Characterizing obsidian sources with portable XRF ...members.peak.org/~obsidian/pdf/frahm_2014_2.pdf · Characterizing obsidian sources with portable XRF: accuracy, reproducibility,

lable at ScienceDirect

Journal of Archaeological Science 49 (2014) 105e125

Contents lists avai

Journal of Archaeological Science

journal homepage: http : / /www.elsevier .com/locate/ jas

Characterizing obsidian sources with portable XRF: accuracy,reproducibility, and field relationships in a case study from Armenia

Ellery Frahma,b

aDepartment of Anthropology, University of Minnesota, 301 19th Avenue South, Minneapolis, MN 55455, United StatesbDepartment of Earth Sciences, University of Minnesota, 310 Pillsbury Drive SE, Minneapolis, MN 55455, United States

a r t i c l e i n f o

Article history:Received 13 March 2014Received in revised form1 May 2014Accepted 5 May 2014Available online xxx

Keywords:pXRFObsidian sourcingSource characterizationAccuracyReproducibilityField methodsPokr Arteni

E-mail addresses: [email protected], elleryfrahm

http://dx.doi.org/10.1016/j.jas.2014.05.0030305-4403/� 2014 Elsevier Ltd. All rights reserved.

a b s t r a c t

Portable X-ray fluorescence (pXRF) has been demonstrated as a powerful tool to assign obsidian artifactsto sources. Newer instruments can even match artifacts from some regions to their sources in a matter ofjust seconds, not minutes. There remains, however, a reluctance to use pXRF instruments to characterizethe sources themselves. Many past studies have used pXRF in a region where the sources have been wellcharacterized using lab-based techniques. That is, earlier analytical work established compositional typesfor the obsidian sources, and pXRF was later used to sort artifacts into those types. This is due, at least inpart, to notions that pXRF instruments are insufficiently accurate or reliable to characterize the sources.The motivations to use pXRF for characterizing sources are similar to those for sourcing artifacts: manymore specimens can be analyzed without concern for the financial, practical, and legal considerationsassociated with instruments in distant facilities. This paper documents tests conducted to investigate theaccuracy and reproducibility of pXRF data relative to five laboratory-based techniques (NAA, EDXRF,WDXRF, EMPA, and LA-ICP-MS) with a focus on Armenian obsidian sources. These tests demonstrate thatthere is no reason to believe pXRF is inherently inaccurate, unreproducible, or otherwise inadequate forsource characterization. A case study of the Pokr Arteni source highlights the advantages of pXRF,including the capability to analyze large numbers of specimens, recognize variability, and elucidate fieldrelationships. In these respects, pXRF can facilitate more sophisticated obsidian sourcing studies.

� 2014 Elsevier Ltd. All rights reserved.

1. Introduction

Obsidian sourcing, at its core, is a technique for classifying ar-tifacts into compositional types that reflect their geological origins.Like other archaeological typologies, this technique consists of twobasic steps: first, creation of the types (i.e., characterization ofgeological specimens to recognize compositional “fingerprints” ofthe sources), and second, classifying artifacts into those types (i.e.,analyzing artifacts to sort them into the established types) and, inturn, determining their sources. Portable X-ray fluorescence (pXRF)has been demonstrated as a powerful tool to compositionally assignobsidian artifacts to sources (e.g., Frahm, 2007, 2013; Craig et al.,2010; Jia et al., 2010, 2013; Nazaroff et al., 2010; Burley et al.,2011; McCoy et al., 2011; Millhauser et al., 2011; Sheppard et al.,2011; Tykot et al., 2011, 2013a; Forster and Grave, 2012; Goodaleet al., 2012; Kellett et al., 2013). Newer pXRF instruments can

@gmail.com.

even match artifacts in some regions to their sources in a matter ofseconds (Frahm et al., 2014a; Mili�c, 2014).

While the efficacy of pXRF for matching artifacts to their sourceshas been repeatedly validated, there remains a reluctance to usepXRF for the characterization of the sources themselves. Manyrecent studies involved use of pXRF in a region where the sourceshave been well characterized using lab-based techniques. Earlieranalytical work established compositional types for the sources,and pXRF was later used to sort artifacts into those types. Forexample, Rodríguez-Alegría et al. (2013) compared pXRF data forobsidian artifacts to neutron activation analysis (NAA) data fromthe University of Missouri Research Reactor (MURR) for Mexicanobsidian sources. Similarly, Jia et al. (2013) state that, for their studyof Chinese obsidians, “NAAwas undertaken to provide accurate andreliable elemental descriptions for the newly identified volcanicglass sources . with pXRF employed as relative technique formatching archaeological specimens to geochemical sources” (977).The inference is that pXRF analyses are not sufficiently accurate orreliable to characterize the obsidian sources themselves. Otherstatements found in the literature are more explicit but

Delta:1_given name

mailto:[email protected]

mailto:[email protected]

http://crossmark.crossref.org/dialog/?doi=10.1016/j.jas.2014.05.003&domain=pdf

www.sciencedirect.com/science/journal/03054403

http://www.elsevier.com/locate/jas

http://dx.doi.org/10.1016/j.jas.2014.05.003



E. Frahm / Journal of Archaeological Science 49 (2014) 105e125106

unsupported (e.g., Williams et al., 2012:75 claim that pXRF is “lessaccurate than other techniques” but only show variance betweentwo datasets).

The motivations to use pXRF for characterizing sources are akinto those for sourcing artifacts. Specifically, pXRF enables analyses ofmany more specimens without concern for the financial, practical,and legal considerations associated with laboratory-based tech-niques. Consider that Shackley (2005:100) collects at least 200specimens across obsidian flows, but just a fraction is transportedand analyzed with lab-based XRF. With pXRF, characterizing a flowcan be influenced less by the logistics and legalities of shipping andmore by robust sampling and datasets. Glascock et al. (1998) statethat obsidian flows, with few exceptions, are compositionally ho-mogeneous, but in the same volume, Shackley (1998) argues thatintra-source variability is more common thanwidely thought. Suchheterogeneity, he proposes, has often gone unnoticed becausemostobsidian sources have been characterized based on a handful ofspecimens. Instances of intra-source variability were discoveredonly through “intensive and extensive sampling” of obsidiansources (89). Hence, he calls for systematic approaches to (1)characterizing sources as a means to recognize any heterogeneityand (2) elucidating field relationships, which has lagged behindanalytical advances. In these respects, pXRF can facilitate superiorsource characterization.

Fig. 1. Armenian obsidian sources (black circles), select source complexes (dashed lines), andidentical compositions are represented by a single dot (e.g., Djraber, Gyumush, Fontan, etprimary and secondary distribution of these obsidians. Digital elevation data from SRTM3 (Shunder Creative Commons terms from Wikimedia Commons.

