LITERATURE-BASED CHARACTERS AS A SOURCE OF NON-RANDOM BIAS IN PHYLOGENETIC ANALYSES

Phylogenetic trees that include fossil taxa often are generated from a matrix of literature-based characters. This practice can introduce undesirable, non-biological, non-random patterns into the final trees. Literature-based character descriptions are used for a number of reasons: lack of access to specimens, lack of time to examine representative specimens due to taxonomic diversity within the chosen group, or lack of taxonomically intimate knowledge of the group being analyzed. Even if the most current references for a group are used, errors often are perpetuated through the literature. In addition, old photographs and line drawings may be inaccurate or may not show needed characters, leading to incorrect coding of characters. The approach of using literature-based characters can lead to biases in the final trees that may not be immediately apparent until actual type specimens are used to determine characters for creating phylogenies. Using only literature-based characters, trees were generated for species of the crinoid genus Allagecrinus (Crinoidea, Disparida). Distributions of terminal taxa on consensus trees were not related to stratigraphic distribution or paleobiogeography; rather, taxa were distributed by original author with taxa described by the same author showing the closest relationships. When type specimens of Allagecrinus species were examined, many original published descriptions and illustrations were corrected and clarified. A new matrix was created using these specimen-based observations, and the same parameters were used to create consensus trees. The resulting distribution of species was no longer by author, but rather reflected a temporal distribution. This example demonstrates at least one source of non-random bias that may occur using only literature-derived characters, and shows that assessment of characters using specimens can dramatically affect distribution of terminal taxa. Phylogenies on higher taxonomic levels may not need finer-scale characters that species- or genus-level phylogenies necessitate and may be buffered from this effect.