06375022rely on an acute visual system for orientation, feeding and surviving. Therefore these 29features are homoplastic because both groups need them for the same purpose/function, i.e. analogous.
Maximum Likelihood (ML)
ML methods consistently outperform distance methods and MP by being the mostasymptotic consistency (Lartillot & Philippe, 2004), meaning that the method willapproach a constant (the asymptote) continuously and therefore is reliable. Consistencywas proposed by Felsenstein and is the ability to continually arrive at the same value asmore data is added. However ML does require heavy calculations due to comparison of numerous alternative hypotheses (Swofford
, 1996).Various different models exist that vary in the number of different substitution typesincorporated. As a result ML methods are highly dependent on the model they use, if the model is wrong the results could be wrong. This is not to say that a perfect model isrequired, a good result can come from a good model provided it uses the mostimportant evolutionary parameters involved for that sequence (Swofford
, 1996).But this requires knowing what type of model to use and what evolutionary parametersare important. For example is a simple Juke Cantor model good enough to explainnucleotide substitution or is the more complex GTR (general time reversible model)required. The choice of the ‘right’ model requires a goodness of fit χ
statistic whichcompares observed data with expected from model prediction. It’s important to use amodel that fits the data and that isn’t just comparably better than other models(Swofford
, 1996).The models are based on a Markov chain and as a consequence one of theassumptions is that the nucleotide sites evolve independently at the same rate. This cancause an underestimation of the number of multiple changes and similar to MP can bepositively misleading. However this can be corrected for by adding a new relative rate