In article <x7vikqqk5r.fsf at darwin.darwin.eeb.uconn.edu>,
Kent E. Holsinger <Kent at Darwin.EEB.UConn.Edu> wrote:
>>>>>> "Joe" == Joe Felsenstein <joe at evolution.genetics.washington.edu>
>writes: In article <4giqf9$m06 at nntp3.u.washington.edu>
>joe at evolution.genetics.washington.edu (Joe Felsenstein) writes:
>> Joe> In likelihood methods (such as Ziheng Yang's methods that use
> Joe> gamma-distributed rates in his PAML package, or my Hidden
> Joe> Markov Model approach in DNAML) one can change the rate
> Joe> variation parameters until the overall likelihood of the tree
> Joe> is maximized. With distance methods, I am not as sure how to
> Joe> do this. Perhaps one could try various values until the
> Joe> goodness of fit of the tree (say, the sum of squares) is
> Joe> optimized, but doing this may have biases in it.
>>I don't think comparing goodness-of-fit values would work, unless they
>were normalized in some way. The distance between sequences A and B
>increases as the coefficient of variation increases, so the total tree
>length would also increase as the coefficient of variation
>increases. It's not obvious to me that there is a scale-invariant way
>of comparing the goodness-of-fit statistics, but maybe I'm missing
>something.
I agree for the distance measures, and that was why I was worried about
biases in going this for them. But for maximum likelihood methods it
works well, and has no known problems.
--
Joe Felsenstein joe at genetics.washington.edu (IP No. 128.95.12.41)
Dept. of Genetics, Univ. of Washington, Box 357360, Seattle, WA 98195-7360 USA
