landmark coordinate collection [58]
MorphoJ	User-friendly GM analysis	Alternative for beginners, GUI-based
Quantreg R package	Multi-scaling analysis	Quantile regression for complex allometry [43]
High-Resolution Scanner	Image acquisition	Minimum 300 dpi for landmark precision [58]
Specimen Stabilization	Standardized imaging	Pressing, drying for consistent leaf morphology [58]

Advanced Applications and Considerations

Complex Experimental Designs

Taxonomic studies often involve hierarchical data structures that require specialized analytical approaches:

• Nested Designs: For data structured hierarchically (e.g., leaves within trees, trees within populations)
• Mixed Models: Incorporating random effects for related specimens or repeated measures
• Phylogenetic Correction: Adjusting for evolutionary relationships among taxa

For nested designs, the model structure should account for these hierarchies:

Visualization Techniques

Effective visualization is crucial for interpreting allometry correction results:

• Deformation Grids: Visualize shape changes along allometric vector
• PredLine Plots: Compare allometric trajectories among groups
• Regression Score Plots: Display distribution of specimens along allometric axis
• PCA with Size Overlay: Visualize size-shape relationship in morphospace

Validating allometry correction in taxonomic geometric morphometric studies requires a systematic, multi-faceted approach that combines statistical rigor with biological interpretation. The protocols presented here provide a comprehensive framework for assessing and verifying the effectiveness of allometry correction, ensuring that resulting taxonomic inferences are based on biologically meaningful shape differences rather than size-associated variation.

The field continues to evolve with emerging concepts like multi-scaling allometry challenging traditional uniform scaling models [43], and improved error-structure modeling enhancing the reliability of allometric fits [59]. By adopting these validation protocols, researchers can strengthen the foundation of their taxonomic conclusions and contribute to more accurate understanding of morphological evolution.

As geometric morphometrics continues to integrate with genomic and developmental approaches, robust allometry correction validation will remain essential for disentangling the complex interplay between size, shape, and taxonomy in morphological studies.

Allometry, the study of how organismal shape changes with size, is a foundational concept in evolutionary biology and systematics. In taxonomic studies using geometric morphometrics (GMM), accurately identifying and correcting for allometric variation is crucial for distinguishing true taxonomic signals from size-dependent shape changes [3]. This analysis provides a structured framework for comparing allometric patterns across related taxa, enabling researchers to isolate evolutionary divergence from covariation related to body size. The protocols outlined here are designed for integration within a broader thesis on correcting for allometry in taxonomic GMM studies, addressing the needs of researchers requiring robust methods for analyzing morphological patterns in evolutionary biology and systematics.

The two primary schools of thought in allometric analysis—the Gould-Mosimann school (focusing on covariation between size and shape) and the Huxley-Jolicoeur school (focusing on covariation among morphological features containing size information)—provide complementary approaches for understanding these patterns [3]. This protocol integrates both frameworks to offer a comprehensive toolkit for taxonomic comparisons.

Theoretical Framework and Key Concepts

Schools of Thought in Allometric Analysis

Table 1: Key Concepts in Allometric Analysis

Concept	Definition	Taxonomic Relevance
Allometry	Size-related changes in morphological traits	Fundamental for distinguishing taxonomic signals from size variation
Gould-Mosimann School	Defines allometry as covariation of shape with size	Uses shape spaces; analyzes allometry via regression of shape on size
Huxley-Jolicoeur School	Defines allometry as covariation among morphological features containing size information	Uses conformation space; identifies allometric trajectories via PCA
Static Allometry	Allometric patterns within a single ontogenetic stage (typically adults)	Essential for comparing adult morphology across taxa
Ontogenetic Allometry	Shape changes correlated with size throughout growth	Important for understanding developmental differences between taxa
Evolutionary Allometry	Allometric patterns across evolutionary lineages	Crucial for studying macroevolutionary patterns

Biological Hierarchy of Allometric Patterns

Allometry operates at multiple biological levels, each with distinct implications for taxonomic studies [3]:

• Ontogenetic Allometry: Shape changes throughout growth within a species
• Static Allometry: Covariation between size and shape within a single developmental stage
• Evolutionary Allometry: Divergence in allometric patterns across related taxa over evolutionary time

Each level requires specific sampling strategies and analytical approaches. Confounding these levels can lead to misinterpretation of taxonomic patterns, making careful study design essential.

Experimental Protocols

Specimen Selection and Data Collection

Protocol 3.1.1: Specimen Selection for Taxonomic Allometry Studies

• Sample Size Determination:

  • Include sufficient specimens per taxon to ensure statistical power (minimum 15-20 specimens per group)
  • Balance sample sizes across taxa to avoid bias in multivariate analyses
  • For ontogenetic allometry, include specimens representing entire growth series

• Taxonomic Coverage:

  • Select multiple closely-related taxa for comparative analysis
  • Include outgroups to establish ancestral allometric patterns
  • Verify taxonomic identifications using independent evidence (molecular, ecological)

• Size Range Considerations:

  • Ensure adequate size variation within each taxon to detect allometric patterns
  • For static allometry, focus on adults with natural size variation
  • Document potential sources of size variation (environmental, sexual dimorphism)

Protocol 3.1.2: Landmarking and Data Acquisition

• Landmark Configuration:

  • Select biologically homologous landmarks covering the entire structure
  • Include fixed anatomical landmarks and semi-landmarks for curves
  • Use consistent landmark protocols across all specimens

• Image Standardization:

  • Maintain consistent orientation, scale, and resolution across all images
  • Use standardized imaging equipment and settings
  • Include scale references in all images

• Data Quality Control:

  • Assess measurement error through repeated digitization [2]
  • Identify and address outliers in landmark configurations
  • Test for the impact of missing data and apply appropriate estimation methods if needed

Preliminary Analyses

Protocol 3.2.1: Assessment of Measurement Error

• Replication Design:

  • Digitize a subset of specimens (minimum 10-15%) multiple times
  • Randomize specimen order between digitization sessions
  • Consider both intra- and inter-observer error if multiple researchers involved

• Error Quantification:

  • Perform Procrustes ANOVA to partition variance components
  • Compare variance due to measurement error versus biological variation
  • Establish acceptability thresholds for measurement precision

• Correction Procedures:

  • If error is substantial, apply correction factors or averaging
  • Exclude landmarks with consistently high error rates
  • Document all error assessment procedures for methodological transparency

Protocol 3.2.2: Outlier Detection and Data Cleaning

• Morphological Outlier Identification:

  • Calculate Procrustes distances from mean shape for each specimen
  • Visualize distribution of distances to identify extreme values
  • Examine potential taxonomic or biological explanations for outliers

• Multivariate Assessment:

  • Use Mahalanobis distances in shape space
  • Apply principal component analysis to detect outliers in multivariate space
  • Consider jackknife or bootstrap approaches for small samples

• Outlier Handling:

  • Investigate biological versus methodological causes for outliers
  • Document all excluded specimens with justification
  • Perform sensitivity analyses with and without outliers

Allometric Vector Estimation Methods

Protocol 3.3.1: Gould-Mosimann Approaches (Size-Shape Covariation)

• Multivariate Regression of Shape on Size:

  • Perform Generalized Procrustes Analysis (GPA) to align specimens
  • Calculate centroid size as a measure of overall size
  • Regress Procrustes coordinates on centroid size using multivariate regression
  • Extract regression coefficients as the allometric vector

• Principal Component Analysis in Shape Space:

  • Conduct PCA on covariance matrix of Procrustes coordinates
  • Correlate PC1 scores with centroid size
  • If strong correlation exists, use PC1 as allometric vector