Recently, Chataigner and Gratuze (2014a,b) characterizedArmenian obsidian sources (Fig. 1) using laser ablation inductivelycoupled plasma mass spectrometry (LA-ICP-MS). They argue that,while pXRF is sufficient for initially sorting artifacts, lab-basedanalytical techniques are critical:

fairly reliable results can be obtained by using portable XRF[but] there must be a dual approach, combining both fieldmeasurements with portable instrument carried out on a largepopulation of artifacts and laboratory analyses of a selectednumber of samples. Field measurements done on a large num-ber of objects will allow a first grouping and a first attempt atsource attribution, while more complete laboratory analysis willenable the separation of overlapping groups and the proposal ofmore secure source attribution. (2014b:51)

Their data analysis, however, focuses almost exclusively on the“mid-Z” trace elements (e.g., Rb, Sr, Ba, Nb, Zr) well measured byXRF, implying these techniques are adequate for distinguishingthese sources. Additionally, recent pXRF instruments have superiordetectors to lab-based systems built ten years ago, and for key el-ements, pXRF has uncertainties equal to that of NAA (Frahm et al.,2014a). Given a 1980s-era XRF instrument could distinguishArmenian obsidian sources (Keller and Seifried, 1990; Keller et al.,

modern cities referenced in the text (open squares). Localities with different names butc. localities of Gutansar). No endeavor is made here to precisely represent the entireuttle Radar Topography Mission dataset version 3), and base map shared and modified

E. Frahm / Journal of Archaeological Science 49 (2014) 105e125 107

1996), it would be surprising if a state-of-the-art pXRF instrumentcould not.

This paper examines the accuracy and reproducibility of pXRFdata, judged relative to lab-based techniques, based on analyses ofArmenian obsidians. Considered in light of the inter-laboratoryvariance that exists in published datasets, these new pXRF mea-surements are consistent, on average, with the data from five otherinstruments. The comparison also indicates that the pXRF data havebetter accuracy than the recently published LA-ICP-MS measure-ments. Three levels of reproducibility are also explored: (1) pXRFreproducibility relative to conventional analytical techniques, (2)reproducibility of handheld pXRF data relative to measurementstaken while the instrument was mounted in a stand, and (3)reproducibility with different instruments, analytical choices, andoperators. In each instance, reproducibility, judged by correlationbetween datasets, was high, typically equivalent to (or better than)that observed between any two instruments or techniques. Ulti-mately there is no reason to argue that pXRF is inherently inaccu-rate, lacks reproducibility, or is otherwise inadequate for obsidiansource characterization. A case study of the Pokr Arteni obsidiansource demonstrates the advantages of pXRF, including the abilityto analyze large numbers of specimens and investigate spatial re-lationships whilst in the field.

2. Background: evaluating analyses

Given concerns regarding the accuracy and reliability of pXRFdata, it is worth briefly discussing these concepts and their re-lationships to correlation, correction, and calibration.

2.1. Note about the term “pXRF”

Here “pXRF” refers to the handheld instruments roughly thesize, shape, and weight of a cordless drill. Some researchers (e.g.,Craig et al., 2007; Liritzis and Zacharias, 2011) consider “pXRF” toinclude benchtop instruments that can be transported from ananalytical facility to a similar context in a museum or field house.Such instruments have been used to good effect (e.g., Cecil et al.,2007; Liritzis, 2008), but the focus here is ruggedized in-struments that could be used to analyze outcrops.

2.2. Accuracy

Accuracy, as defined by United States’ National Institute ofStandards and Technology (NIST), is the “closeness of the agree-ment between the result of a measurement” and the actual (i.e.,accepted) value of the quantity being measured (Taylor and Kuyatt,1994). Analytical techniques have no predetermined accuracy, anda researcher’s choices, informed by theoretical and practical know-how (i.e., connaissances and savoir-faire in the nomenclature ofLemonnier, 1992), can affect accuracy as much as the hardwareconfigurations (e.g., Goldstein et al., 1981; Reed, 1996). Mark andWorkman (2003) note that evaluating accuracy is hard “becausethe usual statement of ‘accuracy’ compares the result obtained with‘truth’ [but] ‘truth’ is usually unknown, making this comparisondifficult” (214). Regarding this issue, Hancock and Carter (2010)explain that, “although analytical chemistry is not a democraticprocess, the agreement of specific elemental concentration databetween (among) independent analytical techniques adds credi-bility to the relative accuracy of their numbers” (245). Thus,analyzing specimens using multiple techniques is the principalapproach to evaluating accuracy, and generally, it can be assumedthe resulting mean values are reasonable approximations of the“true” values.

2.3. Reproducibility (reliability)

Precision is the degree to which measurements yield identicalvalues (Taylor, 1996). This can be a complex issue in sourcingstudies, which often involve data acquired over years and some-times with different instruments. For such reasons, NIST has notrecognized the concept of precision for decades, having divided theconcept into repeatability and reproducibility (Taylor and Kuyatt,1994). The former is the variability among sequential measure-ments under identical conditions, whereas the latter is variabilitywhen conditions, observers, or instruments change and/or aftertime has passed.

In a discussion of reliable obsidian compositional data, Hughes(1998) cites the classic definition from Carmines and Zeller(1979:11): “reliability concerns the extent to which an experi-ment, test, or any measuring procedure yields the same results onrepeated trials” (see also Frahm, 2012a). Thus, reliability is synon-ymous with reproducibility (Bartholomew, 1996:24). Often evalu-ation theorists include different observers in definitions ofreliability (e.g., Kaplan, 1964:200; Bohrnstedt and Knoke, 1988:14).Thus, by extension, reproducibility can involve other laboratoriesand/or analytical techniques.

2.4. Correlation

To assess reproducibility of measurements from different in-struments or techniques, correlation between the datasets must bedetermined (Glascock, 2011: Fig. 8.7; Speakman and Shackley,2013: Fig. 1). Correlation coefficients, often Pearson’s r and/or R2,describe the relationship between two datasets. A value of 1 for r orR2 would indicate perfect correlation. This assessment is usuallyaccomplished using regression analysis, whereby the correlationcoefficient describes the fit between the data and a predictorfunction (the slope of which can be used to assess accuracy orintercalibrate datasets).

2.5. Correction and calibration

For all forms of XRF, accuracy is principally affected by correc-tion and calibration, that is, how X-ray measurements are adjustedfor various phenomena (e.g., specimen density, X-ray absorption)and converted into element concentrations. Heginbotham et al.(2010) corroborated this based on data from 12 historical copperalloys measured by 19 XRF instruments in various laboratories.They concluded that the most important variable in accuracy wasuse of the fundamental parameters (FP) correction scheme withstandards for calibration:

Through this study, one common characteristic of higher-performing laboratories has become clear: the use of funda-mental parameters software, calibrated with standards. Incomparison, all other factors examined in this study appear tobe relatively poorly correlated with laboratory accuracy. [I]t isvery clear that laboratories using fundamental parameterssoftware calibrated with standards . performed consistentlymore accurately than laboratories using other methods.Remarkably, of these seven laboratories, no two used the sametype of instrument or the same brand of software. (187, 185)

Use of the “FP with standards” approach (rather than olderempirical or standardless approaches) trumps manufacturer orsoftware. Additionally, the number of standards did not correlatewith accuracy, nor did longer measurement times (e.g., 3 h/spec-imen in Gauthier and Burke, 2011), which can only improve


uncertainties and detection limits (i.e., phenomena affected bycounting statistics).

