Reader Comments

Post a new comment on this article

treedater

Posted by evolz on 23 Jan 2020 at 16:52 GMT

The finding that treedater is not robust to autocorrelated rates is much appreciated. I have reproduced this finding using the latest version of treedater. However I would also note that
1) the distribution of errors is highly skewed,
2) the median error in ABR scenarios is closer to the other methods,
3) in a real analysis, the outlying rates would likely raise red flags. I find that most of the error can be resolved by using the 'meanRateLimits' option in treedater. Constraining the value to within 2x of the true rate gives much better results.

No competing interests declared.