Figure 1

Accuracy of ambiguity approaches. Comparison of four different methods for handling ambiguity Symbols on sequences of length 1000 with 2% ambiguities inserted uniformly at random. Each method was used to compute a distance matrix for the data containing ambiguities. Thereafter, three different matrix norms were computed from the distance matrices computed by the methods to the correct distance matrix, i.e., the matrix computed from the data before ambiguities were inserted. The fourth graph shows the Robinson-Foulds distance between the model tree and the Neighbor Joining trees computed from the different distance matrices. To get statistically sound results the average was taken over 20 runs. fastdist is the combined method described in Figure 3. noresolve is the general technique without nearest neighbor resolution. Swofford is the method suggested by Swofford. Phylip is the Output from the Phylip package which uses Felsenstein's approach. original represents the accuracy of NJ on the data before the ambiguities were inserted (notice how NJ for some cases is more accurate on data with ambiguities).