Parsimony, likelihood, and simplicity

The latest charge against parsimony in phylogenetic inference is that it involves estimating too many parameters. The charge is derived from the fact that, when each character is allowed a branch length vector of its own (instead of the homogeneous branch lengths assumed in current likelihood models), the results for likelihood and parsimony are identical. Parsimony, however, can also be derived from simpler models, involving fewer parameters. Therefore, parsimony provides (as many authors had argued before) the simplest explanation of the data, or the most realistic, depending on one's views. If (as argued by likelihoodists) phylogenetic inference is to use the simplest model that provides sufficient explanation of the data, the starting point of phylogenetic analyses should be parsimony, not maximum likelihood. If the addition of new parameters (which increase the likelihood) to a parsimony estimation is seen as desirable, this may lead to a preference for results based on current likelihood models. If the addition of parameters is continued, however, the results will eventually come back to the same place where they had started, since allowing each character a branch length of its own also produces parsimony. Parsimony can be justified by very different types of models—either very complex or very simple. This suggests that parsimony does have a unique place among methods of phylogenetic estimation.     

Parsimony and explanatory power

Parsimony can be related to explanatory power, either by noting that each additional requirement for a separate origin of a feature reduces the number of observed similarities that can be explained as inheritance from a common ancestor; or else by applying Popper’s formula for explanatory power together with the fact that parsimony yields maximum likelihood trees under No Common Mechanism (NCM). Despite deceptive claims made by some likelihoodists, most maximum likelihood methods cannot be justified in this way because they rely on unrealistic background assumptions. These facts have been disputed on the various grounds that ad hoc hypotheses of homoplasy are explanatory, that they are not explanatory, that character states are ontological individuals, that character data do not comprise evidence, that unrealistic theories can be used as background knowledge, that NCM is unrealistic, and that likelihoods cannot be used to evaluate explanatory power. None of these objections is even remotely well founded, and indeed most of them do not even seem to have been meant seriously, having instead been put forward merely to obstruct the development of phylogenetic methods. © The Willi Hennig Society 2008.     

Likelihood, parsimony, and heterogeneous evolution

Evolutionary rates vary among sites and across the phylogenetic tree (heterotachy). A recent analysis suggested that parsimony can be better than standard likelihood at recovering the true tree given heterotachy. The authors recommended that results from parsimony, which they consider to be nonparametric, be reported alongside likelihood results. They also proposed a mixture model, which was inconsistent but better than either parsimony or standard likelihood under heterotachy. We show that their main conclusion is limited to a special case for the type of model they study. Their mixture model was inconsistent because it was incorrectly implemented. A useful nonparametric model should perform well over a wide range of possible evolutionary models, but parsimony does not have this property. Likelihood-based methods are therefore the best way to deal with heterotachy.     

Outgroup misplacement and phylogenetic inaccuracy under a molecular clock--a simulation study

We conducted a simulation study of the phylogenetic methods UPGMA, neighbor joining, maximum parsimony, and maximum likelihood for a five-taxon tree under a molecular clock. The parameter space included a small region where maximum parsimony is inconsistent, so we tested inconsistency correction for parsimony and distance correction for neighbor joining. As expected, corrected parsimony was consistent. For these data, maximum likelihood with the clock assumption outperformed each of the other methods tested. The distance-based methods performed marginally better than did maximum parsimony and maximum likelihood without the clock assumption. Data correction was generally detrimental to accuracy, especially for short sequence lengths. We identified another region of the parameter space where, although consistent for a given method, some incorrect trees were each selected with up to twice the frequency of the correct (generating) tree for sequences of bounded length. These incorrect trees are those where the outgroup has been incorrectly placed. In addition to this problem, the placement of the outgroup sequence can have a confounding effect on the ingroup tree, whereby the ingroup is correct when using the ingroup sequences alone, but with the inclusion of the outgroup the ingroup tree becomes incorrect.     

