The Information Science of Dog Phylogeny – Naish’s Important Opinion

In 2006 I read an interesting blog posting by Darren Naish, on the subject of dog ancestry, which was good but was the last good thing I’ve seen him say on it. Now he’s undoing what good he did there.

The original posting revealed the idea that dogs were not from wolves, but from… something else. Then as now, the main supporter for this idea was Janis Koler-Matznick (though she didn’t instigate it). I do appreciate Darren’s airing of interesting issues like this, because new ideas are not only entertaining but the essence of science. Interestingly, he claims now that it was a mistake on his part to have supported the new idea, but he actually didn’t. I recollect him going no further than saying it was interesting. That’s fine and I wish more scientists did that, though of course the best experts will not only put interesting ideas before us but tell us which are right… and end up being right themselves.

It is not widely enough appreciated that most people take their opinions on scientific issues from others, without trying to think things through for themselves. This is how pseudo-science proliferates. In the case of dog ancestry, if they try to work it out at all, people usually assume it’s right to find the nearest wild animal alive today, and assume that was the ancestor.

Obviously that rule doesn’t work because some of our most common domesticated mammals don’t have a never-tamed ancestor that is still extant in their original habitat (dromedaries) or at all (horses, cattle). That’s not what’s happening here though, because the new model ancestor still is alive today in its original habitat.

Another problem is the way a complex genetic analysis can convince people, just by its complexity, that an idea is right. They’re tricky enough for someone with a background in the information sciences to have to work hard to see through, but anyone who has no expertise at all in the field will not be making any meaningful conclusions worth listening to. I’ve put some reasonable effort and time into getting to the bottom of the dog papers, and reasons not to be cheerful about their conclusions are legion:

1: The vast majority of them don’t even claim to choose the right ancestor, but instead just look into the inter-relationships between dog breeds. Many just try to find out when dogs were domesticated. The actual number of papers making a decent fist of investigating the ancestor is tiny. The best I’ve seen is Lindblad-Toh et al (2005), and it significantly disagrees with the Tedford et al. (2009) tree on relationships between wild canids. Although Lindblad-Toh et al (2005) does look quite convincing, remember, they have selectively sampled DNA, and selecting sections of dog/wolf DNA confused by interbreeding would corrupt the result. Remember also that they would not print any tree that did not give the wolf as the ancestor, but that they could get the tree they wanted by choosing the right bits of DNA.

2: Even when they do claim to be pursuing the wild ancestor or seem to want to give the impression that they are, they almost always simply start with the assumption that wolves are dogs’ ancestors! Er, No. To choose the best theory you explicitly line up the competing theories and see which best predicts the observations. Not to do so but to claim you have been looking for the best theory is scientific fraud. Producing a family tree where the true ancestor is not or might not be included is a waste of time and money when only one person is working on it, let alone dozens. Remember, a cladogram is an undirected binary tree until you announce which root – ancestor – you have magically chosen – which is where half the error is often inserted. I mentioned the habit of running a competition where only one competitor was allowed, to some co-authors of a paper with dozens of authors, and one actually deigned to reply. Cheekily, they said that Janis Koler-Matznick’s paper was not very “complete”! Well, her competition with two competitors was always going to be more complete than his with just one! The nearest his got to “complete” was being complete rubbish.

3: Pretending to do a comparative experiment on multiple theories when you’re actually honing a single theory you assume to be right is one deceitful error. But some experimenters are just confusing themselves with their sums. The last one I looked at didn’t actually generate a cladogram, but investigated using a nearest-neighbour analysis. Cladograms are potentially a very bad form of investigation, but all the others are… well, you need a good reason for using something else. Maybe a tree not allowing any cross-breeding between branches is unrealistic when all the members can interbreed, but that wasn’t the only problem: they also felt the need to use some esoteric statistical metric for the first time, which was too complex to explain in the paper, and looked to me like a perfect pot for brewing statistical mumbo-jumbo. It’s not just that people like Darren Naish are confused and misled by the statistical confuscations, but the experiementers themselves don’t understand what they’re doing wrong. And we can be certain that something is going horribly wrong somewhere because all the estimates of the time of dog domestication show no signs of converging anytime soon. Check the end of this posting detailing the unobviousness of it all, especially date disagreements. If it’s so obvious, why do all the experimental results wildly disagree? Even the dinobird simplistic cladograms tend agree with each other better than this even though they’re all confused by parallelisms. Also, if the studies are as hugely inconsistent as they still are, exactly when did it become “obvious” that wolves were dog ancestors? This massive disagreement is the only way nature can get a hint out to us that the experiments are faulty. The “obviousness” became established before DNA studies, so was obviously bogus, and “dog-from-wolf” not only has never been obvious, but obviously never has been, and, actually, is probably wrong.

Something else that’s bogus is Darren Naish’s expertise in subject areas dependent on information science, which in his case, he has not got:

“All I’ll say now is that the ‘dogs are not wolves’ model is not viable in view of the mountains of molecular work now done on dogs, wolves and other canids (e.g., Savolainen et al. 2002, Lindblad-Toh et al. 2005, Pang et al. 2009, vonHoldt et al. 2010, Larson et al. 2012, Freedman et al. 2014). Seriously, it’s dead: dogs really are deeply nested within wolves,…”

Science isn’t measured in mountains but by the validity of arguments, and making the same mistakes over and over again does not add weight to an argument. However, bullshit can be measured in mountains; bear this in mind when considering the scientific validity of the dog papers, no matter how impressive they may look to the layman.

And Darren will never be worth consulting over tricky issues in bioinformatics.

DN’s original 206 blog posting.

DN’s recent view appears near the end of this posting.

Advertisements
This entry was posted in Dog evolution and tagged , , , , , . Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s