ABSTRACT
The building of an appropriate Bayesian model that reflects as satisfactorily as possible the observed Iron Age stratigraphy at Tel Rehov cannot incorporate all the 64 Groningen dates from Tel Reh ov. There are archaeological constraints in terms of some stratigraphic uncertainties (Mazar et al. [Chapter 13, this volume]). The dates used, therefore, are mainly from Area D and Area C on archaeological grounds. Moreover, the 14C dates from a certain Locus or ‘Phase’ should be internally coherent in the sense that recognisable outliers should be removed before the Bayesian analysis. Yet samples that seem to have every stratigraphic reason to be correct but appear to have less 14C coherence should not be arbitrarily withheld from the model. Such samples may end up as not agreeing after the Bayesian analysis. Their low agreement index indicates a problem (maybe instances of reworked material, humic acid infiltration, erroneous stratigraphic interpretation, and so on). Nevertheless, a few of such erroneous samples will have little impact on the overall Bayesian dating results, if the number of coherent dates is sufficiently large. This is an important strength of employing a holistic analytical framework. The model that has been developed is presented in Table 15.1, showing all stratigraphic and 14C date components. A concise description of the model is given below, as extensive information about each Stratum and Locus is available (Bruins, Mazar and van der Plicht, in press; Bruins, van der Plicht and Mazar 2003a; Mazar 1999, 2003; Mazar et al. [Chapter 13, this volume]; van der Plicht and Bruins [Chapter 14, this volume]). Most Groningen radiocarbon dates from Tel Reh ov are based on seeds. Therefore, a calibration curve based on single year dendrochronological measurements would have been preferable, as stated by Mook and Waterbolk (1985: 22): ‘the 14C sample and the calibration data should have the same time-width (growth-period)’. Such a curve is not available for the approximate time-period 1200-600 BCE of the Levantine Iron Age. Since the 1998 calibration curve (Stuiver et al. 1998; Stuiver and van der Plicht [eds.] 1998) is more detailed than the smoothed 2004 version (Reimer et al. 2004), the former has been used rather than the latter. The dendrochronological database for the IntCal04 curve is largely similar to the dataset of the IntCal98 curve, but also includes new measurements for the Iron Age period, for example, on German Oak samples run for the East Mediterranean Radiocarbon Intercomparison Project (see also Manning et al. [Chapter 10, this volume]). A trial run of the model against the IntCal04
calibration curve gave essentially similar results, albeit that the dates become slightly older. A few examples of these IntCal04 results are included for comparison in relation to the Iron IB-IIA boundary and the Stratum V destruction event. There is another basic aspect that should be mentioned here briefly in relation to the Groningen 14C dates of Tel Rehov: ‘The statistical (random) nature of radioactive decay causes the results of repeated measurements to spread around a “true” value. The possible discrepancy between a measured value and the “true” value is indicated by the standard deviation ()’ (Mook and Waterbolk 1985: 10). Therefore, the midpoint value of a single date may be 1 (68.2%) or 2 (95.4%) away from the ‘true’ value. Making two or three measured values of the same sample (subsamples), each with its own pre-treatment, results in a much firmer dating basis, which we consider important in Near Eastern archaeology, as the 14C dating method is pushed to its very limit of resolution (van der Plicht and Bruins 2001). Though two midpoint dates on both ends of a mutual 2 range are considered the same in physical-mathematical terms, the calibrated age of each of them may be substantially different from an archaeological-historical perspective. It is imperative in our methodology of duplicate or triplicate measurements of single samples, employed for many of the Tel Rehov Loci, to calculate the weighted average of the separate dating measurements. Thus, the outcome will be more precise and possibly also more accurate, closer to the ‘true’ value, if the radiocarbon laboratory involved does not have any systematic measurement bias (van der Plicht and Bruins [Chapter 14, this volume]). Hence the ‘R_Combine’ command is often used in the developed Bayesian model, so that the weighted average results of multiple measurements of one sample of a certain Locus are calculated by the model prior to the Markov Chain Monte Carlo sampling process (Bronk Ramsey 2003; Gilks, Richardson and Speigelhalter 1996). The underlying assumption for calculation of the weighted average is that the organic materials from the Locus are truly contemporary. The oldest Iron Age layer at Tel Rehov is Stratum D-6, containing Iron IA ceramics. This Stratum is represented in the model as ‘Phase D6’. The term ‘Phase’ here is the OxCal model language terminology to characterise an archaeological layer that cannot overlap in time with another ‘Phase’, due to stratigraphic succession (Bronk Ramsey 2003). Hence, each Phase is contained within an upper and lower boundary in the model. ‘Phase D6’ is represented by two Loci (2874 and 2836), as each Locus has two coherent radiocarbon dates. The sample of charred olive stones from Locus 2874 was split and subsequently separately pre-treated and dated by the two different radiocarbon dating systems available at Groningen (van der Plicht and Bruins [Chapter 14, this volume]): Proportional Gas Counting (PGC) and Accelerator Mass Spectrometry (AMS). Thus, we have two dates from the same sample (Basket 28701) and, therefore, the ‘R_Combine’ command is used in the Bayesian model, as only the calculation of the weighted average of the same sample renders the correct representation for this Locus. The same procedure was followed for Locus 2836, as the charred olive stones from Basket 28352 were dated by both PGC and AMS. Such a procedure does not only increase precision but probably also accuracy, because two independent pre-treatment and radiocarbon dating systems are involved. Both Groningen systems having an ongoing record of accuracy against one another and against other high precision laboratories active in the field of calibration of the Radiocarbon timescale (Seattle, Belfast, Heidelberg, Groningen-conventional, Pretoria, and Tucson-conventional). Differences between the above laboratories are in the range of 0-20 14C years, with the difference between Pretoria and Groningen (conventional) being only 7 14C years, Groningen being slightly younger. The results of an intercomparison (LeClercq, van der Plicht and Gröning 1998) between Seattle, Belfast, Heidelberg, Groningen (both conventional and AMS), Waikato and Tucson (conventional) are available (see van der Plicht and Bruins [Chapter 14, this volume] for more details).