Protocol 3.3.2: Huxley-Jolicoeur Approaches (Form Space Analyses)

• PCA in Conformation Space:

  • Align specimens using Procrustes superimposition without scaling
  • Perform PCA on the resulting size-and-shape coordinates
  • Use PC1 as the allometric vector representing the primary axis of form variation

• Boas Coordinates Analysis:

  • Calculate Boas coordinates as alternative size-preserved shape variables
  • Perform PCA on Boas coordinates
  • Use PC1 as allometric vector for comparison with other methods

Statistical Comparison of Allometric Patterns

Protocol 3.4.1: Testing for Common Allometric Patterns

• Multivariate Analysis of Covariance (MANCOVA):

  • Use MANCOVA with shape as dependent variables, size as covariate, and taxon as factor
  • Test for homogeneity of slopes (size × taxon interaction)
  • Significant interaction indicates divergent allometric patterns among taxa

• Vector Correlation Analysis:

  • Calculate allometric vectors for each taxon separately
  • Compute pairwise correlations between taxon-specific vectors
  • Test statistical significance of vector correlations using permutation tests

• Angle Between Vectors:

  • Calculate angles between allometric vectors of different taxa
  • Compare observed angles to null distribution from permutation testing
  • Small angles indicate conserved allometric patterns across taxa

Protocol 3.4.2: Visualization of Allometric Patterns

• Allometric Trajectory Plotting:

  • Plot specimens in shape space along allometric vectors
  • Use different symbols/colors for different taxa
  • Include confidence ellipses for trajectory estimation

• Thin-Plate Spline Visualization:

  • Visualize shape changes along allometric vectors using deformation grids
  • Compare deformation patterns across taxa
  • Highlight regions of divergent allometric development

Visualization and Analytical Workflows

Allometric Analysis Decision Framework

Figure 1: Allometric Analysis Decision Workflow. This diagram outlines the key decision points in comparative allometric analysis, showing the relationship between different analytical approaches.

Allometric Vector Comparison Methods

Figure 2: Allometric Vector Comparison Methodology. This workflow shows the process for comparing allometric patterns across different taxonomic groups.

Research Reagent Solutions

Table 2: Essential Materials for Taxonomic Allometric Studies

Category	Specific Tools/Software	Function in Analysis
Imaging Equipment	High-resolution digital camera with macro lens, standardized mounting system	Acquisition of consistent 2D images for landmark digitization
Landmark Digitization	TPSDig2, MorphoJ, IMP series	Precise placement of landmarks and semi-landmarks on digital images
Shape Analysis	MorphoJ, geomorph R package, EVAN Toolbox	Generalized Procrustes Analysis, shape variable extraction, and visualization
Statistical Analysis	R with geomorph, shapes, Momocs packages; PAST	Multivariate statistics, permutation tests, allometric vector calculations
Visualization	MorphoJ, R ggplot2, Thin-Plate Spline software	Visualization of shape differences, allometric trajectories, and deformation patterns
Data Management	Custom spreadsheets, R data frames, TPS file series	Organization of landmark data, metadata, and analysis results

Data Presentation and Analysis

Performance Comparison of Allometric Methods

Table 3: Comparison of Allometric Vector Estimation Methods [4]

Method	Theoretical Framework	Strengths	Limitations	Recommended Use
Multivariate Regression of Shape on Size	Gould-Mosimann	Direct test of size-shape relationship; clear interpretation; good statistical properties	Assumes linear relationship; sensitive to size range and outliers	Primary method for testing allometric hypotheses
PC1 in Shape Space	Gould-Mosimann	Captures major axis of shape variation; may correlate with size	PC1 may represent non-allometric variation; requires correlation with size	Supplemental analysis when PC1 strongly correlates with size
PC1 in Conformation Space	Huxley-Jolicoeur	Uses form space (size+shape); no artificial separation of size and shape	Global structure of space differs from shape space; interpretation less straightforward	Alternative approach particularly for growth series
PC1 of Boas Coordinates	Huxley-Jolicoeur	Similar to conformation space; handles variation in localized geometry	Less familiar to most researchers; limited software implementation	Specialized applications requiring localized geometry

Taxonomic Application Case Studies

Table 4: Exemplary Taxonomic Applications of Allometric Analysis

Taxonomic Group	Study Focus	Methods Applied	Key Findings	Reference
Marmots (Marmota spp.)	Mandibular shape evolution	Procrustes GMM, allometric vector comparison	Interspecific morphological variation patterns in mammalian sociobiology	[2]
Leaf-footed bugs (Acanthocephala spp.)	Species delimitation using pronotum shape	PCA, discriminant analysis, allometry assessment	Pronotum shape reliably distinguishes species despite morphological overlaps	[50]
Triatoma pallidipennis haplogroups	Cryptic species differentiation	Landmark and semilandmark analysis, ecological niche modeling	Head shape provided higher taxonomic value than pronotum for differentiation	[61]
Rockfish (Sebastes spp.)	Ontogenetic allometry in body shape	Multivariate regression, trajectory analysis	Distinct allometric patterns related to ecological specialization	[4]

Implementation Protocols

Complete Analytical Pipeline

Protocol 7.1.1: Integrated Allometric Analysis for Taxonomic Comparisons

• Data Acquisition and Preparation (2-4 weeks)

  • Specimen imaging and landmarking following standardized protocols
  • Data quality checks (measurement error, outliers)
  • Format data for multivariate analyses (TPS or equivalent format)

• Preliminary Shape Analysis (1-2 weeks)

  • Generalized Procrustes Analysis to align specimens
  • Principal Component Analysis for exploratory shape variation assessment
  • Calculation of centroid size and other size metrics

• Allometric Vector Estimation (1-2 weeks)

  • Apply multiple methods (regression, PCA in different spaces)
  • Compare results across methods for consistency
  • Select appropriate allometric vector(s) for further analysis

• Taxonomic Comparisons (2-3 weeks)

  • Test for common allometric patterns using MANCOVA
  • Calculate vector correlations and angles between taxa
  • Perform statistical tests (permutation approaches) for significance

• Size Correction (if required) (1 week)

  • Apply appropriate size correction methods based on research questions
  • Validate correction effectiveness through diagnostic plots
  • Document all correction procedures for reproducibility

• Interpretation and Visualization (1-2 weeks)

  • Create allometric trajectory plots
  • Generate deformation grids for shape changes
  • Relate morphological patterns to taxonomic hypotheses

Troubleshooting Common Issues

Issue: Confounding Allometric Levels

• Problem: Ontogenetic, static, and evolutionary allometry confounded in analysis
• Solution: Implement stratified sampling design; analyze levels separately before comparison

Issue: Method-Dependent Results

• Problem: Different allometric methods yield conflicting patterns
• Solution: Report results from multiple methods; evaluate biological plausibility; consider methodological assumptions

Issue: Small Sample Sizes

• Problem: Insufficient statistical power for reliable allometric vector estimation
• Solution: Use resampling methods; focus on effect sizes with confidence intervals; acknowledge limitations

Issue: Missing Data in Landmark Configurations

• Problem: Incomplete specimens compromise analyses
• Solution: Apply appropriate estimation methods; conduct sensitivity analyses; document missing data patterns

Comparative analysis of allometric patterns across related taxa provides powerful insights into evolutionary processes underlying morphological diversification. The integrated protocols presented here, drawing from both major schools of allometric thought, offer a comprehensive framework for taxonomic studies using geometric morphometrics. By applying these standardized approaches, researchers can robustly distinguish true taxonomic signals from size-dependent variation, advancing our understanding of evolutionary relationships and morphological evolution.