As Heginbotham et al. (2010) point out, not all XRF instrumentsare capable of implementing FP correction algorithms: “Theconsensus of the group is that options should be explored for waysin which existing instruments that currently use empirical orstandardless FP methods could be upgraded to use FP with stan-dards” (187). It must be emphasized that, with respect to pXRF,advocacy of non-FP approaches (e.g., empirical calibration curves,influence coefficients) is apparently tied to instruments that lackrapid and robust FP algorithms (e.g., Speakman, 2012; Conrey et al.,2014). This is evidenced by the dearth of comparisons between FP-and non-FP-correctedmeasurements from the same instrument (cf.Frahm et al., 2014b). Instead, as demonstrated by Conrey et al.(2014), comparisons involve the same basic correction approachapplied to a pXRF and an XRF instrument, canceling out systematicerror. As shown in Section 4.3, comparing independent analyticaltechniques (NAA, EMPA, LA-ICP-MS) to empirically corrected XRFand FP-corrected pXRF supports the conclusions of Heginbothamet al. (2010).

3. Evaluating accuracy

In this section, Armenian obsidian analyses with pXRF (thisstudy) and LA-ICP-MS (Chataigner and Gratuze, 2014a) arecompared to earlier datasets. Of the two new datasets, the pXRFmeasurements exhibit better agreement with previously publishedvalues (i.e., accuracy).

3.1. Previous studies

Prior assessments often compared pXRF values to those fromanother technique (e.g., Craig et al., 2010; Shackley, 2011;Dybowski, 2012; Williams et al., 2012; Kellett et al., 2013; Conreyet al., 2014 – there are, however, a few notable exceptions: Mili�c,2014 and Jia et al., 2013, for example, compare pXRF to threeother techniques). Caution is warranted with such comparisonsbecause neither dataset reflects “truth.” These comparisons simplyestablish the relative difference between two techniques, not theerror of either, and ignore “baseline” variability in measurementsfrom different labs.

Glascock (1999) published the results of a “round robin” testinvolving two obsidian specimens and ten datasets from eight labs,all experienced in obsidian analyses. Table 1 shows the measure-ments from this test and establishes the variance that should beanticipated among datasets from different labs and techniques. Themean coefficient of variation (CV; equal to the relative standarddeviation) was 0.24 for the Sierra de Pachuca obsidian and 0.19 forLittle Glass Buttes obsidian. There is no more complete or recentinter-laboratory dataset for obsidian analyses. Even NIST’s standardreference material (SRM) #278 (powdered obsidian rock) wascertified based on analyses in only four laboratories. In fact, no solidcertified obsidian standards currently exist, only powdered mate-rial (e.g., USGS RGM-1 and -2) and tiny fragments with demon-strated errors (Smithsonian VG-568, Frahm, 2010:386).

Table 1 also shows that mean concentration values and standarddeviations for the ten datasets in Glascock (1999). These means,which can be reasonably assumed to approximate the elements’“true” concentrations, are compared to recently published mea-surements of the same obsidians. Specifically, the relative differ-ences between the means and the newly reported values werecalculated. For instance, the new LA-TOF-ICP-MS data for LittleGlass Buttes obsidian was collected at the Institute for IntegrativeResearch in Materials, Environments, and Societies (IIRMES) atCalifornia State University-Long Beach (Scharlotta, 2010). It is

worth nothing that these IIRMES measurements are, on average,more than 30% different from the means in Glascock (1999).However, this has not, to my knowledge, yielded warningsregarding the perils of LA-TOF-ICP-MS for the future of obsidiansourcing.

3.2. Evaluation approach

There are two principal approaches to evaluating accuracy:first, measurements can be checked against specimens withestablished/accepted compositions, and second, specimens can bemeasured using multiple analytical techniques for comparison.Both approaches are considered here using a combination ofpublished and new datasets. Existing datasets for Armenianobsidian sources are scant. Several prior endeavors have goneunpublished (summarized in Badalyan et al., 2004) or focused ongeochronological, not geochemical, characterization of Armenianobsidians (e.g., Oddone et al., 2000; Badalian et al., 2001;Chataigner et al., 2003). Keller et al. (1996) analyzed 35 Arme-nian obsidian specimens with wavelength-dispersive XRF(WDXRF) for major and mid-Z elements. Several sources, how-ever, were absent, such as Aghvorik and the Tsakhkunyats sources.Blackman et al. (1998) used NAA to analyze 87 Armenian speci-mens, over half of which originated from one source. I analyzed151 Armenian obsidian specimens using EMPA, and I also sent 90matched specimens to MURR for energy-dispersive XRF (EDXRF)and NAA (Frahm, 2010, 2012b). These are the only multi-sourcedatasets currently available for systemic comparison to the newpXRF and LA-ICP-MS datasets. Cherry et al. (2010) tested 68Armenian obsidian specimens using NAA but only from the Syuniksource complexes in the southeast (Fig. 1).

First, as mentioned earlier, the mean values from multiple labsgenerally approximate elements’ true concentrations in a material.It is common for accuracy to be expressed as a percent relative erroror difference between the measurements and the true (i.e.,accepted) values. Table 1, for example, shows the relative differencebetween recent datasets and the mean values from ten differentdatasets summarized by Glascock (1999), which approximates thetrue values. For Sierra de Pachuca obsidian, Glascock’s (2011)EDXRF measurements have an average difference of 14% (trim-med mean) to 24% (mean) relative to the ten-dataset mean values.In contrast, for the Little Glass Buttes obsidian, Scharlotta’s (2010)LA-TOF-ICP-MS measurements have an average difference of 31%(trimmed mean) to 34% (mean) relative to the ten-dataset meanvalues. Consequently, accuracy would, in general, be consideredhigher for Glascock’s (2011) measurements and lower forScharlotta’s (2010) measurements. To assess the pXRF measure-ment accuracy (and that of recent LA-ICP-MS data from Chataignerand Gratuze, 2014a), the mean values from the five publisheddatasets using four analytical techniques (EMPA, NAA, EDXRF,WDXRF) are used as approximations of the true values, and percentrelative difference was calculated.

Second, as noted earlier, Glascock (1999) established that vari-ability exists among techniques and laboratories. If a new series ofmeasurements worsens the existing inter-technique variation, theCV will increase. Thus, accuracy evaluation can also involve addingthe pXRF and LA-ICP-MS measurements to the published datasetsfor Armenian obsidian sources. For the six Armenian obsidiansources in Table 2, CVs were calculated for the five pre-2014 data-sets. If one of the new datasets (pXRF or LA-ICP-MS) is inaccuraterelative to the published ones, adding it to the calculations willincrease variability among the measurements and, in turn, the CV.If, however, the new dataset is consistent with the publishedvalues, the CV will remain constant or decrease. In other words, anaccurate dataset would tend to decrease the variance, while an

Table 1Data from an obsidian “round robin” test conducted by Glascock (1999). Eight laboratories, all experienced in obsidian sourcing, provided ten datasets for two specimens: one from Sierra de Pachuca and Little Glass Buttes. Thedatasets demonstrate a “baseline” variability that should be expected among measurements from various laboratories and analytical techniques. For the elements of interest here, the mean coefficient of variation (CV) for theSierra de Pachuca obsidian, for example, is 0.24, which equals a relative standard deviation (RSD) of 24%. Here the “trimmed mean” reflects removal of the maximum and minimum values, resulting in a CV of 0.21 for Sierra dePachuca obsidian. Mean values and standard deviations were calculated for the datasets in Glascock (1999). These are compared to recently published measurements of the same obsidians, including the relative differencesbetween the means from Glascock (1999) and the newly reported values. For example, the new dataset for Little Glass Buttes obsidian was collected at the Institute for Integrative Research in Materials, Environments, andSocieties (IIRMES) at California State UniversityeLong Beach. The IIRMES LA-TOF-ICP-MS measurements are, on average, over 30% different from the mean values in Glascock (1999).

