Early Success with TL Dating of Archaeological Ceramics

Beginning in the late 1970s, the laboratory at the Pontificia Universidad Católica de Chile (UCTL) made promising headway with TL dates that agreed with ceramic sequences in northern Chile (Brito et al. Reference Brito, Deza, Román and Concha1979; Román and Deza Reference Román and Deza1985; Berenguer et al. Reference Berenguer, Deza, Román and Llagostera1986; Muñoz Ovalle and Chacama Rodríguez Reference Muñoz Ovalle and Chacama Rodriguez1988; Schiappacasse et al. Reference Schiappacasse, Román, Muñoz, Deza and Focacci1991). Initial efforts mentioned ¹⁴C dates, but these were often generalized comparisons to dates associated with similar ceramic styles from other sites. Early on, it was acknowledged that TL dates tended to underestimate ¹⁴C dates, in both Chile and Ecuador (Stothert Reference Stothert1988; Schiappacasse et al. Reference Schiappacasse, Román, Muñoz, Deza and Focacci1991). However, minor differences did not significantly impact low-resolution regional chronologies or ceramic sequences. This made it common to ignore inconsistencies, which were not even apparent because of large error ranges. For example, at the site Turi Aldea in northern Chile, there are 3 ¹⁴C and 2 TL dates from a single occupation layer (Castro et al. Reference Castro, Aldunate, Berenguer, Cornejo, Sinclaire and Varela1994; Sinclaire Reference Sinclaire2004). The TL dates’ medians are 145–225 years older, but the error ranges of 120–215 years make them statistically indistinguishable (Figure 2; Table 2).

Figure 2 ¹⁴C and TL dates from occupational component I at the site Turi Aldea in northern Chile. Because of their large error ranges, these dates pass individual and group chi-square tests. They can be statistically combined with a high agreement index (A=139). ¹⁴C probability ranges are in blue; TL in orange. Medians indicated as vertical lines at the base of each distribution. UCTL lab codes were not published.

Table 2 Paired TL and ¹⁴C dates from Argentina and Chile. The 22 Inca-period pairs are shaded. Positive differences between medians indicate luminescence dates underestimate the paired ¹⁴C dates. Many ¹⁴C dates are available in recent compilations (Campbell and Quiroz Reference Campbell and Quiroz2015; Gayo et al. Reference Gayo, Latorre and Santoro2015; Gil et al. Reference Gil, Giardina, Neme and Ugan2014). ¹⁴C dates were run on charcoal except the three from Arenal I, which were run on chicken bones.

In Ecuador, phases for the Valdivia culture were defined with TL dates that had large error ranges of ±245–578 years, even though the sherds had excellent characteristics for TL dating (Marcos and Michczyński Reference Marcos and Michczyński1996; Galli et al. Reference Galli, Sibilia and Martini2020: 190). Phasing based on TL and ¹⁴C dates tends to agree, but for a few ceramic phases, TL dates consistently underestimate age (Martini and Sibilia Reference Martini and Sibilia2001: 243; Galli et al. Reference Galli, Sibilia and Martini2020: Figure 5). Comparisons have only been made between regional ceramic phases, rather than more rigorous tests of paired dates from the same depositional event. Recently, calibrated ¹⁴C dates on pottery residues suggest that early pottery in Ecuador is much older than the TL-based chronology (Tabarev et al. Reference Tabarev, Kanomata, Marcos, Popov and Lazin2016; Kanomata et al. Reference Kanomata, Marcos, Popov, Lazin and Tabarev2019), a significant finding for some of the oldest pottery in the Western Hemisphere.

The accumulation of TL dates led researchers to build cultural and ceramic sequences with both TL and ¹⁴C dates, despite some inconsistencies. In part, the paucity of dates motivated them to use as many as possible. It was also more legitimate to question ¹⁴C dates in the 1980s, prior to methodological refinements, extensive inter-laboratory comparisons, and refined calibration curves (Bayliss Reference Bayliss2009), in addition to potential problems with old wood and sample–artifact associations. A significant factor in the popularity of TL dates in both Chile and Ecuador was price: they were 75% cheaper than ¹⁴C dates (Berenguer et al. Reference Berenguer, Roman, Deza and Llagostera1988: 343; Marcos and Michczyński Reference Marcos and Michczyński1996: 102). Generally, TL dates fit stratigraphic sequences, which built trust in the method. This may partly explain why recent efforts tend to be less careful about using on-site dosimetry and reporting contextual details.

