Talk:Main Page

From MrBayes

Can someone please explain what "standard deviation of split frequencies" actually is. Okay, I get the standard deviation part, but what is "split frequencies"? How is it calculated? Does it factor in all of the estimated parameters, or only the -lnL? Is there an accepted cut-off for terminating the analysis? In other words, how does SDSF correlate to stationarity? Should you just wait until the SDSF gets to zero? How long should it *stay* at zero before you terminate the analyis? The tutorial says to set your burnin at 25% of your total trees once you stop the analysis, but this seems rather arbitrary. How is this better than plotting your -lnL and other parameters against the ngen to determine burnin?

Thanks for any enlightenment.