Sierra de Pachuca, Hidalgo, Mexico

Glascock, 1999 Glascock 2011

MURR CNRS Orléans Rio CNRS Grenoble ANSTO Rome Ashe Analytics NW Research Mean values EDXRF

NAA NAA LA-ICP-MS ICP-AES/MS ICP-AES/MS PIXE PIXE/PIGME WDXRF EDXRF EDXRF Mean 1s C.V. Mean 1s %Diff New C.V.

Ba 30 � 12 21 � 6 9 � 1 13 17 � 4 20 � 9 14 � 14 18 � 7 0.44Ca 1160 � 240 2362 � 105 640 750 � 11 769 � 45 894 � 31 786 1052 � 601 0.65Fe 15800 � 200 17200 � 1000 16600 � 100 16600 16200 � 600 15600 � 200 18100 � 1200 18900 16400 � 1100 17200 � 600 16860 � 1023 0.07 15,900 � 1200 6% 0.06 Y

Mn 1149 � 20 990 � 140 1048 � 40 837 1231 � 50 1008 � 24 1265 � 85 1161 1095 � 70 1124 � 48 1091 � 126 0.13 797 � 64 31% 0.14 [

Nb 91 � 1 116 � 9 84 91 � 1 97 � 13 91 91 � 4 99 � 2 95 � 10 0.11 84 � 8 12% 0.10 Y

Rb 192 � 3 219 � 13 203 � 1 181 185 � 2 211 � 8 224 � 15 194 200 � 8 225 � 3 203 � 16 0.09 189 � 3 7% 0.08 Y

Sr 2 � 1 2 2 � 1 3 � 1 5 2 � 1 5 � 9 3 � 1 0.54 10 � 4 108% 0.72 [

Ti 1300 � 60 1190 � 7 1050 1049 � 2 1028 � 29 1118 � 65 1258 1141 � 96 1142 � 101 0.10 945 � 144 19% 0.10Zn 191 � 12 240 � 2 210 � 2 206 221 � 4 292 � 19 224 � 14 224 � 8 226 � 30 0.15 207 � 19 9% 0.13 Y

Zr 888 � 40 1020 � 5 1058 � 143 796 1005 � 12 1008 � 13 1097 � 92 1055 991 � 35 965 � 9 988 � 88 0.10 957 � 62 3% 0.09 Y

Mean: 0.24 24% 0.18 Y

Trimmed mean: 0.21 14% 0.11 Y

Little Glass Buttes, Oregon, United States

Glascock, 1999 Scharlotta, 2010

MURR CNRS Orléans Rio CNRS Grenoble ANSTO Rome Ashe Analytics NW Research Mean values LA-TOF-ICP-MS

NAA NAA LA-ICP-MS ICP-AES/MS ICP-AES/MS PIXE PIXE/PIGME WDXRF EDXRF EDXRF Mean 1s C.V. Mean 1s %Diff New C.V.

Ba 1270 � 20 1550 � 130 1080 � 2 843 1237 � 14 1270 � 13 1338 � 14 1227 � 220 0.20 1173 � 93 4% 0.17 Y

Ca 5900 � 30 6813 � 69 5432 6219 � 246 5565 � 97 6230 � 50 5933 6013 � 463 0.09 2652 � 1083 78% 0.23 [

Fe 6200 � 100 6500 � 300 6840 � 280 6100 6500 � 360 6070 � 110 7020 � 230 7900 5730 � 150 6600 � 800 6546 � 611 0.10 5353 � 631 20% 0.11 [

Mn 327 � 6 297 � 30 269 � 5 303 333 � 10 291 � 11 357 � 9 387 298 � 13 349 � 47 321 � 36 0.12 299 � 36 7% 0.11 Y

Nb 12 � 1 9 � 1 8 8 � 1 6 7 � 1 8 � 2 8 � 2 0.26Rb 95 � 1 110 � 1 97 � 1 95 94 � 1 105 � 6 109 � 6 97 96 � 2 101 � 3 100 � 6 0.07 79 � 13 23% 0.09 [

Sr 78 � 20 52 � 1 66 87 � 1 73 � 5 81 � 6 71 69 � 3 73 � 9 72 � 10 0.15 45 � 17 46% 0.18 [

Ti 690 � 25 595 � 5 600 527 � 34 788 � 38 691 � 16 659 570 � 97 640 � 83 0.15 420 � 135 42% 0.17 [

Zn 31 � 7 90 � 3 27 � 1 29 27 � 2 36 � 1 24 � 3 41 � 7 38 � 22 0.64 27 � 26 34% 0.56 Y

Zr 118 � 7 99 � 10 83 � 7 106 106 � 2 105 � 8 107 � 3 105 96 � 3 109 � 8 103 � 9 0.10 64 � 9 47% 0.15 [

Mean: 0.19 34% 0.20 [

Trimmed mean: 0.15 31% 0.16 [

E.Frahm/Journal

ofArchaeological

Science49

(2014)105

e125

109

Table 2A summary of five pre-2014 datasetseWDXRF (Keller et al.,1996; abbreviated in the table as “Ke et al.1996”) andNAA (Blackman et al., 1998; abbreviated in the table as “Bl et al.1998”) aswell as EMPA, EDXRF, andNAA (Frahm,2010)e and twonewonesepXRF (this study) andLA-ICP-MS (Chataigner andGratuze, 2014a; abbreviated in the table as “ChandGr 2014a” and “Ch&Gr”) for six Armenian obsidian sources. Relative differences between the pre-2014meanvalues and each of the recent datasets are shown.Here the “trimmedmean” reflects removal of themaximumandminimumvalues. For each source, themean relative difference is smaller for the pXRFdata than the LA-ICP-MSdata (e.g., 9% and24%,respectively, for theAghvorik source). Additionally, three sets of CVswere calculated for (1) thefivepre-2014datasets, (2) thosedatasetswith thepXRFvalues added, and (3) thosedatasets with the LA-ICP-MS values added. The CVs indicate that the pXRF measurements are more consistent with the pre-2014 datasets than the LA-ICP-MS measurements.

Source: Aghvorik

Pre-2014 data 2014 data Coefficients of variation

Specimen AR.2009.63.1 Published values This study Ch and Gr, 2014a Pre-2014values

WithpXRFvalues

WithCh & Grvalues

Fr 2010 MURR 2010 Ke et al., 1996 Bl et al., 1998 Pre-2014 Mean pXRF LA-ICP-MS

EMPA EDXRF NAA WDXRF NAA Mean 1s Mean 1s %Diff Mean 1s %Diff

Ba 994 898 992 � 29 961 � 55 937 � 33 4% 752 � 25 41% 0.06 0.05 Y 0.13 [

Ca 9207 9207 9239 � 332 1% 10147 � 428 16% 0.00 0.07Fe 9224 10420 12158 12500 � 213 11076 � 1534 10487 � 77 9% 11766 � 756 10% 0.14 0.12 Y 0.12 Y

Mn 395 373 484 417 � 58 434 � 28 7% 412 � 39 2% 0.14 0.11 Y 0.11 Y

Nb 14 14 16 � 1 18% 18 � 1 38% 0.08 0.16Rb 90 99 115 � 2 101 � 13 97 � 1 7% 94 � 5 12% 0.12 0.11 Y 0.11 Y