Dates in Chile have the additional advantage of comparability: all dates have been estimated with the same procedures at the same laboratory. UCTL measures the equivalent dose with three methods: plateau, additive with superlinearity correction, and pre-dose (Supplementary Material 1; Brito et al. Reference Brito, Deza, Román and Concha1979; Concha et al. Reference Concha, Román, Brito and Deza1980; Román et al. Reference Román, Deza and Brito1983; Román and Deza Reference Román and Deza1985; Deza and Román Reference Deza and Román1986; Román and Deza Reference Román and Deza1998). UCTL reported that for younger sherds, they prefer the pre-dose method since the luminescence signal is weaker and there is a greater chance of uncertainty; for older sherds, they prefer the plateau method (Bárcena Reference Bárcena1998: 365). The laboratory’s procedures have not changed for decades. They are considered reliable and are used at other luminescence laboratories (for updates to luminescence methods, see Roberts et al. Reference Roberts, Jacobs, Li, Jankowski, Cunningham and Rosenfeld2015).

Elsewhere in the Andes, results are mixed. Some TL dates match high-precision ¹⁴C dates such as the exemplary Bayesian chronology of the Sipán tomb complex, even though problematic dates are discarded (Aimi et al. Reference Aimi, Alva, Chero, Martini and Maspero2016; see also Roque et al. Reference Roque, Guibert, Vartanian, Vieillevigne and Bechtel2004). Near Lake Titicaca, TL ages on ceramics from raised fields seem unreliable and have unwieldy error ranges of 90–660 years (Erickson Reference Erickson1988: 194; Janusek and Kolata Reference Janusek and Kolata2004: 410). At the lakeshore site Huajje, TL dates were corrected for feldspar fading and mostly matched temporal expectations (Schultze Reference Schultze2008: 391–397). In the southern Lake Titicaca Basin, TL dates from Qeya-style museum pieces in France underestimate the style’s expected date range, based on Bayesian models of ¹⁴C dates and stratigraphy (Marsh et al. Reference Marsh, Roddick, Bruno, Smith, Janusek and Hastorf2019; A. Roddick, personal communication 2021). In the southern Nazca region in Peru, a large OSL study (Vaughn et al. Reference Vaughn, Eerkens, Lipo, Sakai and Schreiber2014) had a number of unexpected results compared to both ceramic seriation and Bayesian models of ¹⁴C dates (Unkel et al. Reference Unkel, Markus, Hermann, Johny, Bernd and Volker2012). Since so many factors affect archaeological dates, it is challenging to identify the cause of disagreement between methods, especially when dates do match in some cases. To begin to do this, the next section assesses individual paired dates.

Approach 1: Individual Paired Dates

To evaluate associated ¹⁴C and TL samples that archaeologists expect to be the same age, we have adapted Waterbolk’s (Reference Waterbolk1971) categories for more or less certainty of the depositional association between dated samples. Date pairs with higher grades are more reliable data points for testing dating methods.

1. A. Very high probability. For example, a ¹⁴C date on organic residue from a pot and a luminescence date from the same vessel, or when a production stamp or seal indicates the production date of the vessel.

2. B. High probability. For example, both samples are from a single depositional event such as a hearth or occupational floor. Samples are spatially close to one another and the context is small and carefully excavated. This includes tombs that were not looted or reused (otherwise tombs are C).

3. C. Probability. For example, the two samples are from the same architectural structure or occupational layer. Contexts are larger and excavations less controlled.

4. D. Reasonable possibility. For example, the samples are from the same occupational layer but different sectors of a site, or from a site with only one period of occupation. These pairs are of little relevance to testing the dating methods and not generally included here.

