Technical Comments

Response to Comment on “Late Upper Paleolithic occupation at Cooper’s Ferry, Idaho, USA, ~16,000 years ago”

See allHide authors and affiliations

Science  10 Apr 2020:
Vol. 368, Issue 6487, eaaz6626
DOI: 10.1126/science.aaz6626


Manning builds an inappropriate Bayesian age model to assert that the initial occupation at Cooper’s Ferry began only ~15,935 ± 75 to 15,130 ± 20 cal yr B.P., suggesting that our estimation of ~16,560 to 15,280 cal yr B.P. is unsupported. However, this analysis both ignores evidence of human occupation from the earliest undated cultural deposits and reflects a misapplication of Bayesian age-modeling techniques. Consequently, his results are unreliable.

Manning (1) suggests that our conclusion that human occupation at Cooper’s Ferry likely began 16,560 to 15,280 cal yr B.P. (95.4% confidence) (2) is unsupported by the chronometric data and analysis presented. He takes issue with the Bayesian age modeling performed and with the geoarchaeological evidence we offered.

In Bayesian age modeling [via OxCal (3, 4)], chronological models for archaeological sites are built variously. Misunderstanding the statistical analyses that underlie the functions used to build a model, however, might produce poor model construction and unreliable results. This is exhibited in Manning’s comment (1). Here, he recommends that instead of applying a (start) Boundary function, we should have used the First query, which calculates a date for the first probable event in a Phase. This, Manning suggests, provides a more reliable start estimate of human occupation. Statistically, however, a First query assumes that dates within a Phase are entirely independent. This is not the case for an archaeological unit or feature. Therefore, a First query function is inappropriate. Instead, a (start) Boundary takes interdependence into account. This approach is recommended by specialists and explicitly noted in the OxCal manual (5). In addition, although Manning argues that the “modeling will determine too early a start Boundary simply given the data scatter,” we note that the analysis performed by a First query is equally affected by scatter. Manning also applies a “Charcoal” Outlier_Model on charcoal dates and manually removes outliers identified through the statistical analysis. This assumes that charcoal samples of unknown species have significant inbuilt age. The “Charcoal” Outlier_Model, however, samples solutions from only the younger end of the radiocarbon likelihood, making inbuilt age for the charcoal determinations a prior. At Cooper’s Ferry, inbuilt age is not a certainty and sample displacement is likely both older and younger. Manning makes this very point when discussing the association of samples with human activity. As a result, our use of a “General” Outlier_Model, which allows samples to be either older or younger, best reflects the likely possibilities for this sequence. Moreover, manual removal of outliers defeats the purpose of outlier analysis (whether “Charcoal” or “General”), which objectively identifies outliers within a sequence and provides automatic down-weighting (6).

Manning questions the quality of dates OxA-38051 and OxA-X2792-48 at the base of LU3 because of missing information. We regret that %C and δ13C values for OxA-38051 were omitted by accident, and a formal correction has been made. These (δ13C = –23.7; %C = 44.3) are within accepted parameters (7) in any case, and the measurement is robust (8). Note that OxA-38051 was not, contrary to Manning, “flagged as a possible outlier (residual, in-built age?)” in our model. At a posterior probability of 8% [see figure 3 in (2)], this is marginally above the prior outlier probability (set at 5%), and the date is in clear agreement with the stratigraphic context. Manning’s exclusion of OxA-38051 from his analyses (figures 1 and 2) is therefore based on an erroneous understanding or interpretation of the data originally presented.

Finally, we question the purpose of the analysis performed in Manning’s figure 1 and its technical viability. Here, he argues that age estimates obtained from start Boundaries are older than the oldest date obtained, implying that this is problematic. It is not. This is the function and intended output of a start Boundary: to provide an age estimate for an event that is not directly dated (the same applies with the First query). After all, a radiocarbon date on its own is a minimum-age estimate for an archaeological event and this, in turn, postdates human presence. Therefore, the start Boundary is particularly useful in defining when human occupation began at a site or region. Given existing archaeo-chronometric data for Cooper’s Ferry and the appropriate use of Bayesian age modeling techniques, 16,560 to 15,280 cal yr B.P. (95.4% confidence) remains the most likely estimate for the start of human occupation at Cooper’s Ferry. With respect to the analysis itself, it is incorrect to arbitrarily define a temporal range for the “beginning” of a (start) Boundary distribution (blue bar) and its median (green bar). This treatment reveals a misunderstanding of the probability density function (PDF; at 95.4% confidence, a PDF is an age range that includes the 95.4% most likely results based on the prior). As such, counting years between either end of the colored bars and the median of the oldest cal yr B.P. age is inappropriate. The ubiquitous use of bars is also incorrect, as these imply uniform distributions and the temporal ranges are treated as such, with either end carrying the same probability. These distributions are not uniform, however.

Manning questions the association of dated samples outside archaeological features with past human activity, arguing that features alone establish chronostratigraphic control at the site. In his view, the earliest archaeological evidence at Cooper’s Ferry is solely associated with the hearth (F129) and a pit feature with extinct horse remains (F143). He disregards artifacts, bone fragments, and charcoal stratigraphically beneath these features in LU3. This is a radical departure from how radiocarbon and luminescence dating methods are used to establish archaeological chronologies because many sites lack discrete cultural features. If his approach were strictly enforced, much of the evidence used to establish the late Pleistocene human occupation of Beringia (9) or the age of Clovis sites (10) should be rejected. Manning points out the presence of early Holocene-aged radiocarbon dates in LU3 as evidence of “pervasive bioturbation.” However, as we explained in the supplementary materials, the three outlier radiocarbon ages reported from LU3 were collected on the margins of observed rodent burrows and submitted as an experiment to determine whether such samples, which were typically avoided, would return anomalous radiocarbon ages. They did (11).

Even if Manning’s model were correct—which, we stress, it is not—the initial start date for human occupation at Cooper’s Ferry that his model predicts is ~15,935 ± 75 to 15,130 ± 20 cal yr B.P. (when the age he incorrectly claims as an outlier is included). This is far earlier than any other well-dated archaeological site in North America, earlier than the earliest estimates for the separation of the Cordilleran and Laurentide ice sheets at the southern end of the ice-free corridor at ~14,900 ± 900 years ago (12), and considerably earlier than current estimates for the bioavailability of the ice-free corridor at ~13,000 years ago (13, 14)"" \h . This, in line with genetic evidence (15)"" \h , further supports our conclusion that the initial dispersals into mid-latitude North America occurred elsewhere, probably along the Pacific coast.

References and Notes

  1. “OxA-X-” dates denote samples that are research measurements using nonstandard or experimental methods, or determinations that come with a caution based on the chemical parameters observed during pretreatment and measurement. In this case, the latter applies because of a low C% value (28%). Given that at 95.4% confidence, the date was in agreement with OxA-38051—both samples located within the same unit, at a similar elevation—the measurement was included in the analysis. It is important to note that neither was identified as an outlier in the modeling.
  2. It was also explained that all other objects mapped in place, including dated samples, were determined to be in good stratigraphic context outside of rodent burrow traces. Our methods of observation and excavation were careful enough to enable a precise separation of intact versus bioturbated sediments. The fact that the dates in question are outliers reinforces this, and their inclusion in our modeling shows transparency.
View Abstract

Stay Connected to Science

Navigate This Article