Sr 236 206 221 � 21 226 � 4 4% 143 � 6 72% 0.09 0.07 Y 0.24 [

Ti 1865 1946 1906 � 57 2024 � 25 10% 1855 � 65 5% 0.03 0.04 [ 0.03Zn 53 51 43 � 1 49 � 5 58 � 3 29% 42 � 6 25% 0.11 0.12 [ 0.11Zr 205 224 218 244 � 15 223 � 16 229 � 4 5% 192 � 25 25% 0.07 0.06 Y 0.09 [

Mean: 9% 24% 0.10 0.08 Y 0.12 [

Trimmed mean: 8% 21% 0.10 0.08 Y 0.11 [

Source: Gutansar


Specimen AR.2009.47.5 Published values This study Ch and Gr 2014a Pre-2014values

WithpXRFvalues

WithCh & Grvalues

Fr 2010 MURR 2010 Ke et al. 1996 Bl et al. 1998 Pre-2014 Mean pXRF LA-ICP-MS


Ba 483 437 436 � 10 606 � 29 490 � 80 456 � 24 12% 363 � 25 50% 0.16 0.15 Y 0.19 [

Ca 6928 7147 � 101 7037 � 155 6495 � 220 13% 6717 � 714 8% 0.02 0.05 [ 0.03 [

Fe 8015 7726 7982 9572 � 83 8170 � 188 8293 � 733 7370 � 53 20% 8048 � 943 5% 0.09 0.09 0.08 Y

Mn 570 489 637 642 � 38 585 � 71 533 � 26 15% 566 � 51 5% 0.12 0.11 Y 0.11 Y

Nb 32 35 � 1 34 � 2 32 � 1 8% 33 � 2 3% 0.06 0.05 Y 0.04 Y

Rb 139 139 136 � 2 153 � 4 142 � 8 139 � 1 3% 137 � 9 6% 0.05 0.05 0.05Sr 133 139 126 � 2 133 � 6 140 � 2 9% 87 � 12 69% 0.05 0.05 0.19 [

Ti 1044 1903 1079 � 35 1342 � 486 1155 � 13 25% 1002 � 71 48% 0.36 0.31 Y 0.34 Y

Zn 52 43 39 � 1 44 � 7 52 � 2 26% 41 � 6 14% 0.15 0.14 Y 0.13 Y

Zr 140 177 131 169 � 5 191 � 18 162 � 25 156 � 2 6% 121 � 13 48% 0.16 0.14 Y 0.18 [

Mean: 14% 26% 0.12 0.11 Y 0.14 [

Trimmed mean: 14% 23% 0.12 0.10 0.12 [

Source: Hatis


Specimen AR.2009.48.1 Published values This study Ch and Gr, 2014a Pre-2014values

WithpXRFvalues

WithCh & Grvalues

Fr 2010 MURR 2010 Ke et al., 1996 Bl et al., 1998 Pre-2014 mean pXRF LA-ICP-MS


Ba 599 522 519 � 4 726 � 40 592 � 97 586 � 18 2% 450 � 24 45% 0.16 0.14 Y 0.19 [

Ca 7727 7040 � 51 7383 � 486 7621 � 230 5% 7074 � 786 7% 0.07 0.05 Y 0.05 Y

Fe 6746 7358 7575 7112 � 275 6100 � 55 6978 � 580 6872 � 55 3% 6184 � 924 20% 0.08 0.07 Y 0.09 [

Mn 388 408 493 503 � 55 448 � 58 443 � 21 2% 462 � 26 5% 0.13 0.11 Y 0.11 Y

Nb 17 19 � 1 18 � 1 19 � 1 7% 21 � 2 24% 0.06 0.05 Y 0.09 [

Rb 105 106 107 � 1 126 � 6 111 � 10 104 � 2 11% 108 � 7 5% 0.09 0.08 Y 0.08 Y

Sr 148 183 112 � 1 148 � 35 152 � 2 5% 82 � 6 95% 0.24 0.19 Y 0.33 [

Ti 781 1418 629 � 40 943 � 419 952 � 13 2% 626 � 18 67% 0.44 0.36 Y 0.44Zn 35 33 31 � 1 33 � 2 54 � 10 79% 37 � 4 17% 0.07 0.27 [ 0.08 [

Zr 60 107 77 87 � 6 149 � 13 96 � 34 87 � 2 17% 63 � 4 69% 0.35 0.32 Y 0.37 [

Mean: 13% 36% 0.17 0.17 0.18 [

Trimmed mean: 6% 32% 0.15 0.16 [ 0.17 [

Source: Mets Satankar



WithpXRFvalues

WithCh & Grvalues



Ba 42 0 17 � 8 65 31 � 28 32 � 13 6% 13 � 1 136% 0.92 0.79 Y 0.95 [

Ca 3185 3466 � 277 3325 � 199 3528 � 189 10% 3501 � 357 9% 0.06 0.05 Y 0.05 Y


Table 2 (continued )

Source: Mets Satankar



WithpXRFvalues

WithCh & Grvalues



Fe 3951 4956 4177 5344 � 204 4920 4670 � 583 4631 � 337 1% 4696 � 182 1% 0.12 0.11 Y 0.11 Y

Mn 450 365 560 523 � 74 475 � 86 468 � 23 2% 510 12% 0.18 0.16 Y 0.16 Y

Nb 30 35 � 2 33 � 3 32 � 1 4% 35 � 1 11% 0.10 0.07 Y 0.08 Y

Rb 193 198 188 � 15 192 193 � 4 190 � 3 2% 193 � 1 0% 0.02 0.02 0.02Sr 9 0 18 � 7 9 � 9 5 � 1 93% 6 � 1 65% 1.01 0.96 Y 0.92 Y

Ti 512 504 480 � 49 499 � 17 564 � 65 20% 484 � 9 5% 0.03 0.07 [ 0.03Zn 34 35 35 35 � 1 46 � 3 48% 46 48% 0.02 0.15 [ 0.15 [

Zr 62 92 47 85 � 3 123 82 � 29 83 � 4 2% 57 � 1 60% 0.36 0.32 Y 0.36Mean: 19% 35% 0.28 0.27 Y 0.28Trimmed mean: 19% 26% 0.22 0.27 0.23 [

Source: Pokr Sevkar



WithpXRFvalues

WithCh & Grvalues



Ba 64 37 43 � 3 78 55 � 19 39 � 12 58% 32 � 4 89% 0.34 0.34 0.38 [

Ca 3520 3395 � 51 3458 � 89 3695 � 225 11% 3858 � 428 18% 0.03 0.04 [ 0.07 [

Fe 4741 5573 4713 5635 � 165 4840 5100 � 463 4758 � 52 12% 4260 � 286 30% 0.09 0.09 0.11 [

Mn 407 330 463 426 � 55 407 � 56 405 � 27 1% 391 � 22 7% 0.14 0.12 Y 0.12 Y

Nb 26 32 � 1 29 � 4 28 � 1 6% 30 � 2 6% 0.15 0.11 Y 0.10 Y

Rb 173 172 170 � 1 191 177 � 10 170 � 2 6% 167 � 8 9% 0.05 0.05 0.05Sr 19 0 19 � 1 13 � 11 14 � 1 17% 11 � 1 23% 0.87 0.69 Y 0.73 Y