Since ¹⁴C and luminescence date different events, we should expect a lag between them. Luminescence dates a pot’s firing, which should be earlier than its use or deposition, which is dated by ¹⁴C dates. Heirloom vessels in graves may have been fired generations prior to the death of the associated individual (e.g., Fitzpatrick et al. Reference Fitzpatrick, Kaye, Feathers, Pavia and Marsaglia2009), but in most cases, the lag is likely no more than a few decades so the dates’ probability ranges should overlap. This overlap can be evaluated with a chi-square test (Ward and Wilson Reference Ward and Wilson1978) implemented with the Combine command in OxCal 4.4 (Bronk Ramsey Reference Bronk Ramsey2009a). Here, we use this approach to compare normal TL probability distributions, irregular ¹⁴C calibrations, and uniform distributions for historic ranges. If a date pair passes the test, “we have no statistical evidence to doubt the consistency of the two determinations” (Ward and Wilson Reference Ward and Wilson1978: 30). Radiocarbon dates are calibrated with SHCal20 (Hogg et al. Reference Hogg, Heaton, Hua, Palmer, Turney, Southon, Bayliss, Blackwell, Boswijk, Ramsey, Pearson, Petchey, Reimer, Reimer and Wacker2020), the most appropriate curve for the southern Andes (Marsh et al. Reference Marsh, Bruno, Fritz, Baker, Capriles and Hastorf2018); calibrated medians and probability ranges are rounded by 10 years. Modeled results are presented in italics. We compare medians in years and percentages (Tables 2–4), for example, the ¹⁴C date Beta-69935 from Chile is ∼780 years old (calibrated) and its pair UCTL-540 is 240 years younger, a difference that is 31% of its ¹⁴C age. This follows the convention in luminescence dating of reporting errors as a percentage of the age, for example, Alpha-2076 from Mendoza was reported as AD 1490±20%. When available, UCTL uses on-site CaSO₄:D_y dosimeters to correct TL ages (Deza and Román Reference Deza and Román1986). In this region, most dosimetry corrections are small, reflecting the generally low external dose (Table 1).

Table 3 Comparisons of 28 historic-period TL dates on floor tiles, bricks, and sherds with known production dates from northern Mendoza, as in Table 2.

Table 4 Paired dates from the pilot study in Mendoza. Crossed-out dates were not used by the laboratory to estimate the age. Valid OSL and IRSL dates were combined following Ward and Wilson (Reference Ward and Wilson1978) to arrive at estimated sherd ages. Inca-period pair is shaded.

Date Pairs from Chile and Argentina

Despite the fact that the UCTL laboratory has run over three thousand TL dates, surprisingly few are paired with ¹⁴C or historic dates. We identified 59 luminescence-¹⁴C date pairs and 28 TL dates on items with known historic production ages from Chile and Argentina (Tables 2 and 3).

Approach 2: Bayesian Models of ¹⁴C and TL Dates

We compared ¹⁴C-only and TL-only Bayesian models for 3 ceramic styles in northern Chile and the Inca occupation of Mendoza. Since we cannot assess all factors in individual date pairs, we turn to larger samples that should be less sensitive to case-specific problems. If both methods are accurate, independent Bayesian models for each method should show overlap in Kernel Density Estimates (KDE) and starting and ending boundaries (Bronk Ramsey Reference Bronk Ramsey2017).

We modeled temporal ranges for 3 ceramic styles with dates from cemeteries in the Azapa Valley in northern Chile, where there are 66 ¹⁴C and 60 TL dates (Muñoz Ovalle Reference Muñoz Ovalle2019). The lack of agreement between the dates has led to long-running debates on the ceramic sequence (see Korpisaari et al. Reference Korpisaari, Oinonen and Chacama2014: 411–4). The models for Maytas-Chiribaya, Cabuza, and San Miguel ceramics all had acceptable agreement indices (Supplementary Material 2). For the Cabuza models, 3 outliers were removed from the ¹⁴C model and 2 from the TL model, following Korpisaari et al. (Reference Korpisaari, Oinonen and Chacama2014).

