A model for coding pollen size in reference to phylogeny using examples from the Ebenaceae

Abstract

Most genetically based features should be available for use in cladistic analysis. Palynologists routinely measure polar (P) and equatorial (E) axes and place pollen into size classes defined by earlier pollen workers. Grouping of pollen into globally arbitrary classes may not correspond to statistically significant differences among the taxa of a study. We propose a model using conventional statistical procedures coupled with data visualization and Monte Carlo simulation. This approach is not a final solution to the general problem of coding continuous characters into discrete states; it is an attempt to address the problems of character state delimitation in pollen morphology. We suggest that the coding of continuous measurement variables (e.g., P, E) into character states should be done following a logical sequence of interactive visualization (2D and 3D) of bivariate frequency distributions including the inspection of prediction and confidence ellipses (e.g., 99%), and use of ANOVA. We illustrate our approach using realistic pollen data sets generated by a computer program (POLSIM) written to perform Monte Carlo sampling from normally distributed statistical populations of polar and equatorial axes. Our model is then applied to an original data set of 4,134 pollen grains from the Ebenaceae, resulting in the coding of the four genera into three character states for pollen size.