In their recent Bayesian survey of early archaeological dates from North America in Nature, Becerra-Valdivia and Higham (2020) conclude that “The data obtained show that humans were probably present before, during and immediately after the Last Glacial Maximum (about 26.5–19 thousand years ago), but that more widespread occupation began during a period of abrupt warming, Greenland Interstadial 1 (14.7–12.9 thousand years before ad 2000).”

Closer examination of their data and choices raises some very serious questions about their main conclusion. They do a good job of showing that there is a solid and rapidly growing archaeological signal in North America in the 14-15k years ago range, and that it most likely extend into the 16k range, which puts us squarely after the Last Glacial Maximum (LGM), but not during or before. The dates that the authors classify as Pre-Clovis “yielded a distribution with a peak centred on around 14,250 cal. bp, which represents the bulk of the chronometric evidence”. At this point, that is not a surprising result. The pre-LGM part is more surprising, but unfortunately much less solid.

At first glance, I have no problem with the Bayesian age estimates on the dataset used by Becerra-Valdivia and Higham. The dataset itself, from which the estimate is drawn, is more problematic. The conclusions of the authors make perfect sense if one accepts all the dates they consider archaeological. While the authors properly identify outliers within sites in their data, some of their decisions about which sites to accept and which to reject are questionable.

The earliest date in their data set, at about 26k years ago, is from the recently published Chiquihuite Cave site in Mexico (Ardelean et al 2020), which I covered here. I have to question why Becerra-Valdivia and Higham accept the earliest Chiquihuite date, while at the same time rejecting a site like Topper as “equivocal”. They even cite Waters et al.’s (2009:1309) assessment of Topper as the basis for their rejection. This is how Waters et al explain their rejection of Topper:

“The human origin of the ‘‘Topper assemblage’’ has not yet been unequivocally demonstrated. Alternatively, the pieces making up the ‘‘Topper assemblage’’ could have been produced naturally as a result of thermal fracturing (forest fires and freeze-thaw cycles) or physical fracturing during stream transport. Finally, the ‘‘Topper assemblage’’ is highly diachronous, occurring in sediments ranging  in  age  from >50,000  to 15,000 yr B.P. It is unusual that there was no lithic technological change for ca. 35,000 years.”

These are precisely the critiques leveled at the Chiquihuite claim since it was published, and which need to be addressed by the Ardelean et al. (2020). Accepting Chiquihuite while rejecting Topper on that basis, seems inconsistent to say the least, and points to some issues in data selection and quality control in the paper.

The other two very early dates they use, and that justify their estimate of an age greater than the LGM, are from “Meadowcroft Rockshelter (24,335–18,620 cal. bp) and Cactus Hill (20,585–18,970 cal. bp)”. From what I can tell the Meadowcroft date represents the possible fragment of basketry in one of the lower layers, that Adovasio and Dillehay (2020) describe as “a single element of intentionally cut birch-like (cf. Betula sp.) bark which is quite similar in overall morphology to the strips employed in all later Meadowcroft plaiting.”

As a boreal archaeologist, I collect and carry birch bark with me at all times as a survival tool while in the field. It can naturally come in very regular strips, and I would never hazard a guess as to whether a strip is intentionally cut. It would be difficult to say whether the fragment from the lower levels of Meadowcroft is naturally occurring raw material of the kind later used by people at Meadowcroft to make baskets, or whether it represents manufacture at that early date.

At this time, the archaeological status of that earliest Meadowcroft date has to be considered equivocal. It could be archaeological, or not. I would exclude that date from the dataset for now. The rest of the early and more clearly archaeological dates at Meadowcroft are much closer to the usual 14k-15k range.

