BROADLY CONSISTENT WATCH I

I'm starting a new tradition here, the "Broadly Consistent Watch." If you see that headline, you can be sure I'll be noting an abuse of the term "broadly consistent" --- indeed, in most cases, I'll be pointing out the use of the term for things that are actually not consistent at all.

Here's the first edition, from Kivisild et al. (2005:10) (also discussed in a previous post):

The coalescent date of the human mitochondrial DNA tree using this rate is 160,000 (S.D. 22,000) years. This coalescent date is broadly consistent with the dates of the Homo sapiens fossils recognized so far from Ethiopia (CLARK et al. 2003; MCDOUGALL et al. 2005; WHITE et al. 2003).

This is an excellent example of the au courant use of the term. Here, the paper shows its familiarity with the recent literature on fossil hominids, correctly citing the recent Omo Kibish dates and Herto fossils. And indeed the Herto fossils are dated to between 154,000 and 160,000 years ago, and the Omo Kibish hominids between 190,000 and 200,000 years, so these "early modern" humans do appear to be "broadly consistent" with the mtDNA coalescence estimate.

But that's the beauty of "broadly consistent": it can apply to anything within a ballpark or two (or four), especially if (a) you're talking about data from another field, and (b) you don't look too closely at the numbers.

It's so tempting just to say "broadly consistent" and let the minds of the readers connect the dots: "Aha! It proves the theory! This can't be a mere coincidence! The dates are broadly consistent!" It's so tempting almost no one can resist using it from time to time.

Let's look more closely at these "broadly consistent" dates. First of all, the Omo Kibish hominids simply fall outside the standard error of the mtDNA date. They're not "broadly consistent" at all --- if anything, they appear to be inconsistent, although they probably are close enough to be within a 95 percent confidence interval (if it were reported, which it isn't).

That assumes that the important thing is for the dates to be the same. But if the human mtDNA type supposedly came from the population represented by these Ethiopian Homo sapiens fossils, then its variation must coalesce before these fossils. The same date is not evidence for consistency; a consistent date would be earlier. How much earlier depends on the demography, but 10 or 20 thousand years would seem like a bare minimum.

And then there's the "hotspot" problem that is the subject of the Kivisild et al (2005) paper. The 160,000 year estimate assumes equality of rates among sites, but the data indicate that some sites mutate much more frequently than others, and repeatedly during human evolution. If these sites mutate more rapidly and have saturated on the human lineage compared to chimpanzees, then the 160,000 year date should be an overestimate because humans should have more variation than expected from the long-term evolutionary comparisons. The data do not indicate how extensive this overestimate may be, but it makes the coalescence less consistent with the dates of the fossils, not more.

Now, can we say in this case that the dates are really not "broadly consistent"? No, indeed we can't. There are just too many sources of error in the genetic estimate to say whether it might be within the range of possible mtDNA ancestors of these Ethiopian fossils. The date could be as high as 210,000 - 220,000 years, if the mutation rate has been overestimated (e.g., if many rare sites that currently segregate are in fact selected). From that perspective, the dates are "broadly consistent" with every event in the Late Pleistocene.

But that's far from a vote of confidence. It is not a significant coincidence; it is the overlap of uncertainty. And that's usually what "broadly consistent" means.

References:

Kivisild T et al. 2005. The role of selection in the evolution of human mitochondrial genomes. Genetics (online before print).