Beside the fact that maximum parsimony is computationally cheap, what other good arguments are there for it? Is there any model behind this principle? Why would one expect this principle to provide right phylogeny in any situation at all ?
The idea of parsimony is that when constructing phylogenetic trees a simple hypothesis (e.g., four evolutionary changes are necessary to connect two taxa) is more likely to be true than a more complex hypothesis (e.g., 15 evolutionary changes) (source: Berkeley University).
The reason is that for a certain taxon to evolve, there must be a certain number of evolutionary changes. In the end, this happens by more or less stochastic processes (Dingli & Pacheco, 2011) that randomly generate genetic changes (mutations). The characteristics of better fit individuals survive, while those of the least fit are terminated from the gene pool. Hence, every additional change that must be adopted to create a new taxon from another means that the chance that that happens decreases exponentially; it's all a matter of statistics. The less the number of changes needed, the higher the chance that that might have actually happened.
- Dingli & Pacheco, BMC Biology (2011);9:41
Generally speaking, parsimony consists in not making unnecessary ad hoc hypotheses.
In the case of phylogeny, parsimony can be used at two stages: when inferring the number of character changes for a given topology and then when choosing among topologies.
One does not expect parsimony to provide the right phylogeny in any situation. It is just the most sensible we can do when no further information about evolutionary processes is available. This is why parsimony is a popular choice when it comes to deal with morphological data: models of character evolution are not as easy to devise for morphological data as for molecular data.
In the case of phylogeny based on molecular data such as protein or DNA sequences, models of evolutionary changes exist, which are not based on ad hoc hypotheses, but on reasonably good knowledge gathered from disciplines such as molecular biology, biochemistry, etc. Under these models, "hidden" multiple character changes are expected along long enough branches. This may lead to differences with respect to the most simple parsimony criterion, when inferring the number of character changes in a topology.
In such a case, simple parsimony may not be the most sensible approach to phylogeny resconstruction.