For Maytas-Chiribaya ceramics, the TL and ¹⁴C models agree: the phase boundaries overlap and the KDE trends are similar (Figure 3). In contrast, the models for both the Cabuza and San Miguel ceramics strongly disagree. The TL dates are internally coherent but fall centuries earlier than the ¹⁴C dates associated with the same ceramic styles. The TL dates significantly overestimate the radiocarbon dates, the opposite trend seen in individual date pairs. It is unclear why the two dating methods agree for Maytas-Chiribaya ceramics but not for Cabuza ceramics, since microscopic and chemical analysis of the 2 styles show they have very similar paste compositions, firing patterns, and were likely made with local temper and water (Ogalde Reference Ogalde2019: 143–144, 175–176). The reporting and use of dosimetry corrections is inconsistent, a trend seen elsewhere in Chile, but this probably would not have had a major effect on the results (Korpisaari et al. Reference Korpisaari, Oinonen and Chacama2014: 422; Puerto Mundt and Marsh Reference Puerto Mundt and Marsh2021). Most dates were run on individual sherds, which could have been out of primary context since many of these cemeteries have been looted. It is also possible that some of these are heirloom vessels, which would explain the older TL dates. However, individual outliers should not affect Bayesian trends. The ¹⁴C models agree with the interpretation that all three styles emerged from the post-Tiwanaku diaspora and should fall after Tiwanaku’s collapse ∼AD 950–1000. This is supported by a large set of ¹⁴C dates from textiles and human bones (Cassman Reference Cassman1997; Sutter Reference Sutter2000). Hence, future research may find it productive to proceed without the TL dates.

Figure 3 Bayesian models of ¹⁴C and TL dates for 3 ceramic styles from the Azapa Valley, Chile. Green and red curves indicate starting and ending boundaries, respectively. Kernel density estimates (KDE) are indicated in between the boundaries, in blue for ¹⁴C dates and orange for TL dates. Vertical lines below the curves indicate modeled and unmodeled medians.

The apparent reliability of Inca-period TL dates led Marsh et al. (Reference Marsh, Kidd, Ogburn and Durán2017: 126) to include them in a Bayesian model of the Inca occupation of Mendoza. This model required complex outlier models that discarded 10 (19%) of 54 dates. Here we compare two single-phase models, one for each dating method (Figure 4; Supplementary Material 2). The ¹⁴C model has 31 dates, including eight dates not included in the previous model (Bárcena Reference Bárcena2010; Morgan et al. Reference Morgan, Neme, Sugrañes, Salgan, Gil, Otaola, Giardina and Llano2017; Terraza et al. Reference Terraza, Bárcena and Aguilar2019; Durán et al. Reference Durán, Zárate, Yebra, Frigolé, Gasco, Castro, Winocur, Barberena, Cortegoso, Marsh, Lucero, Andreoni and Zonana2021). Two dates (6%) had low agreement indices (A=22 and 33), which follows the general expectation that a set of dates will have around 5% outliers (Bronk Ramsey Reference Bronk Ramsey2009b). The agreement indices are above 60 and hence acceptable (A_model=84, A_overall=76). The Bayesian algorithm converged on a precise starting boundary, cal AD 1380 (1350–1430, 95% probability), despite dispersed medians and large error ranges. This agrees with estimates for the earliest evidence of the Inca empire in northwestern Argentina; a model of ¹⁴C dates has the same boundary medians, cal AD 1380 and 1520 (Greco Reference Greco2012: 408–411). The ending boundaries are coherent with the sequence of historically documented events: the first Spanish expedition into northwestern Argentina in AD 1536 (Vitry Reference Vitry2007), the initial Spanish occupation of Santiago in AD 1540, the founding of Mendoza in AD 1561, and two travelers’ reports that the Inca sites were in ruins by AD 1595 (Parisii Reference Parisii1994: 55).

Figure 4 Bayesian models for the Inca occupation of northern Mendoza, as in Figure 3.

Figure 5 Early pit house from Barrancas, site B61, northern Mendoza, indicating the location, calibrated median, and lab code for each date. ¹⁴C samples marked as white squares and the sherd dated by OSRL and IRSL marked as an orange square. Based on original drawing by Diego Estrella.