There are questions also about the earliest dates and about the entire sequence at Cactus Hill. I know that some of the regular readers of this blog are Cactus Hill fans, but in Bayesian terms, I would have to assign its earliest dates a low value for prior probability. Hranicky (2010) reports that the stratigraphy is compressed and not very clear (which makes sense in a sand dune), but more importantly, that the types of artifacts dated to pre-clovis times at Cactus Hill are found in Clovis era sites elsewhere in Virginia. Until this is resolved, I don’t think Cactus Hill can be considered solid.

The three sites that pull the age estimate back toward the LGM, Chiquihuite, Meadowcroft, and Cactus Hill, all have significant problems and can’t be considered solid at this time. Used in a Bayesian procedure, I would expect them at least to be downweighted, if not excluded completely. The problem with looking for outliers within sites only, which Becerra-Valdivia and Higham explicitly acknowledge they do, is that it doesn’t tell you which sites are outliers as a whole.

I have to note that removing these highly uncertain dates from their data set takes care of one of the oddities in the paper: It moots the strange Extended Data Figure 4, which shows Becerra-Valdivia and Higham’s Out of Mexico model of archaeological expansion in North America.


Extended data Figure 4 from Becerra-Valdivia and Higham 2020, showing the Out of Mexico model

For the post LGM, pre-15k dates in their data set, this leaves us with Cooper’s Ferry and Friedkin (they strangely omit Gault from this part of the discussion), both of which I have argued (here for Coopers’ Ferry and here for Friedkin) have clear archaeological material down to the 14k range, but not earlier for now.

The rest of the sites in the list, whether one accepts them as archaeological or not (See Grayson and Meltzer 2015 for Hebior and Schaefer, for example), are well within the expected 14k-15k range or later. In other words, I don’t even need to rerun their analysis to know that my revised age estimate for the earliest archaeological material in North America would be in the 14k-15k years ago range, rather than their claimed pre-LGM estimate.

A closer look at the data set shows that the “more widespread” archaeological presence Becerra-Valdivia and Higham see starting in Greenland Interstadial 1, close to 15k years ago, is the earliest well documented age estimate. The older estimate relies on dates whose archaeological status is currently very uncertain.

As an aside, it seems to me that the second major claim in their abstract, that “widespread expansion of humans through North America was a key factor in the extinction of large terrestrial mammals,” is not well supported in the paper and is an over-reach. It seems to belong to a different paper, which I encourage them to write. The claim is further weakened if we take what I would call a more realistic approach to site selection in their data set.

They definitely document a chronological overlap between a sharp increase in archaeological signal and some of the late Pleistocene extinctions around 14k years ago, but this is already well known. In the body of the article, they soften their language to say that their result “raises the distinct possibility that widespread human expansion in population and space was a key factor in the extinction of large terrestrial mammals.” That distinct possibility still needs to be evaluated, and their paper doesn’t do that in any detail. That’s not a fault in itself.

In short, Becerra-Valdivia and Higham present a high quality, internally consistent age estimate for the oldest archaeological material in North America, but there are serious problems with their criteria for accepting or rejecting archaeological sites in their data set.


Adovasio JM and TD Dillehay 2020. Perishable Technology and the Successful Peopling of South America, PaleoAmerica, 6:3, 210-222, DOI: 10.1080/20555563.2019.1686849

Ardelean CF et al. 2020. Evidence of human occupation in Mexico around the Last Glacial Maximum Nature 584:87–92.

Becerra-Valdivia L and T Higham 2020. The timing and effect of the earliest human arrivals in North America, Nature 584:93-97.

Grayson DK and DJ Meltzer 2015. Revisiting Paleoindian exploitation of extinct North American mammals, Journal of Archaeological Science 56:177-193.

Hranicky WJ 2010. Pre-Clovis in Virginia: a matter of Antiquity, Archaeology of Eastern North America 38:53-61.

Waters MR et al 2009. Geoarchaeological investigations at the Topper and Big Pine Tree sites, Allendale County, South Carolina, Journal of Archaeological Science 36 (2009) 1300–1311.

Leave a Reply

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

You are commenting using your 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 )

Connecting to %s