Statistical consistency and phylogenetic inference: a brief review

The claim that parsimony can be statistically inconsistent remains the chief criticism of the cladistic approach, and also the main justification for alternative model‐based approaches such as maximum likelihood and Bayesian inference. Despite its refutation in the 1980s, this persistent myth of parsimony's Achilles’ heel is entrenched in the primary literature, and has metastasized into textbooks, as well. Here, I review historical controversies, and offer three short arguments as to why statistical consistency is not only irrelevant to systematics, but to empirical science in general.     

Advantages and Disadvantages of Molecular Phylogenetics: A Case Study of Ascaridoid Nematodes

The advantages of nucleotide sequence data for studying phylogeny have been shown to include number of potential characters available for comparison, rate independence between molecular and morphological evolution, and utility of molecular data for modeling patterns of nucleotide substitution. Potential pitfalls have also been revealed and include difficulties of inferring positional homology, incongruence between organismal and gene genealogies, and low likelihood of recovering the correct phylogeny given certain patterns in the timing of speciation events. Statistical methods for comparing phylogenetic hypotheses have been used to assess the reliability of alternative trees for ascaridoid nematodes. Based on partial ribosomal RNA sequences, tree topologies inconsistent with monophyly of the Ascaridinae were significantly worse by maximum likelihood inference. The topology of the maximum parsimony tree based on full-length sequences of 18S rRNA and 300 nucleotides of Cytochrome oxidase II for 13 ascaridoid species was generally consistent with traditional taxonomic expectations at lower ranks, but inconsistent with most proposed arrangements at higher taxonomic levels.     

Evaluating the clade size effect in alternative measures of branch support

The clade size effect refers to a bias that causes middle‐sized clades to be less supported than small or large‐sized clades. This bias is present in resampling measures of support calculated under maximum likelihood and maximum parsimony and in Bayesian posterior probabilities. Previous analyses indicated that the clade size effect is worst in maximum parsimony, followed by maximum likelihood, while Bayesian inference is the least affected. Homoplasy was interpreted as the main cause of the effect. In this study, we explored the presence of the clade size effect in alternative measures of branch support under maximum parsimony: Bremer support and symmetric resampling, expressed as absolute frequencies and frequency differences. Analyses were performed using 50 molecular and morphological matrices. Symmetric resampling showed the same tendency that bootstrap and jackknife did for maximum parsimony and maximum likelihood. Few matrices showed a significant bias using Bremer support, presenting a better performance than resampling measures of support and comparable to Bayesian posterior probabilities. Our results indicate that the problem is not maximum parsimony, but resampling measures of support. We corroborated the role of homoplasy as a possible cause of the clade size effect, increasing the number of random trees during the resampling, which together with the higher chances that medium‐sized clades have of being contradicted generates the bias during the perturbation of the original matrix, making it stronger in resampling measures of support.     

Reconstructing phylogenies from allozyme data: comparing method performance with congruence

Allozyme data are widely used to infer the phylogenies of populations and closely-related species. Numerous parsimony, distance, and likelihood methods have been proposed for phylogenetic analysis of these data; the relative merits of these methods have been debated vigorously, but their accuracy has not been well explored. In this study, I compare the performance of 13 phylogenetic methods (six parsimony, six distance, and continuous maximum likelihood) by applying a congruence approach to eight allozyme data sets from the literature. Clades are identified that are supported by multiple data sets other than allozymes (e.g. morphology, DNA sequences), and the ability of different methods to recover these 'known' clades is compared. The results suggest that (1) distance and likelihood methods generally outperform parsimony methods, (2) methods that utilize frequency data tend to perform well, and (3) continuous maximum likelihood is among the most accurate methods, and appears to be robust to violations of its assumptions. These results are in agreement with those from recent simulation studies, and help provide a basis for empirical workers to choose among the many methods available for analysing allozyme characters.     