In contrast, the TL-only model would not converge unless constrained by an ending date of AD 1595. We included two additional dates not included in the previous model (Bárcena et al. Reference Bárcena, Terraza and Iniesta2015). The model’s agreement indices are just below the acceptable threshold (A_model=58, A_overall=48). Four of the 21 dates (19%) have low agreement indices (A=28–42), more than expected. Compared to the ¹⁴C model, the two boundaries do not overlap and the TL starting boundary is later and much less precise. The KDE suggests continued Inca occupation until AD 1595, which disagrees with historically documented events. The TL model underestimates the dates of Inca occupation, even though overlapping error ranges allow individual date pairs to pass chi-square tests. Hence, we suggest that the updated ¹⁴C-only model is our current best estimate of the timing of the Inca occupation in Mendoza.

Approach 3: A Pilot Study of Paired ¹⁴C, TL, OSL, and IRSL Dates in Mendoza

The final approach was a pilot study with a set of six sherds from Mendoza paired with ¹⁴C dates, all run at the same laboratory in Argentina. The luminescence dates were run at the University of Washington by James Feathers, who used three methods for each sherd: TL, OSL, and IRSL. All sherds were undecorated (Table 4; for methods and photos see Supplementary Materials 3 and 4). These sites were excavated in 5-cm levels, so they have grade-B associations between dated pairs.

One limitation is the lack of on-site dosimeter corrections. Studies in the region show low external rates (Bárcena Reference Bárcena1998; Schmidt et al. Reference Schmidt, Tsukamoto, Salomon, Frechen and Hetzel2012), so we use dates that assume a low dose of 0.5% K, 6 ppm Th, and 2 ppm U, as suggested by the laboratory (Supplementary Material 3). Most likely, on-site dosimeters would result in minor adjustments to the dates, as they did in Bárcena’s (Reference Bárcena1998) study (Table 1). TL dating showed anomalous fading in all sherds, and despite fading corrections, only one sherd produced useful data with a high error of ∼30% (UW-3759). This is a notable result, since anomalous fading may not have been fully considered at UCTL (Román et al. Reference Román, Deza and Brito1983: 10). Hence, for the pilot study, only OSL and IRSL ages are considered.

Barrancas, B61. In the southern Andes, ceramics were first adopted in multiple regions around ~150 cal BC (Marsh Reference Marsh2017). This includes Barrancas, a lowland area with the region’s earliest cemetery (Novellino et al. Reference Novellino, Barchetta Aporta, D’Addona, Estrella, Bernal, Devincenzi, Cortegoso and Durán2013) and pit house at the site B61, which has two very similar ¹⁴C dates (LP-2997 and LP-3088; Marsh Reference Marsh2017). The first sample is from a hearth adjacent to the ramp entrance; the second is from a carbon concentration on the other side of the same house (Figure 5). The luminescence-dated sherd is from the same depth (10–15 cm) on the house’s original floor. All three samples are from the better-preserved northern half of the structure, which included details such as the impressions left by branches used to build wattle-and-daub walls.

Structures like this are rarely maintained for more than a generation, so floor refuse is often deposited within a few decades and covered by roof fall, reducing the possibility of mixing with earlier or later material. The sherd from the house floor returned a luminescence date of AD 560±100 (UW-3758). This is many centuries later than the date suggested by the two paired ¹⁴C dates, which have a combined median of 50 cal BC. While unlikely, we cannot rule out the possibility that the sherd washed into the pit from a later occupation. It is also possible that differential radioactivity above and below the floor surface affected the age.

El Manzano Histórico. At this site, a similar pit house was found below a modern road. Carbon and artifact concentrations were dense within the house. A ¹⁴C date from the floor returned a median of AD 1000 (LP-1637; Cambria Reference Cambria2010; Marsh et al. Reference Marsh, Frigolé and Moyano2014), which is consistent with expectations for the associated Agrelo pottery. An undecorated sherd from a different part of the house was submitted for luminescence dating (UW-3757). Unlike other sherds in the pilot study, the recovered dose was higher than the administrated dose, which may explain the overestimated OSL age, AD 480±100. Only OSL was used, since the TL and IRSL signals suffered from anomalous fading. A high b-value hinted that the age may be underestimated, but in fact, the OSL age is centuries older than the ¹⁴C date and all other regional ¹⁴C dates associated with this ceramic style (García Reference García2004).