The decision framework and troubleshooting guidance address common challenges in allometric analysis, while the performance comparisons inform method selection based on specific research questions. As geometric morphometrics continues to evolve, these protocols provide a foundation for rigorous taxonomic comparisons of allometric patterns across diverse organismal groups.

In modern taxonomic research, the integration of multiple lines of evidence has become essential for robust species delimitation and understanding evolutionary relationships. This approach, often termed "integrative taxonomy," is particularly valuable when studying groups with subtle morphological differences, high phenotypic plasticity, or complex evolutionary histories [62]. Geometric morphometrics (GM) provides powerful tools for quantifying shape variation, but these analyses are frequently confounded by allometry—the pattern of covariation between shape and size [3] [4]. Allometric variation can obscure phylogenetic signal or mimic patterns of divergent evolution if not properly accounted for in analyses [63]. This protocol details comprehensive methodologies for integrating morphometric, molecular, and ecological data within a framework that explicitly addresses allometric corrections in taxonomic studies.

Theoretical Background: Allometry in Geometric Morphometrics

Allometry remains an essential concept for evolutionary and developmental studies, referring to size-related changes in morphological traits [3]. In geometric morphometrics, two primary conceptual frameworks guide allometric studies:

• Gould-Mosimann School: Defines allometry as the covariation between shape and size, typically implemented through multivariate regression of shape variables on a size measure [3] [4].
• Huxley-Jolicoeur School: Characterizes allometry as covariation among morphological features that all contain size information, implemented by analyzing allometric trajectories along the first principal component in form space [3] [4].

Each framework offers distinct advantages. The Gould-Mosimann approach provides a direct measure of shape-size covariation, while the Huxley-Jolicoeur method can reveal more complex morphological integration patterns [4]. For taxonomic studies, failing to account for allometric variation can lead to misinterpretation of shape differences that are actually consequences of size variation rather than evolutionary divergence.

Experimental Design and Workflow

The following integrated workflow provides a systematic approach for taxonomic studies that incorporate multiple data types while controlling for allometric effects:

Figure 1: Integrated Workflow for Taxonomic Studies Incorporating Multiple Data Types

Sample Design Considerations

• Sample Sizes: Ensure adequate sampling (typically ≥15-20 specimens per group) to achieve statistical power for morphometric analyses [2].
• Size Range Representation: Include specimens representing the full biological size range for each putative taxon to properly characterize allometric relationships.
• Reference Collections: Utilize museum specimens with verified identifications when possible, and ensure vouchering of new specimens for future reference.

Morphometric Data Acquisition and Processing Protocols

Landmark Digitization and Data Collection

• Image Acquisition: Capture standardized, high-resolution images or 3D scans of specimens. Ensure consistent orientation and scale across all samples [50].
• Landmark Configuration: Digitize two-dimensional or three-dimensional landmarks using software such as TPSDig2 or MorphoJ [50]. Include biologically homologous landmarks that capture key morphological structures.
• Landmark Types:
  • Type I Landmarks: Discrete anatomical loci (e.g., suture intersections)
  • Type II Landmarks: Maxima of curvature or other local geometry features
  • Type III Landmarks: Extremal points that may be less phylogenetically conserved

Shape Processing and Allometry Assessment

• Generalized Procrustes Analysis (GPA): Superimpose landmark configurations to remove variation due to position, orientation, and scale [50].
• Centroid Size Calculation: Compute centroid size as the square root of the sum of squared distances of all landmarks from their centroid [3].
• Allometry Detection: Perform multivariate regression of Procrustes coordinates on centroid size to test for significant allometric effects [4] [50].
• Allometric Pattern Visualization: Create deformation plots or vector displacement diagrams to visualize shape changes associated with size variation.

Table 1: Methods for Allometric Analysis in Geometric Morphometrics

Method	Theoretical Framework	Implementation	Best Use Cases
Multivariate Regression of Shape on Size	Gould-Mosimann	Regression of Procrustes coordinates on centroid size	Testing specific allometry hypotheses; size correction
PC1 of Shape Space	Gould-Mosimann	Principal component analysis of shape variables	Exploratory analysis of major shape variation patterns
PC1 of Conformation Space	Huxley-Jolicoeur	PCA without size standardization	Studying integrated form variation
PC1 of Boas Coordinates	Huxley-Jolicoeur	PCA of Boas coordinates	Analyzing form variation with minimal size-shape separation

Molecular Data Integration Protocols

Molecular Marker Selection and Sequencing

• DNA Extraction: Use standardized extraction protocols appropriate for the specimen type (fresh, ethanol-preserved, or historical samples).
• Marker Selection: Employ standard taxonomic markers:
  • Animal Systems: COI, ITS2, 16S rRNA
  • Plant Systems: ITS, matK, rbcL
• Sequence Verification: Assemble and edit contigs, then align sequences using algorithms such as MUSCLE or MAFFT.

Phylogenetic Analysis

• Tree Reconstruction: Implement multiple methods (Maximum Likelihood, Bayesian Inference) to estimate phylogenetic relationships.
• Support Assessment: Calculate bootstrap values or posterior probabilities to assess node support.
• Concordance Testing: Compare topological concordance between molecular phylogenies and morphometric groupings.

Ecological Data Collection and Integration

Ecological Variable Assessment

• Geographic Data: Record precise collection localities and environmental variables (climate, elevation, vegetation).
• Host/Habitat Association: Document specific ecological associations (host plants, microhabitats, seasonal patterns).
• Symbiont Data: When relevant, characterize microbial symbionts that may influence morphology [62].

Statistical Integration of Ecological Data

• Multivariate Analyses: Implement canonical correspondence analysis or redundancy analysis to test morphology-environment correlations.
• Niche Modeling: Develop ecological niche models to test for association between morphological variation and environmental factors.

Data Integration and Statistical Framework

Simultaneous Analysis Framework

• Data Standardization: Normalize variables to comparable scales when combining different data types.
• Concordance Testing: Assess congruence between different data partitions (morphometric, molecular, ecological) before combined analysis.
• Iterative Corroboration: Implement an iterative approach where hypotheses generated from one data source are tested against others [62].

Allometry-Corrected Morphometric Integration

• Allometry Correction: Compute allometry-free shape residuals by extracting the residuals from the regression of shape on size [4].
• Phylogenetic Comparative Methods: When combining with molecular phylogenies, use phylogenetic generalized least squares (PGLS) to account for evolutionary non-independence.
• Discriminant Analysis: Apply canonical variate analysis (CVA) to allometry-corrected shape data to maximize separation among putative taxa [50].

Table 2: Statistical Methods for Data Integration in Taxonomic Studies

Analysis Type	Data Input	Function	Software Implementation
Partial Least Squares (PLS)	Shape + Ecological variables	Tests association between shape and ecology	MorphoJ, R geomorph
Phylogenetic PCA	Shape + Phylogeny	Identifies shape variation independent of phylogeny	R phytools, geomorph
Canonical Variate Analysis (CVA)	Allometry-corrected shape	Maximizes group separation	MorphoJ, PAST
Mahalanobis Distances	Allometry-corrected shape	Quantifies morphological divergence	MorphoJ, PAST
Procrustes ANOVA	Shape + Grouping factors	Tests group differences in shape	MorphoJ, R geomorph

Validation and Species Delimitation Protocol

Hypothesis Testing Framework

• Initial Group Formation: Define preliminary taxonomic hypotheses based on initial morphological assessment.
• Molecular Validation: Test monophyly of morphologically defined groups using molecular phylogenies.
• Ecological Consistency: Assess whether putative taxa show ecological differentiation or niche conservatism.
• Iterative Refinement: Modify taxonomic hypotheses based on discordant patterns among data types.

