I appreciate Zac inviting me to share some thoughts about radiocarbon dating (in Texas) for this blog. Assuming that most readers know at least a little bit about radiocarbon dating, I’ll go right to some ideas and concepts that I think can help us construct more accurate and meaningful chronologies.
1) We should continually strive for greatest precision in terms of what we’re dating. This is probably the most important point to keep in mind. Prehistoric events occurred on the same time scale that we all experience on an everyday basis, and would be better understood by us if we could achieve some approximation of that time scale in our work. Reconstructed site components, Analytical Units, or whatever they’re called that are as many as 2000 years long aren’t very useful for understanding the past. Well-supported temporal units that are precise and concise, so long as they’re meaningful in terms of some important event, style, process, etc., are far better than the multi-millennial time periods that characterize a lot of Texas’s regional chronologies. As dating techniques and approaches continue to improve, we’ll better understand how culture change occurred over surprisingly short intervals.
2) Working toward this precision means (in part) reducing the difference between the target event (what we’re trying to date) and the dated event, or what a sample actually represents. Old wood adds to this difference, as does carbon “floating” in midden deposits. Old wood floating in a midden is twice as bad. Once, I had to work with a field school-excavated assemblage that used 15-20 cm arbitrary levels and collected charcoal samples with no depth control from each level. This meant that the best we could do was to assign a sample’s radiocarbon age to that general 15-20 cm elevation range. Because depth = age, generally speaking, this collection strategy doesn’t contribute to precision; better would be giving each sample its own elevation measurement. Performing botanical taxonomic IDs and then selecting samples for submission from short-lived samples (seeds, twigs, plant foods, etc.) helps avoid old wood, and thus adds precision. Dating bone using reliable pretreatments is essentially the same as dating plant food, and can yield very important results.
3) It can be important to understand whether the date you get back from a lab is indicative of the target event or not. To accomplish this, I try to select samples in “couplets,” or as paired samples where one provides a check on the other, at least at the outset of dating work for a site. Because of old wood, floating carbon, unnoticed disturbances, and so on, how do we know that a single radiocarbon date is an accurate indication of the target event? Some kind of temporal information that can provide a check on the submitted assay is essential except in the case of range finders (single samples submitted in order to get a general idea of the age of a deposit). Think of this: the only thing worse than one date is no dates. Even if samples submitted as couplets come back out of order, knowing whether or how your site/deposit/context is out of order is important information.
4) Not all labs provide the same quality of service, and radiocarbon measurements today are far better than they were even a few years ago. Standard deviations (SD) of 10, 15, or 20 years can now be achieved with relative ease and no additional cost. I presume this has to do either with the kind of mass spectrometer that a lab owns, or with the amount of time the sample is counted. If costs are otherwise the same, there’s no reason to opt for dates with 40-year SDs. Most older dates have large error margins simply because radiocarbon technology has improved tremendously over the years. Nevertheless, this can make big differences when calibrating your dates.
Consider three hypothetical dates (EX-1, EX-2, EX-3), all measuring 1200 BP but with SDs of 20, 50, and 120 years, respectively. Calibrated at two sigma, these assays are:
EX-1=AD 770-887 (95.4%)
EX-2=AD 686-903 (86.9%) and AD 918-964 (8.5%)
EX-3=AD 606-1040 (95.2%) and 1110-1115 (0.2%)
The fact that EX-2 and EX-3 each have more than one probability only means that once calibrated, their distributions intercept the calibration curve in more than one place. This complication aside, the resulting age spans clearly become wider and less precise, from a spread of about a hundred years to over 400 years, with larger SDs. Put another way, the probability distribution of EX-1 is about 27% the size of that of EX-3. Today, it should be possible to reconstruct entire cultural periods of less than 400 years, making many older dates with large SDs or new dates from labs still reporting larger SDs less useful.
5) Take some time to understand what the calibration curve looks like and what this means for your dates. The wiggles, “plateaus,” and reversals that make up the curve have a lot to do with the distribution of radiocarbon ages for certain time periods. For instance, the part of the curve that defines the Toyah period contains a sharp reversal at its early part (ca. AD1350) and a couple of minor reversals and a relatively flat part at the later portion (ca. AD1600). This means that radiocarbon dates will tend to “stack up” at these two intervals, lending to the impression that more dates fall into these two time periods than in between them. The Montell-Castroville-Marcos part of the curve is also difficult to work with (Figure 1).
Figure 1. The radiocarbon calibration curve with the approximate Montell-Castroville-Marcos interval indicated.
6) When reporting your dates, it’s important to present all the relevant information. This includes not just sample number (both field and lab), provenience, standard deviation, and corrected (or conventional) radiocarbon age (readers can calibrate the dates themselves if you prefer working with radiocarbon ages), but also a thorough discussion of the contexts and associations of that date. What is the dated material? If this date is part of a couplet, how does it compare with its matching sample? Dates by themselves really aren’t very useful without this kind of attending detail, and this can make all the difference for subsequent interpretations, both yours and that of your readers.
7) Because a radiocarbon date is only a statistical probability, presenting them as single-year ages (e.g., 580 BP, AD 662, etc.) implies a precision that doesn’t exist. It seems important to strike a balance between accuracy and precision; this is where tight standard deviations matter. I use calibrated dates at two standard deviations, about a 95% chance that the probability being discussed is an accurate age for the sample in question. Resulting losses in precision can be overcome by using dates to construct different kinds of models.
8) There are a number of ways that radiocarbon data can be presented (calibrated years BP, radiocarbon years BP, BC/AD, Common Era, and so on). It is very difficult to work with models that that don’t explain which kind of temporal unit they’re using. Radiocarbon BP and calibrated BP are about the same in recent times, but get farther apart the older the samples get. It is a serious hindrance for later researchers if they don’t know what kind of time unit is being discussed. This isn’t the problem today that it was many years ago, but a lot of older chronologies are in need of revision, perhaps using altogether new dates.