Paso de Paramillos I. This rock shelter is near a mountain pass west of Mendoza (Bárcena Reference Bárcena1998: 220–2). The deposition of the site is cleanly divided into two layers. The lower one has sparse bits of carbon and very few, small lithic flakes. The upper layer has a distinct soil texture and color and includes a much higher density of artifacts and some ceramics. The sherd dated by luminescence (UW-3759) was found lying flat on the stratigraphic boundary between the two layers. The IRSL signal was too weak to obtain an age, but OSL dated the sherd to AD 780±90. It was found within a few centimeters of a dated carbon concentration (LP-3629), which has a median of AD 350, consistent with regional expectations for early highland pottery. The luminescence age underestimates the ¹⁴C date by multiple centuries. While there was no sign of stratigraphic disturbance near the sherd, rodent activity could have moved this sherd. A ¹⁴C–TL date pair from the site’s upper layer site did pass the chi-square test, though the association between samples is less certain (Bárcena Reference Bárcena1998: 222; Table 2).

Las Cuevas 2. This is a high-mountain rock shelter in the Cuevas River Valley (3160 m.a.s.l.). The area has lush summer pastures and is located along the natural pass between Mendoza and Santiago (Gasco et al. Reference Gasco, Llano, Durán, Winocur, Zonana and Paiva2021). Excavation material suggested an occupation from around AD 1300 to Spanish contact. The paired samples are from a depth of 30–35 cm in a stratum with carbon and ash lenses and a clayey matrix (Figure 6). The samples were found immediately below field stones that were part of an informal wall. The radiocarbon date has a median of AD 1490 (LP-3602), consistent with diagnostic Inca-period ceramics from this level and the other Inca ¹⁴C dates from northern Mendoza. In contrast, the luminescence age of AD 180 (UW-3756) is much older, and there is little agreement between the individual TL, OSL, and IRSL ages. While highly unlikely, it is not impossible that the sherd was made more than a thousand years before it was deposited in this site’s layers of refuse.

Figure 6 West profile of unit B7, Las Cuevas 2. The white square indicates the approximate depth of the ¹⁴C sample with its calibrated median and laboratory code; the orange square indicates the same for the sherd dated with luminescence. It is crossed out because it does not match expectations based on diagnostic decorated ceramics or stratigraphy.

Agua de la Cueva. This site has the most complete sequence of human occupation in the province. Occupation began in the Late Pleistocene (García Reference García2003), but its intensity was much higher over the last two thousand years (Durán and García Reference Durán and García1989; Castro and Yebra Reference Castro and Yebra2018). With one exception (AD-1562, level 29), the nine ¹⁴C dates follow the stratigraphic sequence, consistent with the generally undistributed and horizontal deposition of strata (Figure 7; Durán et al. Reference Durán, Gasco, Paiva, Zonana and Barberena2020). The most recent luminescence date (UCTL-1172, level 11) fits stratigraphic expectations, similar to other Inca-period TL dates. Three other luminescence ages (UCTL-1173, UCTL-1173, and UW-3755) underestimate the date of the strata where they were found, which fits the overall trend for individual paired dates. One age (UW-3754, level 24) agrees very well with the ¹⁴C date from the same level; both have medians of AD 860. Another luminescence date from the same level (UW-3755) barely passes the chi-square test, but its median is 3 centuries younger. This difference between the luminescence ages is difficult to explain because the sherds’ depositional histories should be very similar. This part of the sequence has few pits or burning events that might affect luminescence ages.