Species Delimitation Criteria

• Morphological Diagnosis: Statistically significant differences in allometry-corrected shape [50].
• Molecular Differentiation: Monophyletic groupings with adequate genetic distances.
• Ecological Differentiation: Distinct ecological associations or distributions.
• Sympatric Co-occurrence: Evidence of distinct forms maintaining separation in sympatry.

Research Reagent Solutions and Computational Tools

Table 3: Essential Software Tools for Integrative Taxonomic Studies

Software/Tool	Primary Function	Application in Protocol	Access
TPSDig2	Landmark digitization	Initial morphometric data collection	Free
MorphoJ	Geometric morphometrics analysis	Allometry assessment, CVA, visualization	Free
R geomorph package	Comprehensive GM analysis	Multivariate regression, PLS, Procrustes ANOVA	Free
Geneious	Molecular data assembly and alignment	Sequence management, alignment, phylogenetic setup	Commercial/Free
PAUP*/MrBayes	Phylogenetic analysis	Tree building, support assessment	Free
R phytools	Phylogenetic comparative methods	Phylogenetic signal, PGLS analyses	Free

Case Study Application: Leaf-Footed Bugs (Acanthocephala)

A recent study on the genus Acanthocephala demonstrates the application of these integrated approaches [50]. Researchers analyzed pronotum shape variation across 11 species using geometric morphometrics, finding that principal component analysis accounted for 67% of total shape variation and revealed distinct shape patterns useful for species discrimination. The study employed:

• Allometry Assessment: Multivariate regression of shape on centroid size to test for allometric effects
• Group Discrimination: Canonical variate analysis on allometry-corrected shape data
• Statistical Validation: Mahalanobis distances and Procrustes ANOVA to quantify morphological divergence
• Taxonomic Application: Using resulting morphological clusters to refine species boundaries

This approach successfully distinguished multiple species of quarantine concern, demonstrating the practical taxonomic utility of these methods in agriculturally important insect groups [50].

Troubleshooting and Technical Considerations

Common Challenges and Solutions

• Allometry-Phylogeny Confounding: When allometric patterns are confounded with phylogenetic relationships, use phylogenetic comparative methods to separate these effects.
• Missing Data: Implement estimation approaches for incomplete specimens while accounting for potential biases [2].
• Measurement Error: Conduct replicate measurements to quantify and account for measurement error, particularly when subtle shape differences are taxonomically informative.
• Conflicting Signals: Develop explicit protocols for reconciling discordant patterns from different data types, including consideration of evolutionary processes that might generate such discordance.

Methodological Validation

• Measurement Error Assessment: Conduct replicate landmarking on a subset of specimens to quantify precision [2].
• Statistical Power Analysis: Evaluate sample size adequacy for detecting meaningful biological effects.
• Method Sensitivity: Test analytical sensitivity to alternative parameter choices or methodological approaches.

This integrated protocol provides a comprehensive framework for taxonomic studies that leverage multiple evidence types while explicitly addressing the confounding effects of allometry in morphological datasets.

Geometric morphometrics (GM) has emerged as a powerful tool for resolving taxonomic uncertainties in insects, proving particularly valuable for species identification in agriculturally important pests where traditional keys are lacking [50]. This application note details a protocol for applying GM to analyze pronotum shape variation within the leaf-footed bug genus Acanthocephala (Hemiptera: Coreidae). The methodology is explicitly framed within a research context that requires correcting for allometric effects—the influence of size on shape—to isolate pure shape variation for robust taxonomic discrimination [4] [3]. The approach described herein successfully distinguished among 11 Acanthocephala species, several of which are of quarantine concern to the United States, demonstrating the practical utility of this method for pest monitoring and agricultural biosecurity [50].

Background and Principles

In taxonomic morphometric studies, failing to account for allometry can confound species-specific shape differences with changes in shape that are a direct consequence of size variation [3]. Two primary conceptual frameworks guide the study of allometry:

• The Gould-Mosimann School: Defines allometry as the covariation between shape and size. In GM, this is typically implemented through the multivariate regression of shape variables on a size measure (e.g., centroid size) [3].
• The Huxley-Jolicoeur School: Defines allometry as the covariation among morphological features that all contain size information. This framework characterizes allometric trajectories using the first principal component in form space (size-and-shape space) [4] [3].

The protocol that follows is grounded in the Gould-Mosimann school, using multivariate regression to test for and correct allometric effects, thereby ensuring that subsequent taxonomic discrimination is based on size-independent shape characters [50] [4].

Experimental Workflow and Protocols

Comprehensive Workflow Diagram

The following diagram outlines the complete analytical pipeline, from specimen preparation to statistical analysis and allometry correction.

Detailed Step-by-Step Protocols

Protocol 1: Specimen Preparation and Image Acquisition

Objective: To obtain standardized, high-resolution digital images of Acanthocephala pronota for morphometric analysis.

Materials:

• Curated insect specimens (pinned or preserved in ethanol)
• Stereomicroscope with consistent illumination
• High-resolution digital camera mounted on microscope
• Image capture software
• Scale bar for calibration

Procedure:

• Selection: Select adult specimens verified to species level by taxonomic experts. The referenced study analyzed 54 specimens across 11 species [50].
• Positioning: Position the specimen under the stereomicroscope so the pronotum is in a consistent dorsal view, with the anterior and posterior margins parallel to the image plane.
• Imaging: Capture high-resolution images using the digital camera system. Ensure lighting is uniform to avoid shadows that obscure morphological landmarks.
• Calibration: Include a scale bar in each image for spatial calibration during landmark digitization.
• Curation: Store images in a standardized format (e.g., JPG, TIF) with filenames that encode specimen ID and species designation.

Protocol 2: Landmark Digitization and Data Collection

Objective: To capture pronotum shape by digitizing homologous anatomical landmarks across all specimens.

Materials:

• Computer workstation
• TPSDig2 software (v2.17 or later) [50]

Procedure:

• Software Setup: Install and launch TPSDig2.
• Landmark Scheme: Apply a consistent scheme of 40 Type I and Type II landmarks to cover the pronotum geometry [50]. Example landmarks include:
  • Anterior and posterior points of the medial line
  • Lateral extremes of the pronotal margins
  • Points of maximum curvature on the anterior and posterior edges
  • Bases of prominent spines or projections (if present and homologous)
• Digitization: For each specimen image, digitize all landmarks in a predefined order.
• Data Export: Save the landmark coordinates in TPS or NTS format for subsequent analysis.

Protocol 3: Shape Data Preprocessing and Allometry Correction

Objective: To remove non-shape variation (position, orientation, scale) and statistically account for allometric effects.

Materials:

• MorphoJ software (v1.06d or later) or R with the geomorph package [50]

Procedure:

• Generalized Procrustes Analysis (GPA):
  • Import the raw landmark coordinates into MorphoJ.
  • Perform GPA to superimpose landmark configurations. This step removes differences in translation, rotation, and scale, producing a matrix of Procrustes shape coordinates for analysis [50].
• Allometry Assessment:
  • Run a multivariate regression of shape coordinates on centroid size (a geometric measure of size computed as the square root of the sum of squared distances of all landmarks from their centroid) [4] [3].
  • Assess the statistical significance of the regression (e.g., using a permutation test with 10,000 rounds).
• Allometry Correction:
  • If allometry is non-significant, proceed with the original Procrustes coordinates for taxonomic analysis.
  • If allometry is significant, calculate the regression residuals. These residuals represent the shape variation after removing the component that covaries with size [4]. Use these size-corrected residuals for all subsequent taxonomic analyses.

Protocol 4: Multivariate Analysis for Species Discrimination

Objective: To visualize and test for significant differences in pronotum shape among Acanthocephala species.

