|
|
Alex
Thanks for this nasty one. I spent two hours just trying to get this sorted out. I found that the problem is not of bug as much as that of bias. Because many ratios are incalculable (many pairs have no tranversions and even transitions), many ratios are invalid and their average compounds the problem. This happens under all distance methods for estimating R. Therefore, a better way of dealing with this is to compute avg. ts and avg tv, and then take the ratio. In fact, even that is not necessary if you were to use our new method (Maximum composite likelihood [MCL], which will be included in the new version soon). I will update the help file and caution people to not use average R approach (instead, they should employ the alternatives).
In a future version, I might disable this option and only provide MCL based R estiamtes.
Sudhir
|
|