Figure 7 East profile of unit B, Agua de la Cueva (North) with dates, as in Figure 6. UW-3755’s median is three centuries too young, but it does barely pass a chi-square test (Table 4). Squares shifted to the left are from unit A, which is adjacent to unit B. Dates recalibrated from Cortegoso et al. (Reference Cortegoso, Durán and Gasco2014), Durán et al. (Reference Durán, Gasco, Paiva, Zonana and Barberena2020), and Gil et al. (Reference Gil, Giardina, Neme and Ugan2014). Profile redrawn from Castro and Yebra (Reference Castro and Yebra2018: Figure 4), based on original drawings by Víctor Durán and Gustavo Lucero.

The pilot study of ceramics in Mendoza shows that luminescence ages are inconsistently older and younger than paired ¹⁴C dates. One possibility is that there was a residual signal left in the sherds that resulted in partial bleaching, but on the other hand, they had good plateaus. Another possibility is that energy levels vitrified the quartz, but regional firing practices make it is quite unlikely that vessels were heated above 1000°C. The dose rate calculations could clearly be improved with local corrections using on-site dosimeters. This may have a minor effect, like Bárcena’s (Reference Bárcena1998) dosimeter corrections (Table 1), or be significantly higher, since luminescence sensitivity is unstable in the Andes. In this case, the dates may be accurate, but with error ranges that are far too large to be useful in constructing archaeological chronologies. It is also possible that post-depositional processes moved samples out of primary position; hence, these are not grade-A date pairs. Overall, the pilot study’s inconsistent results echo those of the other two approaches.

Discussion: Disagreement between ¹⁴C, Historic, and Luminescence Ages

We took three approaches to comparing dating methods, which all show that luminescence dates are inconsistent. Of the 94 paired dates from the literature and the pilot study, 56% fail the chi-square test, with a median difference between paired dates of 300 years or 27% (Figure 8; Tables 2–4). For date pairs that passed the test, agreement was not strong, with a median difference of 90 years or 12%. For date pairs that failed the test, differences increase with age, marking a clear trend line (r²=0.50). This trend line crosses the horizontal axis at AD 1500, which may reflect the fact that some laboratories use this date as an a priori reference point for sherd age, since it is usually clear whether sherds are pre- or post-Hispanic. This may help explain why most Inca-period dates pass individual chi-square tests. Method choice may also help explain this; for example, UCTL prefers pre-dose for younger sherds and plateau for older sherds (Bárcena Reference Bárcena1998: 365). The Mendoza pilot study identified anomalous fading in all sherds, which may be another confounding factor for some TL dates.

Figure 8 Summary of differences between paired luminescence and ¹⁴C or historic medians for 94 date pairs (Tables 2–4). Most prehistoric luminescence dates underestimate paired ¹⁴C dates while historic-period dates nearly always overestimate age.

Dates that failed the test showed different tendencies for three age ranges. First, older luminescence dates underestimate ¹⁴C dates, a tendency identified in the 1980s (Stothert Reference Stothert1988; Schiappacasse et al. Reference Schiappacasse, Román, Muñoz, Deza and Focacci1991). Second, during the Inca period, 78% of pairs pass the chi-square test. Third, in the historic period, dates overestimate the true age. In Mendoza, 25 of 28 TL dates on European-fired sherds, floor tiles, and bricks fail the chi-square test; all but one overestimate the expected dates based on historic records. Sediment dates from northern Mendoza have also reported this trend. In a stratum with three consistent ¹⁴C dates, with medians of AD 1760–1850, the associated OSL age was some three centuries older (Schmidt et al. Reference Schmidt, Tsukamoto, Salomon, Frechen and Hetzel2012: 71). In sediments from Chile, refined p-IRIR methods also overestimated the age of two historic events (Brill and Cisternas Reference Brill and Cisternas2020: 9).

UCTL ran 83% of the luminescence dates compared here, but laboratories in Denmark, Peru, and the United States have had no more success despite using updated methods. It is possible that this trend reflects unclear geological variability in northern Mendoza, where 60% of the date pairs come from. However, within this region and others, there is no apparent spatial trend in chi-square results (Figure 1). Furthermore, date pairs with higher association grades are no more reliable: 8 of 11 grade-A pairs failed chi-square tests.