Ti 573 489 570 � 42 544 � 47 617 � 7 21% 522 � 40 7% 0.09 0.09 0.07 Y

Zn 33 33 34 33 � 1 45 � 3 50% 31 � 8 11% 0.02 0.16 [ 0.04 [

Zr 68 98 71 92 � 10 117 89 � 20 90 � 5 2% 64 � 4 55% 0.23 0.20 Y 0.25 [

Mean: 18% 26% 0.20 0.19 Y 0.19 Y

Trimmed mean: 18% 20% 0.20 0.19 [ 0.14

Source: Tsahkhunyats-Damlik



WithpXRFvalues

WithCh & Grvalues



Ba 830 700 902 � 41 811 � 102 773 � 26 8% 586 � 76 54% 0.13 0.11 Y 0.19 [

Ca 5558 5558 5362 � 188 6% 6145 � 357 17% 0.03 0.07Fe 6183 6101 6208 5680 � 102 6043 � 246 5915 � 80 4% 5622 � 914 12% 0.04 0.04 0.05 [

Mn 456 399 458 438 � 34 420 � 26 7% 437 � 31 0% 0.08 0.07 Y 0.06 Y

Nb 16 16 20 � 1 41% 21 � 2 49% 0.17 0.21Rb 82 93 122 � 7 99 � 21 93 � 2 11% 103 � 9 6% 0.21 0.18 Y 0.17 Y

Sr 197 246 221 � 35 211 � 5 8% 116 � 19 104% 0.16 0.12 Y 0.35 [

Ti 651 1237 944 � 414 915 � 12 5% 612 � 48 71% 0.44 0.31 Y 0.42 Y

Zn 36 29 30 � 1 32 � 4 42 � 3 46% 32 � 1 1% 0.13 0.18 [ 0.11 Y

Zr 64 106 81 130 � 11 95 � 29 93 � 3 4% 65 � 11 63% 0.30 0.26 Y 0.32 [

Mean: 14% 38% 0.18 0.15 Y 0.19 [

Trimmed mean: 11% 38% 0.17 0.15 Y 0.18 [


inaccurate one would increase it. A perfectly accurate set of in-struments would yield data with a CV of zero (CVs are also used atMURR to assess data quality; Glascock, 2011). Adding an inaccurateinstrument would yield higher CVs. That is, an inaccurate mea-surement raises variance since it is farther from the dataset’s mean,which estimates the “true” value.

Table 1, for example, shows the CVs for the ten datasets inGlascock (1999) and how they change with the addition of therecent datasets. For Sierra de Pachuca obsidian, Glascock’s (2011)EDXRF data decrease the overall variation, reflecting good agree-ment with the other datasets and, thus, high accuracy. In contrast,for Little Glass Buttes obsidian, Scharlotta’s (2010) LA-TOF-ICP-MSmeasurements increase the overall variation, suggesting pooragreement and, thus, lower accuracy. It should be emphasized that

this is not an assessment of LA-TOF-ICP-MS as an analytical tech-nique. Rather, the particular methods and choices applied byScharlotta (2010) affected the resulting accuracy.

3.3. Materials and methods

This test used a Thermo Scientific Niton XL3t GOLDDþ pXRFinstrument. It generates X-rays with a 50-kV silver-anode tube, isoutfitted with a silicon drift detector (SDD), and has, in practice, anX-ray resolution better than 155 eV, comparable to lab-basedEDXRF systems. Niton analyzers monitor and automatically adjusttheir tube currents (up to 200 mA) to attain optimal X-ray countrates. A built-in camera assisted specimen positioning, minimizinggeometric errors. Each analysis took 120 s (i.e., sequential 40-


s measurements with three built-in X-ray filters). Ten well-measured elements e Ba, Ca, Fe, Mn, Nb, Rb, Ti, Sr, Zn, and Zr e

are the focus of interest here.Factory-set calibrations of pXRF analyzers strive for accuracy

over a wide range of materials and element concentrations. Forexample, for environmental surveying, the calibration must ac-count for, on one hand, loess in which toxic metals occur near thedetections limits and, on the other hand, much higher concentra-tions in contaminated river sediments. The intention is that these“internal” calibrations can be used in various tasks and acquiresuitable data over a wide range of compositions. Applications suchas obsidian sourcing, in which trace elements and the overallcompositions fall into narrow ranges, tend to reveal the limitationsof factory-set calibrations. Analyses of standards are typicallyneeded to “fine tune” the calibrations to yield accurate measure-ments. The “FP with standards” approach has been shown to resultin the most accurate XRF data (Heginbotham et al., 2010), so thiswas the correction and calibration approach used here and for on-going projects in Armenia (Adler et al., 2013 at Lusakert Cave 1 andNor Geghi 1; Glauberman et al., 2013 at Barozh 12; Kandel et al.,2013 at Aghitu 3).

The factory calibrationwas supplemented by regression analysisbased on 24 obsidian specimens measured using NAA and EDXRF atMURR (methods summarized in Glascock et al., 2007) and electronmicroprobe analysis (EMPA; summarized in Frahm, 2012b). Thesespecimens, all calcalkaline obsidians from sources throughoutSouthwest Asia (e.g., Turkey, Georgia), were prepared according toMURR’s practices for solid (not powdered) obsidian standards fornondestructive XRF (Glascock, 2011; Glascock and Ferguson, 2012).

Fig. 2. Linear regression calibrations were derived based on 24 obsidian specimens analyzSouthwest Asia. Peralkaline obsidians (e.g., Nemrut Daǧ, Bingöl A) were excluded. MeasuremEDXRF (MURR), and EMPA (Minnesota) data for the same specimens. The resulting calibrat

Each of these three techniques (NAA and EDXRF at MURR, EMPA atMinnesota) has been independently assessed for accuracy in waystoo extensive to summarize here. For example, the EMPA methodshave been evaluated using certified standards, published data, in-ternational round robins, and other inter-laboratory comparisons(Frahm, 2010:380e435, 2012b).

Fig. 2 shows the regression analysis for six elements. The meanvalues of the EDXRF, NAA, and EMPA measurements were, for thepurposes of calibration, considered suitable approximations of theelements’ true concentrations. Correlations between these meanvalues and the factory-calibrated pXRF data are high (R2 ¼ 0.94e0.99), indicating that the FP algorithms are mitigating nonlinearphenomena in the measured X-ray intensities. Slopes of theseregression equations, however, vary markedly. For two elements(Ca and Zr), the slopes are typical of comparisons between fullycalibrated data from any two XRF instruments (Speakman andShackley, 2013). Other elements have slopes further from 1 (e.g.,1.88 for Rb), demonstrating the factory calibration yields inaccuratevalues for this range (75e225 ppm) in this particular matrix(rhyolitic glass). Yet the high correlation (R2 ¼ 0.99) enables cali-bration and, in turn, accurate values when these regression equa-tions are applied. In the instrument’s onboard software, these slopeand intercept values can be entered as “Cal Factors” in the “AdjustCalibration” menu.

Although the effect of surface geometry on pXRF accuracy hascaused concern (e.g., Shackley, 2010:19; Liritzis and Zacharias,2011:132), geological obsidian specimens were analyzed withoutcutting or polishing, but each measurement was conducted on thefreshest, flattest surface.

ed 14 times each (i.e., 336 analyses). All specimens were calcalkaline obsidians froments using the factory-set calibrations are plotted against mean values for NAA (MURR),ion equations for six elements are shown here as examples.


