On Information Rank Deficiency in Phenotypic Covariance Matrices.

Area of Science:

• Evolutionary Biology
• Quantitative Genetics
• Geometric Morphometrics

Background:

• Phenotypic covariance matrices in geometric morphometrics can exhibit rank deficiency.
• This deficiency impacts the accuracy of phenotypic integration measures.
• Understanding and addressing this issue is crucial for reliable evolutionary and ecological studies.

Purpose of the Study:

• To define and investigate information rank deficiency in phenotypic covariance matrices.
• To demonstrate how this deficiency biases existing phenotypic integration metrics.
• To propose and validate a novel, information entropy-based metric for phenotypic integration.

Main Methods:

• Defined three types of matrix rank: full, mathematical, and information (effective) rank.
• Utilized information theory and eigenvalue spectrum analysis.
• Applied Generalized Procrustes analysis and correlation matrices.
• Developed a new metric combining standardized generalized variance with information entropy.
• Used dire wolf and Smilodon jaw data for empirical validation.

Main Results:

• Information rank deficiency arises from methodological factors and phenotypic covariance.
• Existing metrics (eigenvalue variance, standardized generalized variance) are biased by effective rank deficiency.
• The new information entropy-based metric accurately detects integration shifts in dire wolf jaws.
• Smilodon jaws exhibit higher phenotypic integration than dire wolf jaws.

Conclusions:

• Information rank deficiency is a critical issue in geometric morphometrics.
• The novel information entropy-based metric provides unbiased assessment of phenotypic integration.
• This metric allows for robust comparisons across datasets with varying sample sizes and variables.
• The concept of 'latent dispersion' quantifies shape information content.