25 



1989). Sanderson (1989) has indicated that this is not likely the case for most systematic 

 studies, so bootstrap frequencies are probably not estimates of the true confidence 

 intervals. The use of replacement during sampling may also artificially increase character 

 non-independence by allowing the same character to be sampled more than once (L.R. 

 Linton pers. comm.). As well, the bootstrap can become problematic when parsimony is 

 used to estimate phylogeny and rates of evolution in the various lineages are greatly 

 unequal (Felsenstein 1985). Thus, the bootstrap solution may differ from the most 

 parsimonious one. a difference that arises from the fact that the bootstrap represents a 

 phylogeny estimated from repeated samplings and not the real one (Felsenstein 1985). and 

 from the properties of consensus trees, of which the bootstrap solution is one (Swofford 

 1993). 



These problems seem to become detrimental to the analysis when more than two topologies 

 are possible (as is usually the case in phylogenetic studies), prompting some algorithmic 

 or procedural corrections (Hall & Martin 1988; Rodrigo 1993: Li & Zharkikh 1994. 1995). 

 However, the only "correction" that we have heeded is Hedges's (1992) suggestion that 

 most studies involving the bootstrap do not use enough replications, with at least 500 

 replications being required to ensure that the bootstrap frequency is within one percent of 

 the 95% confidence interval. In recognition of all of these difficulties and the varying 

 opinions as to the utility of the bootstrap (see Felsenstein & Kishino 1993; Hillis & Bull 

 1993), bootstrap frequencies are interpreted here as rough indicators of support for the 

 various nodes of the cladogram, and not as true confidence intervals. 

 Herein, 1,000 bootstrap replicates were conducted using the heuristic search option of 

 PAUR Heuristic searches were identical to that detailed above except that taxa were added 

 with the CLOSE algorithm with HOLD = 10, and with only 100 MAXTREES allowed 

 for each replication. Only the 168 included characters were sampled, and with equal 

 probability (i.e., their inverse weights were not used to designate repeat counts of a 

 character). "Irrelevant" characters (primarily autapomorphies here) were retained with the 

 suggestion that they do not adversely affect bootstrap results (Harshman 1994). 



Permutation tail probabilities (PTP) (Archie 1989; Faith & Cranston 1991) 

 The PTP test seeks to assess the degree of phylogenetic structure in a data set based on 

 the amount of cladistic covariation between its characters, as compared to a matrix that 

 possesses random covariation. A data set with an associated solution that is shorter than 

 a statistically significant proportion of those derived from a number of random data sets 

 (e.g., by being within the lower fifth percentile of tree length) is said to possess "significant 

 cladistic structure" (Faith & Cranston 1991). Random data sets are constructed from the 

 original by randomly permutating character states between the included taxa within each 

 character. Outgroup taxa are excluded from this process to maintain polarity assessments. 

 Thus, each random data set maintains most of the characteristics of the original. 

 Several methods exist to assess the level of significance of a PTP test. The simplest is the 

 PTP statistic which is defined as the proportion of all data sets (original and random) that 

 produce a tree as short or shorter than that derived from the original data set (Faith & 

 Cranston 1991). A critical length value corresponding to the desired level of significance 