3.4. Results

Table 2 summarizes five pre-2014 datasets and two recent onesfor six Armenian sources. Both accuracy assessment approaches areshown. First, for each source, the mean element concentrationswere calculated for the five earlier datasets. The percent relativedifference was calculated for the pXRF (this study) and LA-ICP-MS(Chataigner and Gratuze, 2014a) measurements with respect tothese mean values. In each case, the relative difference is lower forthe pXRF. For Hatis obsidian, as an example, the pXRF measure-ments exhibit an overall relative difference of 6% (trimmed mean)to 13% (mean), whereas the LA-ICP-MS measurements have a dif-ference greater than 30%. For Aghvorik obsidian, the pXRF datahave an overall difference of 8e9% relative to the pre-2014 datasets,whereas the LA-ICP-MS data have a relative difference greater than20%. For all six sources, the overall relative difference is more thantwice as high for the LA-ICP-MS values (i.e., 15% for pXRF versus 31%for LA-ICP-MS based on the means, and 11% for pXRF versus 26% forLA-ICP-MS based on the trimmed means).

Second, three sets of CVs were calculated for each element and,on average, each source: (1) the five pre-2014 datasets, (2) thosedatasets with the pXRF data, and (3) those datasets with the LA-ICP-MS data. When the pXRF data are added to the calculations,five sources have lower overall CVs, whereas one source has anunchanged CV (only for the trimmedmeans are there two instancesof small increases). When the LA-ICP-MS data are added instead,

Fig. 3. Correlations between the pXRF (a) and LA-ICP-MS (b) measureme

the overall CVs are higher for four sources, unchanged for one, andlower for one (although the trimmed mean for this case is un-changed). This further indicates the pXRF data are more consistentwith the pre-2014 datasets than the LA-ICP-MS data.

4. - Reproducibility: laboratories and techniques

Speakman et al. (2011) express concern about pXRF reproduc-ibility: “Researchers who generate (or incorporate) portable XRFdata into their programs should question how the data compare todata generated by other analytical techniques. Are they reproduc-ible?” (3484). This section uses the same methods and measure-ments as Section 3 to examine how well the pXRF and LA-ICP-MSmeasurements reproduce each of the five datasets publishedearlier.

4.1. - Previous studies

In this sense, reproducibility concerns whether pXRF measure-ments can be used with those from other analytical techniques. Afew researchers have examined pXRF reproducibility explicitlyfrom this perspective. Forster and Grave (2012) compared pXRFdata of obsidian artifacts to source data from the literature. Theynoted “a slight offset” (735) but concluded that the pXRF data canbe “integrat[ed] with already published results” (736). Millhauseret al. (2011) compared their pXRF artifact data to EDXRF

nts and the other published datasets for Ba in Armenian obsidians.


geological data and were able to attribute 97% of artifacts to thecorrect obsidian sources. Nazaroff et al. (2010) similarly comparedpXRF and EDXRF values. A systematic difference was identified, butfurther data processing yielded insignificant variance between thetwo datasets. The offset was due to systematic error, so it mighthave been a product of either instrument’s empirical calibration.

Regarding laboratory-based instruments, the literature hintsthat their perceived reproducibility is superior to reality. As acounterpoint to pXRF, Shackley (2012) claims, regarding XRFmeasurements in the “big three” XRF-based North Americanobsidian sourcing laboratories, “most of the measurements arewithin 1%” (also “often only 1%” in Shackley, 2011:33). However, thedatasets provided as evidence (Shackley, 2012: Table 2) differ from3% (Fe, Sr) to 9% or more (Mn, Nb, Rb, Y).

4.2. - Evaluation approach

Reproducibility between techniques or laboratories is oftenjudged by measurements on one or two standards, but overallreproducibility over the range of elemental concentrations found inArmenian obsidians is ultimately of interest. Thus, reproducibility isjudged here based on correlation between the datasets from twotechniques. If two datasets are perfectly correlated, values from onedataset perfectly reproduce the other dataset. If, though, the mea-surements from one dataset do not reproduce those from another,theywill be poorly correlated. Thus, correlation can serve as a proxy

Fig. 4. Correlations between the pXRF (a) and LA-ICP-MS (b) measureme

for reproducibility. This relationship is perhaps unintuitive butstatistically sound. Correlation is a normalized form of covariance(which can involve two different variables or independent mea-surements of a variable such as elemental concentration), whichmay be defined as the inverse or reciprocal of reproducibility.Hence, correlation can indicate reproducibility through this sta-tistical relationship.

4.3. - Results

Figs. 3 and 4 reveal the correlations between (1) the pXRF or LA-ICP-MS data and (2) the other datasets for two elements. For Ba(Fig. 3), correlation between the pXRF data and the other recent(2010e2014) datasets is nearly perfect: R2 and r¼ 1.00. Even for theolder data, correlation remains high: R2¼ 0.93e0.96, r¼ 0.96e0.98.LA-ICP-MS correlation to other datasets is almost identical. Simi-larly, for Rb, (Fig. 4) correlation between the pXRF data and recentdatasets is very high (R2 ¼ 0.98e1.00, r ¼ 0.99e1.00), and corre-lation to the older datasets is high (R2 ¼ 0.96, r ¼ 0.98). LA-ICP-MScorrelation to the other datasets is only marginally lower (circa 1e2% relative).

Table 3 summarizes correlations for all ten elements. For sevenelements, the mean r values for the pXRF and LA-ICP-MS datasetsrelative to the others are 0.98 or better. For the other three ele-ments (Mn, Ti, Zn), the pXRF and LA-ICP-MS data both have loweraverage correlations to the other datasets, but not all of them. For

nts and the other published datasets for Rb in Armenian obsidians.


Zn, this appears to result from its low concentrations and narrowvariation in Armenian obsidians (circa 35e55 ppm). Consider,though, Mn. The pXRF data exhibit high correlations to the MURRNAA and LA-ICP-MS values (r ¼ 0.98 and 0.97, respectively), butcorrelations are lower relative to the EDXRF and WDXRF values(r ¼ 0.77). Similarly, for Ti, pXRF data are highly correlated to theEMPA and LA-ICP-MS data (r ¼ 0.99 and 0.98), but correlations arelower relative to the EDXRF and WDXRF data (r ¼ 0.87 and 0.91).Thus, the pXRF instrument yielded superior reproducibility for Mnand Ti than the two lab-based XRF instruments. It seems that theempirical corrections used for the two lab-based XRF instrumentsresulted in less reproducible measurements than the “FP withstandards” approach used with the pXRF instrument, consistentwith Heginbotham et al. (2010).

It is also worth observing that, in general, the slopes of theregression lines for the pXRF data are closer to one than those forthe LA-ICP-MS data. This, however, is usually interpreted as a proxyfor the measurements’ accuracy, not their reproducibility (cf.Speakman and Shackley, 2013).

5. - Reproducibility: handheld versus test stand

Analyses were also conducted to consider data reproducibility inhandheld versus test-stand use. A series of 16 specimens weremeasured in two trials, once while handheld and once with theinstrument mounted in a test stand. The resulting datasetsdemonstrate high reproducibility.