9) If you amass a decent sample of meaningful dates from a site or study area, it’s possible to use these dates as an additional line of evidence for understanding aspects of the past in ways related to but different from just knowing the age(s) of your deposit/site/study area. I haven’t yet mastered Bayesian statistics for working with dates, but there are on-line tools (like OxCal) that allow archaeologists to propose some interesting and important models. If you use these, remember that the models that come out are only as good as the dates and assumptions that go in (garbage in, garbage out). It’s possible, however, to work with dates-as-data in ways other than complex Bayesian tools. I’ve give two examples.
Radiocarbon dating at 41HY163, the Zatopec site, documented three main occupation periods, what we called Late Archaic 2, Late Archaic 2/Austin, and Toyah (Figure 2; the site’s longer sequence isn’t represented in the radiocarbon evidence). This record includes AMS dates run on charcoal (possibly including old wood), bison (not specifically pre-treated), and human remains (XAD-purified). We calibrated the dates for the Late Archaic 2/Austin occupation period and for the following Toyah period, and examined the distributions of these dates at two standard deviations. What we found was that each period was characterized by approximately four non- or minimally-overlapping occupation “events,” but that the Toyah events spanned only about 200 years whereas the preceding period covered around 400 years (Figures 3, 4). Obviously we can’t know that dates from each event actually resulted from the same visitation. Yet in terms of archaeological and radiocarbon visibility, the frequency of visitations clearly increased. From this, we hypothesized that site occupation frequency changed from around once every hundred years or so in Late Archaic 2/Austin times to about once every 50 years in Toyah times, or approximately doubled in frequency. This is an important change in how hunter-gatherers were covering the landscape during these periods.
Figure 2. Calibrated radiocarbon dates from 41HY163, the Zatopec site (Yelacic and Lohse 2011: Figure 5-25).
Figure 3. Calibrated radiocarbon dates from Late Archaic 2/Austin occupation at 41HY163. We modeled four hypothetical occupation events that altogether spanned about 400 years based on non- or minimally overlapping calibrated probabilities (Yelacic and Lohse 2011:Figure 5-23).
Figure 4. Calibrated radiocarbon dates from Toyah occupation at 41HY163. Again, we modeled four hypothetical occupation events for this period based on non- or minimally overlapping calibrated probabilities, but these spanned only 200 years (Yelacic and Lohse 2011:Figure 5-24).
In the other example, a 3×4 m block at 41HY160, Spring Lake, was intensively dated with over 30 radiocarbon dates for a deposit that reached only about 170 cm. We “cleaned” this record of dates by excluding those from units that showed clear signs of disturbance (utility lines, burrows, etc.) and arranged the remainder in order by depth to create a depth or age-depth model (Figure 5). This sequence includes dates run on bison (XAD-purified) and charcoal (mostly hard woods). If this deposit resulted from consistent, uninterrupted sedimentation, we’d expect to see the dates in order by age, with older ones at the bottom and successively younger ones toward the top. What we see, however, is a lower zone of fairly rapid sedimentation (Phase 1) followed by a zone (Phase 2) where some dates are in inverted order by age. This process began at ca. 4200 cal BP (the beginning of the Late Archaic for our purposes) but included some Middle Archaic dates. This part of the deposit, from about 130-100 cm in depth, seemingly is not intact and does not offer good opportunities for clearly dated target events. During the third Phase, starting by around 3000 cal BP, site depositional processes returned to “normal” and we again see dates in order by depth.
Figure 5. Age-depth model from 41HY160 showing three phases of site formation based on combinations of natural and cultural factors. This model, in particular, is very important to how the Middle and Late Archaic are defined and dated at this site. This kind of model is useful for understanding the depositional history of contiguous blocks, but requires more than only a small handful of dates. Figure compiled by David Yelacic (Lohse et al. 2014:Figure 7).
Overall sedimentation rates are somewhat less in Phase 2 than in Phase 1, but are still higher than in Phase 3, so we know that these inverted dates are not just the result of natural processes (no dirt being deposited on the site). By comparing burned rock frequency by depth with this model and what it tells us of sedimentation by Phase, we conclude that the disturbed nature of Phase 2 probably results in part from an increase in hot-rock and earth oven cooking activities. Perhaps more importantly from a regional perspective, we identify the period from ca. 4200-3000 cal BP as one during which hot-rock cooking was most intensive; this hypothesis could be tested at other sites.
Both cultural and natural factors contributed to the geoarchaeological nature of this deposit at this time; this can probably be seen at other sites in central Texas and likely has much to do with why the cultural chronology for the Middle Archaic and early Late Archaic is poorly resolved.
Not discussed here are any of the many other issues that could be addressed to help Texas archaeologists work better with radiocarbon data. For instance, I’ve been doing a lot of bison bone dating over the last few years, because I found that existing models for when bison were present in the region were far from adequate. Bone requires certain pretreatments, however, in order to yield reliable (accurate) dates. Also not discussed are different options for modeling radiocarbon data using Bayesian tools. Nevertheless, with some understanding of key relationships between target and dated events, of how archaeological associations lend to imprecision (and inaccuracy), and of other sources of imprecision and how we can ascertain the reliability of dates, I think a lot of dating work done across the state could open entirely new understandings of the past.
About the Author
Dr. Jon C. Lohse is a Principal Investigator at Coastal Environments, Inc.
**Next Monday we continue on the topic of radiocarbon with a discussion by Dr. Scott Pletka that focuses upon radiocarbon dates from the Woodland-era of East Texas**