Procedure:

• Principal Component Analysis (PCA):
  • Perform PCA on the (size-corrected) covariance matrix of shape variables.
  • Visualize specimen distribution in morphospace along the first few principal components (PCs), which capture the major axes of shape variation. The referenced study found the first three PCs explained 67% of total shape variation [50].
• Canonical Variate Analysis (CVA):
  • Conduct CVA using species identity as the grouping factor. CVA maximizes separation among pre-defined groups.
  • Visualize group means (canonical variate means) and individual specimens in the space of the first few canonical variates (CVs).
• Statistical Validation:
  • Calculate Mahalanobis distances between species pairs based on the CVA.
  • Perform permutation tests to determine the statistical significance of these distances, validating morphological distinctness.

Key Research Reagents and Solutions

Table 1: Essential Software and Digital Tools for Pronotum Shape Analysis

Tool Name	Specific Function	Application Context in Protocol
TPSDig2 [50]	Landmark digitization from digital images	Protocol 2: Capturing raw coordinate data from pronotum images.
MorphoJ [50]	Integrated geometric morphometric analysis	Protocol 3 & 4: Performing GPA, regression, PCA, CVA, and statistical testing.
R with geomorph package [50]	Programmatic geometric morphometric analysis	Protocol 3 & 4: An alternative, scriptable environment for all statistical analyses.
High-Resolution Digital Camera	Image acquisition	Protocol 1: Creating high-quality input data for landmarking.
Stereomicroscope	Specimen visualization and imaging	Protocol 1: Ensuring precise specimen positioning and detail resolution.

Data Interpretation and Taxonomic Application

Table 2: Representative Results from Geometric Morphometric Analysis of 11 Acanthocephala Species

Analysis Method	Key Outcome	Taxonomic Utility
Principal Component Analysis (PCA)	First 3 PCs accounted for 67% of total shape variation [50].	Identified major axes of pronotum shape variation useful for initial species separation.
Multivariate Regression	Used to test for allometry (shape-size covariation) [50].	Critical pre-processing step to ensure species discrimination is based on size-independent shape.
Canonical Variate Analysis (CVA)	Revealed significant morphological separation among species [50].	Maximized group differences, providing a powerful model for species classification.
Mahalanobis Distances	Most pairwise comparisons between species were statistically significant [50].	Quantified the degree of morphological divergence and provided statistical support for species distinctions.
Morphospace Overlap	Some overlap was observed among closely related taxa [50].	Highlights taxonomic complexity and potential limitations of the method for very recently diverged species.

Interpretation Guidelines

• Successful Discrimination: Significant Mahalanobis distances and non-overlapping confidence ellipses in CVA plots indicate that pronotum shape is a reliable character for distinguishing those species.
• Handling Overlap: Morphological overlap in CVA or PCA plots, particularly between closely related species, suggests the need for integrative taxonomy [64]. Pronotum GM should be combined with other data (e.g., molecular, genitalia morphology) for conclusive identification.
• Allometry as a Feature: In some cases, allometry itself can be a taxonomically informative trait if different species exhibit distinct allometric trajectories.

This protocol provides a standardized, reproducible framework for applying geometric morphometrics to the taxonomic identification of Acanthocephala bugs, with built-in procedures for correcting allometry. The case application demonstrates that pronotum shape is a reliable character for species delimitation within this genus of agricultural and quarantine importance [50]. By isolating size-independent shape variation, researchers can achieve more robust and biologically informative species discrimination, enhancing capabilities in pest monitoring, quarantine inspection, and broader agricultural biosecurity programs.

In taxonomic studies based on geometric morphometrics (GM), allometry—the relationship between size and shape—represents a pervasive source of variation that can confound the identification of genuine taxonomic differences. Failing to account for allometric effects may lead researchers to misinterpret size-related shape changes as taxonomically diagnostic characters, potentially resulting in incorrect classifications. This application note provides a structured comparison of the primary methodological frameworks for correcting allometry in taxonomic GM studies, equipping researchers with evidence-based guidance for selecting appropriate correction approaches. The recommendations are framed within the context of a broader thesis on correcting for allometry, emphasizing practical implementation for researchers engaged in species identification and classification, particularly with fossil specimens or diverse populations where allometric variation is significant.

Theoretical Frameworks of Allometry

The two predominant schools of thought in allometry research offer distinct conceptual and methodological approaches for understanding size-shape relationships.

The Gould–Mosimann School: Covariation of Shape with Size

This framework strictly separates size and shape as distinct biological constructs. Allometry is formally defined as the covariation of shape with size. Methodologically, this approach employs a multivariate regression of shape variables (e.g., Procrustes coordinates) on a measure of size (typically centroid size). The regression vector describes the allometric trajectory, and the residuals from this regression represent size-corrected shape data. This method is particularly powerful for isolating the specific component of shape variation that is predicted by size, making it ideal for testing explicit hypotheses about allometry [3] [4].

The Huxley–Jolicoeur School: Covariation among Morphological Features

This school characterizes allometry as the covariation among multiple morphological traits, all of which contain size information. Instead of separating size and shape a priori, it identifies the primary axis of morphological covariation related to size. The first principal component (PC1) of the untransformed or log-transformed measurements often represents this allometric trajectory. In geometric morphometrics, this is implemented by analyzing data in Procrustes form space (or conformation space), where configurations are aligned for position and orientation but not scaled. The PC1 in this space captures the major axis of form variation, which typically corresponds to allometry [3] [4].

Table 1: Comparison of Allometric Frameworks in Geometric Morphometrics

Feature	Gould–Mosimann School	Huxley–Jolicoeur School
Core Definition	Covariation between shape and size	Covariation among morphological traits containing size information
Size & Shape Relationship	Separated a priori	Integrated as "form"
Primary Method	Multivariate regression of shape on size	First principal component (PC1) in form space
Typical Space Used	Shape tangent space	Conformation space (size-and-shape space)
Key Output	Allometric regression vector	PC1 allometric trajectory
Statistical Emphasis	Explaining shape variation via size	Describing major axis of morphological covariation

The diagram above illustrates the logical progression from raw data to taxonomic application through two distinct methodological pathways.

Performance Comparison of Correction Methods

Computer simulation studies have provided critical insights into the performance characteristics of different allometric correction methods under controlled conditions. These analyses reveal distinct strengths and weaknesses that should inform methodological selection.

Performance Under Idealized and Real-World Conditions

In simulations with no residual variation around the allometric relationship, all four major methods (regression of shape on size, PC1 of shape, PC1 in conformation space, and PC1 of Boas coordinates) demonstrate strong logical consistency, producing nearly identical allometric vectors. This confirms their fundamental validity for analyzing deterministic allometric relationships [4].

However, under more realistic conditions with residual variation, performance differences emerge clearly:

• Multivariate regression of shape on size consistently outperforms the PC1 of shape when residual variation is either isotropic (uniform in all directions) or follows a pattern independent of allometry [4].
• PC1 in conformation space and PC1 of Boas coordinates yield nearly identical results that closely approximate the true simulated allometric vectors across all conditions, with a marginal performance advantage for the conformation space approach [4].
• For taxonomic studies specifically, regression-based correction demonstrates particular utility when the research goal is to remove allometric effects to better visualize or analyze taxonomically informative shape variation independent of size.

