Meeting Abstract
Approaches for quantifying form to assess phenotypic integration vary from linear measurements to high-density surface geometric morphometrics, and each has strengths and weaknesses. High-dimensional approaches better capture shape, particularly for complex structures, and allow more robust comparisons across disparate taxa, which may share few Type 1 landmarks. However, high-density geometric morphometric approaches also bring challenges, e.g., with dimensionality and possible statistical artefacts imposed by Procrustes superimposition. Here, we present simulations and four case studies spanning 730 species of squamates, birds, salamanders, and caecilians that exemplify the promise and diverse challenges of high-dimensional analyses of integration. We assess: 1) does Procrustes superimposition affect analyses of modularity?; 2) is big data actually better?; and 3) how do analyses of integration with high-dimensional semilandmarks compare to those with only landmarks? Procrustes superimposition can mask modularity, especially when the number of landmarks is low (e.g., < 15) and they covary in parallel directions, but the effect decreases with increasing landmark number or more complex modular covariance patterns. Landmark Sampling Evaluation Curve analyses demonstrate that for many skull regions 20-30 landmarks/semilandmarks are needed to accurately characterize shape, and landmark-only analyses do a particularly poor job in vault and rostrum bones. Full, subsampled, and landmark-only analyses of integration are generally congruent, but landmark-only analyses show higher integration between adjacent bones in which landmarks are generally limited to the sutures of those bones.