A likelihood approach to character weighting and what it tells us about parsimony and compatibility

The statistical framework of maximum likelihood estimation is used to examine character weighting in inferring phylogenies. A simple probabilistic model of evolution is used, in which each character evolves independently among two states, and different lineages evolve independently. When different characters have different known probabilities of change, all sufficiently small, the proper maximum likelihood method of estimating phylogenies is a weighted parsimony method in which the weights are logarithmically related to the rates of change. When rates of change are taken extremely small, the weights become more equal and unweighted parsimony methods are obtained. When it is known that a few characters have very high rates of change and the rest very low rates, but it is not known which characters are the ones having the high rates, the maximum likelihood criterion supports use of compatibility methods. By varying the fraction of characters believed to have high rates of change one obtains a ‘threshold method’ whose behavior depends on the value of a parameter. By altering this parameter the method changes smoothly from being a parsimony method to being a compatibility method. This provides us with a spectrum of intermediates between these methods. These intermediate methods may be of use in analysing real data.     

A likelihood approach to character weighting and what it tells us about parsimony and compatibility

The statistical framework of maximum likelihood estimation is used to examine character weighting in inferring phylogenies. A simple probabilistic model of evolution is used, in which each character evolves independently among two states, and different lineages evolve independently. When different characters have different known probabilities of change, all sufficiently small, the proper maximum likelihood method of estimating phylogenies is a weighted parsimony method in which the weights are logarithmically related to the rates of change. When rates of change are taken extremely small, the weights become more equal and unweighted parsimony methods are obtained.
When it is known that a few characters have very high rates of change and the rest very low rates, but it is not known which characters are the ones having the high rates, the maximum likelihood criterion supports use of compatibility methods. By varying the fraction of characters believed to have high rates of change one obtains a 'threshold method' whose behavior depends on the value of a parameter. By altering this parameter the method changes smoothly from being a parsimony method to being a compatibility method. This provides us with a spectrum of intermediates between these methods. These intermediate methods may be of use in analysing real data.     

Dynamic homology and the likelihood criterion

The use of likelihood as an optimality criterion is explored in the context of dynamic homology. Simple models and procedures are described to allow the analysis of large variable length sequence data sets, alone and in combination with qualitative information (such as morphology). Several approaches are discussed that have different likelihood interpretations in terms of maximum parsimony likelihood and maximum average likelihood. Implementation is discussed and an example in arthropod systematics presented. Topological congruence comparisons with parsimony are made.
© The Willi Hennig Society 2006.     

Revisiting an Equivalence Between Maximum Parsimony and Maximum Likelihood Methods in Phylogenetics

Tuffley and Steel (Bull. Math. Biol. 59:581–607, 1997) proved that maximum likelihood and maximum parsimony methods in phylogenetics are equivalent for sequences of characters under a simple symmetric model of substitution with no common mechanism. This result has been widely cited ever since. We show that small changes to the model assumptions suffice to make the two methods inequivalent. In particular, we analyze the case of bounded substitution probabilities as well as the molecular clock assumption. We show that in these cases, even under no common mechanism, maximum parsimony and maximum likelihood might make conflicting choices. We also show that if there is an upper bound on the substitution probabilities which is ‘sufficiently small’, every maximum likelihood tree is also a maximum parsimony tree (but not vice versa).     

Molecular systematics of the Eastern Fence Lizard (Sceloporus undulatus): a comparison of Parsimony, Likelihood, and Bayesian approaches

Phylogenetic analysis of large datasets using complex nucleotide substitution models under a maximum likelihood framework can be computationally infeasible, especially when attempting to infer confidence values by way of nonparametric bootstrapping. Recent developments in phylogenetics suggest the computational burden can be reduced by using Bayesian methods of phylogenetic inference. However, few empirical phylogenetic studies exist that explore the efficiency of Bayesian analysis of large datasets. To this end, we conducted an extensive phylogenetic analysis of the wide-ranging and geographically variable Eastern Fence Lizard (Sceloporus undulatus). Maximum parsimony, maximum likelihood, and Bayesian phylogenetic analyses were performed on a combined mitochondrial DNA dataset (12S and 16S rRNA, ND1 protein-coding gene, and associated tRNA; 3,688 bp total) for 56 populations of S. undulatus (78 total terminals including other S. undulatus group species and outgroups). Maximum parsimony analysis resulted in numerous equally parsimonious trees (82,646 from equally weighted parsimony and 335 from weighted parsimony). The majority rule consensus tree derived from the Bayesian analysis was topologically identical to the single best phylogeny inferred from the maximum likelihood analysis, but required approximately 80% less computational time. The mtDNA data provide strong support for the monophyly of the S. undulatus group and the paraphyly of "S. undulatus" with respect to S. belli, S. cautus, and S. woodi. Parallel evolution of ecomorphs within "S. undulatus" has masked the actual number of species within this group. This evidence, along with convincing patterns of phylogeographic differentiation suggests "S. undulatus" represents at least four lineages that should be recognized as evolutionary species.     