Table 2: Performance Comparison of Allometric Methods Based on Simulation Studies

Method	Theoretical School	Isotropic Noise Performance	Anisotropic Noise Performance	Best Application Context
Regression of Shape on Size	Gould-Mosimann	Excellent	Excellent	Hypothesis-testing about allometry; Size correction for taxonomy
PC1 of Shape	Gould-Mosimann	Good	Moderate (can be biased)	Exploratory analysis when allometry is the dominant signal
PC1 in Conformation Space	Huxley-Jolicoeur	Excellent	Excellent	Identifying major allometric trajectory in form space
PC1 of Boas Coordinates	Huxley-Jolicoeur	Excellent	Excellent	Alternative to conformation space with similar performance

Detailed Experimental Protocols

Protocol 1: Multivariate Regression of Shape on Size (Gould-Mosimann Approach)

This protocol provides a step-by-step methodology for implementing regression-based allometric correction, which is particularly effective for taxonomic studies where isolating non-size-related shape variation is crucial.

Step 1: Image Acquisition and Preparation

• Capture high-resolution images of specimens with a standardized scale and orientation.
• For fish morphology, ensure "the fish was placed horizontally on a solid-colored background directly beneath the camera" with the body axis horizontal [65].
• Use background removal tools if necessary to isolate the specimen from distracting elements.

Step 2: Landmark Digitization

• Select landmarks covering the morphology of interest:
  • Type I Landmarks: Anatomically discrete, homologous points (e.g., jaw joint, tooth cusps).
  • Type II Landmarks: Geometrically defined points (e.g., point of maximum curvature).
  • Type III Landmarks: Constructed points (e.g., midpoints between other landmarks).
• For complex curves, add semi-landmarks to capture outline information [66].
• Use software such as tpsDig2 for digitizing landmarks and semi-landmarks [65].

Step 3: Generalized Procrustes Analysis (GPA)

• Superimpose landmark configurations using GPA to remove non-shape variation (position, orientation).
• This can be performed in MorphoJ or R (using the geomorph package).
• The output is a set of Procrustes shape coordinates for subsequent analysis.

Step 4: Multivariate Regression

• Compute centroid size for each specimen as the square root of the sum of squared distances of all landmarks from their centroid.
• Perform multivariate regression of Procrustes coordinates on centroid size (or log-transformed centroid size).
• In MorphoJ: Select "Regression" from the "Standard" menu.
• In R: Use the procD.lm() function in the geomorph package.

Step 5: Size Correction and Residual Extraction

• Extract the regression residuals, which represent shape variation independent of size.
• These size-corrected shapes can be used for subsequent taxonomic analyses (e.g., discriminant analysis, clustering).

Step 6: Visualization

• Visualize allometric pattern using a vector plot or deformation grid.
• Visualize size-corrected shapes along principal components to explore taxonomic structure.

Protocol 2: PC1 in Conformation Space (Huxley-Jolicoeur Approach)

This protocol outlines the procedure for analyzing allometry as the primary axis of form variation, which is valuable when researchers wish to preserve the integrated nature of size and shape in their analyses.

Step 1: Data Preparation and Landmarking

• Follow the same image acquisition and landmark digitization procedures as in Protocol 1.

Step 2: Procrustes Superimposition Without Scaling

• Align landmark configurations for position and orientation only, without scaling to unit size.
• This preserves size variation in the data, creating "conformation space" (size-and-shape space).
• This step can be performed in MorphoJ or using the gpagen() function in the geomorph R package with the ProcD = FALSE option.

Step 3: Principal Component Analysis (PCA)

• Perform PCA on the superimposed coordinates from conformation space.
• The first principal component (PC1) typically represents the allometric trajectory.

Step 4: Interpretation of Allometric Vector

• Examine the loadings of PC1 to interpret the allometric pattern.
• Correlate PC1 scores with centroid size to confirm it represents allometry (high correlation expected).

Step 5: Group Comparison and Visualization

• Compare PC scores among taxonomic groups to assess whether they share common allometric patterns.
• Visualize shape changes along PC1 using deformation grids.

The workflow diagram above contrasts the two primary methodological pathways for allometric analysis, from initial image data to final taxonomic application.

The Scientist's Toolkit

Table 3: Essential Software and Tools for Allometric Analysis in Geometric Morphometrics

Tool Name	Type	Primary Function	Application in Allometric Analysis
tpsDig2	Desktop Application	Landmark digitization	Collecting x,y coordinates from specimen images [65]
tpsUtil	Desktop Application	TPS file management	Creating, editing, and managing TPS data files [65]
MorphoJ	Desktop Application	Morphometric analysis	Performing Procrustes ANOVA, regression, and PCA [65]
R (geomorph package)	Programming Environment	Comprehensive morphometric analysis	Advanced regression, PCA, and permutation tests [65]
R (Momocs package)	Programming Environment	Outline analysis	Analyzing outline data using Fourier analysis [65]
ImageJ	Desktop Application	Image processing	Preparing and pre-processing specimen images [65]

Application in Taxonomic Studies

The strategic selection of allometric correction methods has demonstrated significant value in practical taxonomic applications, particularly for challenging discrimination tasks.

Case Study: Shark Tooth Taxonomy

A compelling example comes from the taxonomic study of isolated fossil shark teeth, where geometric morphometrics successfully recovered taxonomic separation based on tooth morphology. In this study, researchers digitized a total of seven homologous landmarks and eight semilandmarks on each tooth specimen to capture overall shape. The analysis effectively separated genera including Brachycarcharias, Carcharias, Carcharomodus, and Lamna, demonstrating the power of shape-based discrimination for taxonomically challenging groups. This approach proved particularly valuable as it captured "additional shape variables that traditional methods did not consider," providing more comprehensive morphological information for reliable taxonomic identification [66].

Practical Recommendations for Taxon Discrimination

Based on empirical performance and theoretical considerations, the following recommendations are provided for taxonomic studies:

• When analyzing taxa with significant size overlap, the Huxley-Jolicoeur approach (PC1 in conformation space) often provides the most biologically meaningful discrimination.
• When studying taxa with substantial size differences that may confound true taxonomic shape differences, the Gould-Mosimann approach (regression-based correction) is preferred to isolate size-independent shape variation.
• For fossil taxa with incomplete preservation, focus on Type I landmarks, which offer higher reliability across specimens, and be cautious when applying allometric corrections with limited sample sizes.
• Always validate allometric patterns visually using deformation grids or vector diagrams to ensure biological interpretability of the statistical results.

The selection of appropriate allometric correction methods represents a critical methodological decision in taxonomic geometric morphometrics. The Gould-Mosimann framework, with its regression-based approach, offers superior performance for isolating size-independent shape variation when the research goal is to remove allometric effects for clearer taxonomic discrimination. Conversely, the Huxley-Jolicoeur approach, utilizing PC1 in conformation space, provides a more integrated perspective on allometry as the primary axis of morphological variation. Informed method selection, based on both theoretical alignment with research questions and empirical performance characteristics, significantly enhances the reliability and interpretability of taxonomic distinctions based on shape data.

Assessing Taxonomic Signal Preservation Post-Correction

In geometric morphometric (GMM) analyses, correcting for allometry—the pattern of shape change correlated with size—is a crucial step to isolate taxonomic signals from size-related variation. This protocol provides a structured framework for assessing whether biologically meaningful taxonomic information remains intact after allometric correction. The methods integrate multivariate statistics, shape regression, and validation diagnostics to ensure that size correction clarifies rather than obscures phylogenetic patterns, with particular consideration for applications in paleontology and evolutionary biology.

Allometry, the pattern of covariation between shape and size, presents a significant challenge in taxonomic morphometric studies. When allometric patterns are strong and conserved across taxa, correcting for them is essential to reveal shape differences independent of size [51]. However, such correction risks distorting or removing the very taxonomic signal researchers seek to preserve. This protocol addresses the critical need to verify that allometric correction clarifies, rather than obscures, phylogenetically informative shape variation.

