While Parsimony is derived from a philosophical realm that emphsizes the importance of generating a defensible hypothesis based on Popper's conjecture and refution, Maximum Likelihood is derived from a philosophical realm that emphasizes goodness of fit to existing models.

This may be an oversimplification due to the many derivations now available for Parsimony analyses and the minimal nature of the only computationally realistic models for Maximum Likelihood at the moment but it will become more apparent as hardware catches up with theory.

Good
 

• Probably least affected by sampling error.
• Robust even with some types of seemingly unrobust data.
• Solves the long branch problem
• Can be used in ways that fully support complex models
• Better suited to 4-state data with variable distribution of states in probability space

Bad
 

• Relies on models that are often unrealistic
• Appears robust even with some types of truly unrobust data
• Solves the long branch problem (it may not do this in a biologically meaningful way)
• Computationally difficult and time consuming
• Not really useful for non-molecular data
• Tree measures are less informative because it is generally the model and not the tree that is being tested

Neutral

• Requires a paradigm shift philosophically or it just gives other information - not phylogeny

 

 
 



Method

Basically, you calculate the probability of each base change at each postion on the tree and then add up all of the probabilties over multiple tree topologies until you find the tree that has the highest probability of matching ideal tree for the given model.