A LIKELIHOOD JUSTIFICATION OF PARSIMONY

Abstract— A connection is established between maximally parsimonious cladograms and trees of highest likelihood. The assumptions needed to prove this are derivable from the structure of evolutionary theory and are independent of the frequency of homoplasy. The bearing of this justification on alternative methods of phylogenetic inference and on Felsenstein's (1978) proof that parsimony and other phylogenetic methods can be statistically inconsistent is discussed.     

Frequent inconsistency of parsimony under a simple model of cladogenesis

Although the conditions under which the parsimony method becomes inconsistent have been studied for almost two decades, the probability that the parsimony method would encounter conditions causing inconsistency under simple models of cladogenesis is unknown. Here, we examine the statistical behavior of the parsimony method under a birth-death model of cladogenesis, when the molecular clock holds. The parsimony method can become inconsistent a high proportion of the time even under this simple model of cladogenesis. When taxon sampling is poor or rates of evolution are high, the probability that parsimony will become inconsistent increases.     

Probabilistic methods outperform parsimony in the phylogenetic analysis of data simulated without a probabilistic model

To understand patterns and processes of the diversification of life, we require an accurate understanding of taxon interrelationships. Recent studies have suggested that analyses of morphological character data using the Bayesian and maximum likelihood Mk model provide phylogenies of higher accuracy compared to parsimony methods. This has proved controversial, particularly studies simulating morphology‐data under Markov models that assume shared branch lengths for characters, as it is claimed this leads to bias favouring the Bayesian or maximum likelihood Mk model over parsimony models which do not explicitly make this assumption. We avoid these potential issues by employing a simulation protocol in which character states are randomly assigned to tips, but datasets are constrained to an empirically realistic distribution of homoplasy as measured by the consistency index. Datasets were analysed with equal weights and implied weights parsimony, and the maximum likelihood and Bayesian Mk model. We find that consistent (low homoplasy) datasets render method choice largely irrelevant, as all methods perform well with high consistency (low homoplasy) datasets, but the largest discrepancies in accuracy occur with low consistency datasets (high homoplasy). In such cases, the Bayesian Mk model is significantly more accurate than alternative models and implied weights parsimony never significantly outperforms the Bayesian Mk model. When poorly supported branches are collapsed, the Bayesian Mk model recovers trees with higher resolution compared to other methods. As it is not possible to assess homoplasy independently of a tree estimate, the Bayesian Mk model emerges as the most reliable approach for categorical morphological analyses.     

Distances that perfectly mislead

Given a collection of discrete characters (e.g., aligned DNA sites, gene adjacencies), a common measure of distance between taxa is the proportion of characters for which taxa have different character states. Tree reconstruction based on these (uncorrected) distances can be statistically inconsistent and can lead to trees different from those obtained using character-based methods such as maximum likelihood or maximum parsimony. However, in these cases the distance data often reveal their unreliability by some deviation from additivity, as indicated by conflicting support for more than one tree. We describe two results that show how uncorrected (and miscorrected) distance data can be simultaneously perfectly additive and misleading. First, multistate character data can be perfectly compatible and define one tree, and yet the uncorrected distances derived from these characters are perfectly treelike (and obey a molecular clock), only for a completely different tree. Second, under a Markov model of character evolution a similar phenomenon can occur; not only is there statistical inconsistency using uncorrected distances, but there is no evidence of this inconsistency because the distances look perfectly treelike (this does not occur in the classic two-parameter Felsenstein zone). We characterize precisely when uncorrected distances are additive on the true (and on a false) tree for four taxa. We also extend this result to a more general setting that applies to distances corrected according to an incorrect model.     