The theoretical foundation for this approach bridges two historical schools of allometric analysis: the Gould-Mosimann framework (allometry as shape covariation with size) and the Huxley-Jolicoeur framework (allometry as covariation among size-loaded traits) [3]. In taxonomic contexts, we must distinguish whether observed shape differences reflect genuine phylogenetic divergence or mere allometric consequences of size differences.

Theoretical Framework & Key Concepts

Allometry in Geometric Morphometrics

Table 1: Concepts of Allometry in Morphometrics

Concept	Definition	Analytical Approach
Gould-Mosimann School	Allometry as covariation of shape with size	Multivariate regression of shape coordinates on size
Huxley-Jolicoeur School	Allometry as covariation among morphological features containing size information	First principal component in form space
Static Allometry	Shape-size relationship within a single population/age group	Regression within operational taxonomic units
Evolutionary Allometry	Shape-size relationship across taxa or evolutionary lineages	Phylogenetically informed comparative methods

In geometric morphometrics, allometry is typically quantified as the multivariate regression of Procrustes shape coordinates on a size measure, usually centroid size [3]. The proportion of shape variance explained by size (R²) indicates allometric strength, while the regression vector characterizes allometric direction.

Taxonomic Signal Preservation

Taxonomic signal preservation after allometric correction requires that:

• Phylogenetically informative shape differences remain detectable
• Allometric correction does not introduce mathematical artifacts
• Size-corrected shapes maintain biological interpretability
• Group discrimination power is maintained or improved

Experimental Protocol

The following diagram illustrates the comprehensive workflow for assessing taxonomic signal preservation after allometric correction:

Data Collection & Preprocessing

Landmarking Protocol:

• Landmark Type Selection: Combine Type I (anatomical homology), Type II (geometrically defined), and Type III (sliding semilandmarks for curves)
• Data Collection: Use TpsDig2 for 2D landmarks or specialized software (e.g., MorphoDig) for 3D data
• Error Reduction: Repeat landmarking 3+ times by the same observer; calculate measurement error via Procrustes ANOVA
• Outlier Detection: Use Procrustes distance-based methods to identify potential landmarking errors

For fossil specimens, additional considerations apply. Body curvature in fossil fishes, for instance, requires mathematical "unbending" using specialized TPS functions before allometric analysis [24].

Allometric Correction Methods

Table 2: Allometric Correction Methods in Geometric Morphometrics

Method	Procedure	Taxonomic Signal Preservation
Multivariate Regression	Use residuals from shape ~ size regression	Preserves non-allometric shape variation; may retain phylogenetic signal
Burnaby's Method	Projection perpendicular to allometric vector	Appropriate when allometry is conserved across groups
Group-Specific Correction	Separate regressions per taxon	Preserves inter-group allometric differences
Phylogenetic Correction	PGLS regression incorporating phylogenetic relationships	Explicitly models evolutionary relationships

Implementation of Multivariate Regression Correction:

Taxonomic Signal Validation Protocol

Statistical Validation Steps

• Principal Component Analysis Comparison

  • Perform PCA on raw and corrected shapes
  • Compare variance explained by principal components
  • Assess changes in group separation in morphospace

• Discriminant Function Analysis

  • Calculate cross-validated classification rates pre- and post-correction
  • Compare Mahalanobis distances between groups
  • Perform MANOVA on corrected shapes with taxonomy as factor

• Variance Partitioning Analysis

  • Quantify variance components for size, taxonomy, and their interaction
  • Use VARPART analysis to assess pure taxonomic effects

• Effect Size Monitoring

  • Calculate Procrustes F-statistics for group differences
  • Monitor changes in effect sizes for known taxonomic distinctions

Diagnostic Criteria for Signal Preservation

Taxonomic signal is considered preserved when:

• Cross-validated classification accuracy remains >80% of pre-correction values
• Mahalanobis distances between established taxa remain statistically significant
• Morphological disparities between groups retain biological interpretability
• Known phylogenetic patterns remain discernible in morphospace

The Scientist's Toolkit

Table 3: Essential Research Reagents & Solutions for Taxonomic Morphometrics

Tool/Software	Primary Function	Application Context
TPS Series (tpsDig2, tpsUtil)	Landmark digitization & data management	Raw data collection; "unbending" fossil specimens
R geomorph package	Procrustes ANOVA, allometric analysis, phylogenetic integration	Multivariate statistical analysis of shape data
MorphoJ	User-friendly GM analysis, discriminant analysis	Introductory analyses; visualization
R ape & phytools	Phylogenetic comparative methods	Phylogenetic Generalized Least Squares (PGLS)
EVAN Toolbox	Paleontological morphometrics	Fossil-specific analyses including fragmentary specimens
Landmark Editor (IDAV)	3D landmark collection	3D geometric morphometrics

Case Application: Fossil Fishes with Postmortem Deformation

A study on Cretaceous gonorynchiform fossil fishes demonstrates the protocol's application. Researchers compared regression-based correction and TPS "unbending" to address postmortem body curvature, a source of non-biological deformation [24].

Key Findings:

• Both methods successfully corrected curvature-induced geometric variation
• The "unbending" method was more appropriate for fossil specimens with additional taphonomic alterations
• Taxonomic distinctions between Rubiesichthys gregalis and Gordichthys conquensis remained detectable post-correction
• Larger individuals exhibited less curvature, demonstrating allometric considerations in preservation

Interpretation Guidelines

The following decision diagram guides interpretation of taxonomic signal preservation results:

Interpretation of Outcomes:

• Strong Preservation: Proceed with corrected data for taxonomic analyses
• Moderate Preservation: Use corrected data with caution; report validation results transparently
• Weak Preservation: Reconsider correction method; allometry may be intrinsic to taxonomic differences

Concluding Recommendations

Successful preservation of taxonomic signal following allometric correction requires:

• Study Design Stage: Ensure adequate sample sizes per taxon and coverage of morphological diversity
• Analytical Stage: Apply multiple validation approaches and consistency checks
• Interpretation Stage: Maintain biological context when evaluating statistical outcomes
• Reporting Stage: Transparently document all validation results, including any signal attenuation

This protocol emphasizes that allometric correction should clarify rather than obscure taxonomic signals. When correction substantially weakens taxonomic discrimination, this may indicate that allometry itself carries phylogenetic information worth retaining in analyses.

Conclusion

Correcting for allometry is not merely a statistical procedure but a fundamental requirement for robust taxonomic analysis using geometric morphometrics. By systematically addressing size-related shape variation, researchers can isolate true taxonomic signals essential for accurate species delimitation. The integration of foundational concepts with practical methodologies provides a comprehensive framework applicable across diverse biological systems. Future directions should focus on developing more sophisticated correction algorithms that account for modular allometry and phylogenetic non-independence, alongside improved validation protocols that integrate multiple lines of evidence. As geometric morphometrics continues to evolve, rigorous allometry correction will remain central to advancing taxonomic research and understanding evolutionary patterns.

