User:R. Eric Collins/MBL/Support

From OpenWetWare
Jump to: navigation, search

Branch Support and Tests

  • Bootstrap
    • hat ^ signifies observed
    • not required to use majority-rule, sometimes 'rogue' taxa jump around and ruin support
    • tree/bootstrap as 'summary' of results (i.e. not the end all and be all)
    • asterisk * refers to boostrapped value
  • CAN'T USE KH TEST TO COMPARE AN A PRIORI TREE TO YOUR BEST ML TREE
  • SH test
    • if your model is too simple, you can overestimate branch support
    • in PAUP, include *all* trees for SHtest, not just 2 you're interested in because it takes all of them into account

p-value== how often do you get support for the alternate when the null hypothesis is true?

    • bootstrap proportion > 0.95 can sometimes be used as equivalence to 0.05 p-value
    • but... curved boundaries in treespace invalidate D ~ 1 - BP relationship
    • triple-point (star tree) corrected by Susko 2010
    • un-adjusted (Felsenstein) bootstraps tend to UNDERESTIMATE on a priori tree
    • for highly-supported branches (~ >70%), bootstrap proportions actually work quite well as estimates for p-values
    • things based on bootstraps are more robust to changes in model