A falsificationist perspective on the usage of process frequencies in phylogenetics

By referring to Popperian falsificationism, proponents of cladistic parsimony claim the superiority of parsimony over likelihood. They conclude that likelihood as a statistical approach is inconsistent with falsificationism, and base their argumentation on four claims: (1) congruence tests cladograms against observational evidence and represents the most important test in phylogenetics, in which minimum‐step trees represent most corroborated trees; (2) frequency probabilities cannot be used for evaluating degree of corroboration; (3) phylogeny represents a unique process and thus frequencies cannot be applied as they require statistical reference classes that are necessarily general; (4) likelihood is a verificationist approach. After discussing the deficiencies of the cladistic phylogeneticists’ conceptualisation of the congruence test and the presentation of an alternative conceptualisation, it is shown that these four claims cannot be sustained within a falsificationist framework, and that the weighting of characters is a necessity. A differentiation between the theoretical concept of apomorphy and the epistemological concept of character weight is proposed. While apomorphies have to be independent from each other, the weighting of characters is interdependent due to human inability to distinguish organismic traits that are structurally identical though they do not share a common evolutionary origin. The possibility of this epistemological interdependence can best be dealt with by the application of process frequencies. The importance of process frequencies of specific transformation classes is exemplified in reference to Popper's formula for the measure of degree of corroboration and its consistency is shown. Therefore, the application of statistical methods is reasonable. As a consequence, the question whether likelihood or parsimony methods represent the best approaches in phylogenetics remains a genuinely empirical question that cannot be decided only in reference to Popper's falsificationism.     

以雾灵山夜蛾科为材料评估进化模型对DNA条码分类的影响

为了探究进化模型对DNA条形码分类的影响, 本研究以雾灵山夜蛾科44个种的标本为材料, 获得COI基因序列。使用邻接法（neighbor-joining）、 最大简约法（maximum parsimony）、 最大似然法（maximum likelihood）以及贝叶斯法（Bayesian inference）构建系统发育树, 并且对邻接法的12种模型、 最大似然法的7种模型、 贝叶斯法的2种模型进行模型成功率的评估。结果表明, 邻接法的12种模型成功率相差不大, 较稳定; 最大似然法及贝叶斯法的不同模型成功率存在明显差异, 不稳定; 最大简约法不基于模型, 成功率比较稳定。邻接法及最大似然法共有6种相同的模型, 这6种模型在不同的方法中成功率存在差异。此外, 分子数据中存在单个物种仅有一条序列的情况, 显著降低了模型成功率, 表明在DNA条形码研究中, 每个物种需要有多个样本。     

Should we be worried about long-branch attraction in real data sets? Investigations using metazoan 18S rDNA

Although long-branch attraction (LBA) is frequently cited as the cause of anomalous phylogenetic groupings, few examples of LBA involving real sequence data are known. We have found several cases of probable LBA by analyzing subsamples from an alignment of 18S rDNA sequences for 133 metazoans. In one example, maximum parsimony analysis of sequences from two rotifers, a ctenophore, and a polychaete annelid resulted in strong support for a tree grouping two "long-branch taxa" (a rotifer and the ctenophore). Maximum-likelihood analysis of the same sequences yielded strong support for a more biologically reasonable "rotifer monophyly" tree. Attempts to break up long branches for problematic subsamples through increased taxon sampling reduced, but did not eliminate, LBA problems. Exhaustive analyses of all quartets for a subset of 50 sequences were performed in order to compare the performance of maximum likelihood, equal-weights parsimony, and two additional variants of parsimony; these methods do differ substantially in their rates of failure to recover trees consistent with well established, but highly unresolved phylogenies. Power analyses using simulations suggest that some incorrect inferences by maximum parsimony are due to statistical inconsistency and that when estimates of central branch lengths for certain quartets are very low, maximum-likelihood analyses have difficulty recovering accepted phylogenies even with large amounts of data. These examples demonstrate that LBA problems can occur in real data sets, and they provide an opportunity to investigate causes of incorrect inferences.