References

• 1. Size, shape, and form: concepts of allometry in geometric ... [https://pubmed.ncbi.nlm.nih.gov/27038023/]
• 2. A practical, step-by-step, guide to taxonomic comparisons ... [https://europeanjournaloftaxonomy.eu/index.php/ejt/article/view/2527]
• 3. Size, shape, and form: concepts of allometry in geometric ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC4896994/]
• 4. Methods for studying allometry in geometric morphometrics [https://link.springer.com/article/10.1007/s10682-022-10170-z]
• 5. Are geometric morphometric analyses replicable? Evaluating ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC7140992/]
• 6. Correction to: Methods for studying allometry in geometric ... [https://link.springer.com/article/10.1007/s10682-022-10179-4]
• 7. A geometric morphometric protocol to correct postmortem ... [https://peerj.com/articles/17436/]
• 8. Geometric morphometrics approach for classifying children's ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11785739/]
• 9. Geometric Morphometric Versus Genomic Patterns in a ... [https://www.mdpi.com/2079-7737/12/3/418]
• 10. Quantitative shape analysis with weighted covariance ... [https://frontiersinzoology.biomedcentral.com/articles/10.1186/1742-9994-10-16]
• 11. A novel approach to geometric morphometrics [https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1009061]
• 12. Thirty years of geometric morphometrics - PubMed Central - NIH [https://pmc.ncbi.nlm.nih.gov/articles/PMC9545184/]
• 13. A practical, step-by-step, guide to taxonomic comparisons ... [https://europeanjournaloftaxonomy.eu/index.php/ejt/article/view/2529]
• 14. Allometry: The Study of Biological Scaling [https://www.nature.com/scitable/knowledge/library/allometry-the-study-of-biological-scaling-13228439/]
• 15. Relationships among ontogenetic, static, and evolutionary ... [https://pubmed.ncbi.nlm.nih.gov/7149015/]
• 16. Geometric morphometric analysis [https://bio-protocol.org/exchange/minidetail?id=3214040&type=30]
• 17. Defining, Reconstructing and Investigating Shape Through a ... [https://journal.caa-international.org/articles/10.5334/jcaa.161]
• 18. Testing the reliability of geometric morphometric and ... [https://www.sciencedirect.com/science/article/pii/S2666033425000048]
• 19. gpagen: Generalized Procrustes analysis of points, curves ... [https://rdrr.io/cran/geomorph/man/gpagen.html]
• 20. FACEDIG automated tool for placing landmarks on facial ... [https://www.nature.com/articles/s41598-025-09714-4]
• 21. Geometric Morphometrics for Beginners [http://www.transmittingscience.com/courses/geometric-morphometrics/geometric-morphometrics-for-beginners/]
• 22. MorphoJ [https://morphometrics.uk/MorphoJ_page.html]
• 23. Allometric effect [https://groups.google.com/g/geomorph-r-package/c/1jXm_6dNE3A]
• 24. A geometric morphometric protocol to correct postmortem ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11146327/]
• 25. Allometric test [https://groups.google.com/g/geomorph-r-package/c/OkQ-_wOvKAk]
• 26. MorphoJ User's Guide - Klingenberg lab web site [https://morphometrics.uk/MorphoJ_guide/]
• 27. Geometric Morphometric Analyses of 2D and 3D ... - geomorph [https://geomorphr.r-universe.dev/geomorph]
• 28. Divide and conquer? Size adjustment with allometry and ... [https://bmcbiol.biomedcentral.com/articles/10.1186/s12915-017-0448-5]
• 29. Mandibular Shape Variation, Allometry and Modularity in Adult ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12649264/]
• 30. 2.4. Geometric morphometric analyses [https://bio-protocol.org/exchange/minidetail?id=4780712&type=30]
• 31. Cranial divergence among evolutionary lineages of ... [https://bioone.org/journalArticle/Download?urlId=10.1644%2F11-MAMM-A-260.2]
• 32. Parallelism of mandibular function in therian carnivores [https://www.cambridge.org/core/journals/paleobiology/article/parallelism-of-mandibular-function-in-therian-carnivores-a-morphometric-phylogenetic-and-finite-element-analysis/5775F2C350A864CE3C324D882D64EFD5]
• 33. Endocranial shape variation and allometry in ... [https://www.nature.com/articles/s41598-024-68390-y]
• 34. Principles of Effective Data Visualization - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC7733875/]
• 35. Assesment and interpretation of negative forelimb allometry in ... [https://frontiersinzoology.biomedcentral.com/articles/10.1186/s12983-019-0342-9]
• 36. Including Distorted Specimens in Allometric Studies [https://pmc.ncbi.nlm.nih.gov/articles/PMC8341891/]
• 37. procD.allometry function [https://www.rdocumentation.org/packages/geomorph/versions/3.0.7/topics/procD.allometry]
• 38. The ontogenetic allometry of body morphology and ... [https://www.sciencedirect.com/science/article/pii/S1751731117001884]
• 39. Sample Size and Geometric Morphometrics Methodology ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10833145/]
• 40. Geometric Morphometrics sheds light on the systematics ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10540071/]
• 41. Allometry and phylogeny of within-diaspore biomass allocation [https://pmc.ncbi.nlm.nih.gov/articles/PMC12646799/]
• 42. Model selection and logarithmic transformation in ... [https://pubmed.ncbi.nlm.nih.gov/18513152/]
• 43. Multi-scaling allometry in human development, mammalian ... [https://www.nature.com/articles/s41598-024-69199-5]
• 44. Quick Guide to Geomorph [http://www.emmasherratt.com/uploads/2/1/6/0/21606686/quick_guide_to_geomorph-_introduction.html]
• 45. Geometric Morphometrics in R - geomorph R package [https://geomorphr.github.io/GMcourse/index.html]
• 46. gmShiny v0.1.4 - Video Tutorials [https://www.gmshiny.com/video-tutorials]
• 47. Clovis point allometry, modularity, and integration [https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0289489]
• 48. Geometric morphometrics reveals sex-differential shape ... [https://peerj.com/articles/3617/]
• 49. Cranial integration and modularity: insights into evolution ... [http://www.italian-journal-of-mammalogy.it/Cranial-integration-and-modularity-insights-into-evolution-and-development-from-morphometric,77242,0,2.html]
• 50. When Shape Defines: Geometric Morphometrics Applied to ... [https://www.mdpi.com/1424-2818/17/10/680]
• 51. The differentiated impacts and constraints of allometry ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11920208/]
• 52. Geometric Morphometric Data Augmentation Using ... [https://www.mdpi.com/2076-3417/10/24/9133]
• 53. Classification Efficacy Using K-Fold Cross-Validation and ... [https://www.mdpi.com/2072-4292/13/15/2980]
• 54. A cross-validation-based approach for delimiting reliable ... [https://movementecologyjournal.biomedcentral.com/articles/10.1186/s40462-017-0110-4]
• 55. Deep learning image analysis for filamentous fungi taxonomic ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11387011/]
• 56. Regression - MorphoJ User's Guide [https://morphometrics.uk/MorphoJ_guide/covariation/residFromReg.htm]
• 57. Centric Allometry: Studying Growth Using Landmark Data [https://link.springer.com/article/10.1007/s11692-020-09530-w]
• 58. Leaf Morphology, Taxonomy and Geometric ... Morphometrics [https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0025630]
• 59. A Revision of the Traditional Analysis Method of Allometry to ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC9512611/]
• 60. plotAllometry: Plotting to assist visualization of shape-size ... [https://rdrr.io/cran/geomorph/man/plotAllometry.html]
• 61. Geometric morphometrics and ecological niche modelling ... [https://www.sciencedirect.com/science/article/pii/S2667114X23000079]
• 62. Integration of molecular, ecological, morphological and ... [https://pubmed.ncbi.nlm.nih.gov/22268975/]
• 63. Do morphometric data improve phylogenetic reconstruction ... [https://bmcecolevol.biomedcentral.com/articles/10.1186/s12862-024-02313-3]
• 64. Integrative taxonomy in helminth analysis: protocols and ... [https://parasitesandvectors.biomedcentral.com/articles/10.1186/s13071-025-06682-6]
• 65. Quantitative Analysis of Fish Morphology Through Landmark ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11540051/]
• 66. Geometric morphometrics on fossil shark teeth [https://palaeo-electronica.org/content/2025/5603-geometric-morphometrics-on-fossil-shark-teeth]