Continuous characters

From TNT

While other programs require that continuous characters be discretized, TNT can deal with continuous characters as such. This avoids the use of the rather ad hoc methods that have been proposed to discretize continuous distributions in phylogenetic analysis (gap-coding, Thiele's method, etc.; see Farris, 1990 for discussion of some of these methods). If there is significant variability in one of the terminals, it will probably be best represented by a range of one (or two) standard deviations around its mean. For normal distributions, this automatically takes care of non-significant differences. Continuous characters in TNT can have states 0-65, with 3 decimals (thus, when not doing implied weighting, tree scores are reported always with 3 decimals). They are always optimized as additive, i.e., using Farris optimization. They cannot be turned to nonadditive or sankoff. The continuous characters must be the first ones in the matrix (i.e. the first block or blocks, followed by blocks with other data formats). Each value must be separated by a blank or space; ranges are indicated by two values separated by a dash (-). Missing entries are indicated only as question marks (?). The treatment for the data when there are continuous characters is rather transparent. The commands/options that show reconstructions (recons command, or Optimize/Characters/Reconstructions) or count number of specific transformations (change command, Optimize/CountSpecificChanges) cannot be applied to continuous characters. Continuous characters can be edited with the xread= command, but not with the menu options Data/Edit/Taxon or Data/Edit/Character. Continuous characters can be named with the cnames command, but their states cannot. Standardization of continuous characters (i.e. rescaling so that the values are on a common scale) is left to the user. Under pre-defined weights, standardization will normally be an issue. Under implied weighting, the default formula is applied. Since continuous characters tend to have a larger s-m difference (s: steps, m: minimum possible), implied weighting should normally penalize them rather heavily. Thus, standardization of continuous characters is probably not so important under implied weighting.