Earlier obsidian studies have almost always used a pXRF whilemounted in a test stand due to its use in a laboratory or similarcontext (cf. Frahm et al., 2014a). When researchers report the in-strument was mounted in a test stand, it is often unclear if themotivation was data quality, convenience, or rules imposed by asafety organization. When a reason is given, it frequently relates toreproducible data. For example, Goodale et al. (2012) note that astand “ensure[d] the instrument was physically stable which inprevious experience of the authors was found to be important forobtaining reliable results” (879). The effect on reproducibility,however, has not been quantitatively demonstrated.


This test consisted of two trials in which there was only onedifference: operating the instrument in hand or mounted in a teststand. The measurements from a series of specimens analyzed inboth trials were compared to establish the effect, or lack thereof, ondata reproducibility.

5.3. - Materials and methods

This test used a set of 16 synthetic specimens, which spanned abroader compositional spectrum than possible with obsidians. Thespecimens were composites of (1) the sub-30-mm fraction of levi-gated clay (Kilikoglou et al., 1998) and (2) submillimeter quartzgrains (�99% SiO2; 100e750 mm particles). By adding quartz invaried fractions (5e40%), the specimens’ bulk compositions wereadjusted systemically (e.g., 50e70% SiO2, 12e20% Al2O3, 1e3% K2O,3e7% FeOT), spanning the major-element concentrations typical ofbasalts through rhyolites. After drying and firing at 950 �C, thespecimens were cut to produce rectangular prism bars approxi-mately 10 � 10 � 60 mm in size.

As in Sections 3 and 4, the instrument was a Thermo ScientificNiton XL3t GOLDDþ. The only variable between the two trials was

mode of operation: handheld or test-stand use. Five measurementsper specimenwere takenwhile handheld, and fifteen per specimenin the test stand. Each measurement was taken in a different spot.The built-in camera assisted in aligning specimens with the mea-surement window (10 mm diameter). Measurements were FPcorrected and calibrated using four SRMs: CCRMP TILL-4, NRCCRMGBW07411, IAEA SOIL-7, and NCS DC73308. Each analysis was 90 s(30 s with three filters) to measure nine of the ten elements inSections 3 and 4 (i.e., Ba was unmeasured to remove the need for afourth X-ray filter, yielding shorter measurement times).

5.4. - Results

Fig. 5 shows correlations between the two datasets: R2 ¼ 0.97e1.00 for all elements (cf. R2 ¼ 0.98e1.00 for pXRF vs. EDXRF inSpeakman and Shackley, 2013; R2¼ 0.98e1.00 for EDXRF vs. NAA inGlascock, 2011). Additionally, the slopes fall between 0.92 and 1.08,indicating a relative difference between these data of 8% or less (cf.0.80e1.21 in Speakman and Shackley, 2013; 0.92e1.17 in Glascock,2011). High reproducibility was observed for a trace element nearthe detection limits (Nb,10e40 ppm) and elements at much greaterconcentrations. Thus, pXRF can be highly reproducible betweenhandheld and test-stand operation and, in turn, between laboratoryand field use.

6. - Reproducibility: instruments, methods, and users

Another concern is whether pXRF users can take measurementsthat are reproducible relative to other users with the same ordifferent pXRF instruments. Studies have already started to inte-grate pXRF data collected with the same instrument but bydifferent users (Tykot et al., 2013b), and obsidian sourcing with apXRF instrument has been used as an undergraduate pedagogicalexercise (Clark and Mayo, 2012). Consequently, a test was con-ducted to initially explore the effects of different instruments,users, choices, and skill on the reproducibility of pXRFmeasurements.


The literature seems to contain more criticisms of pXRF usersthan of the technique itself (e.g., Shackley, 2010:18; Speakmanet al., 2011:3483; Speakman and Shackley, 2013:1435). Such criti-cisms are more concerned with theoretical knowledge (e.g., thephysics of X-ray emission) rather than the practical know-howinvolved in acquiring data with pXRF (e.g., how to orient a hand-held instrument with respect to specimens). Skill in wielding theinstrument seems important for reproducibility (Mauss, 1973;Leroi-Gourhan, 1993), but this issue has gone unaddressed. Fewresearchers have examined the reproducibility of pXRF measure-ments with respect to other pXRF datasets. In each instance, themain variable was the instrument model, and other variables wereminimized: Goodale et al. (2012) tested two InnovX models againsteach other using a collection of obsidians, and Brandt et al. (2013)compared a number of Bruker Tracer III analyzers with identicalconditions, calibrations, and procedures.


After the requisite safety training, twenty participants in fourgroups, each with a different pXRF instrument (Table 4), analyzedthe same specimens as Section 5. Participants’ backgrounds variedfrom entirely archaeological to having degrees in the sciences andinformation technology. When asked to rank their prior knowledgeof pXRF from 1 to 5, most participants (71%) selected 2 or 3. For this

Table 3Correlations between the pXRF (left columns) or LA-ICP-MS (right columns) datasets and each of the other six datasets for Armenian obsidian sources. Correlations to other sixdatasets are, on average, slightly higher for the pXRF data (e.g., an overall r of 0.94 for pXRF and 0.93 for LA-ICP-MS). Also the slopes of the regression lines are, on average,slightly closer to 1 for the pXRF data (e.g., an overall 17% difference for pXRF and 21% difference for LA-ICP-MS).

pXRF e This Study LA-ICP-MS e Chataigner and Gratuze, 2014a

Slope Correlation Slope Correlation

Value Diff R2 r Value Diff R2 r

Ba EMPA Frahm 2010 0.96 4% 1.00 1.00 0.77 23% 0.99 0.99EDXRF MURR 2010NAA MURR 2010 1.04 4% 1.00 1.00 0.83 17% 1.00 1.00LA-ICP-MS Chataigner and Gratuze 2014a 1.25 25% 1.00 1.00pXRF This study 0.80 20% 1.00 1.00WDXRF Keller et al., 1996 1.01 1% 0.93 0.96 0.81 19% 0.95 0.97NAA Blackman et al., 1998 0.94 6% 0.96 0.98 0.75 25% 0.94 0.97

Mean: 8% 0.98 0.99 21% 0.98 0.99Ca EMPA Frahm 2010 0.93 7% 0.99 0.99 0.97 3% 0.95 0.97

EDXRF MURR 2010NAA MURR 2010LA-ICP-MS Chataigner and Gratuze 2014a 0.91 9% 0.96 0.98pXRF This study 1.05 5% 0.96 0.98WDXRF Keller et al., 1996 0.93 7% 0.95 0.97 0.85 15% 0.98 0.99NAA Blackman et al., 1998

Mean: 8% 0.97 0.98 8% 0.96 0.98Fe EMPA Frahm 2010 0.98 2% 0.93 0.96 1.20 20% 0.89 0.94

EDXRF MURR 2010 1.10 10% 0.99 0.99 1.35 35% 0.96 0.98NAA MURR 2010 0.76 24% 1.00 1.00 0.93 7% 0.96 0.98LA-ICP-MS Chataigner and Gratuze 2014a 0.79 21% 0.98 0.99pXRF This study 1.24 24% 0.98 0.99WDXRF Keller et al., 1996 0.68 32% 0.91 0.95 0.83 17% 0.97 0.98NAA Blackman et al., 1998 0.75 25% 0.96 0.98 0.95 5% 0.99 0.99

Mean: 19% 0.96 0.98 18% 0.96 0.98Mn EMPA Frahm 2010 0.57 43% 0.87 0.93 0.73 27% 0.82 0.91

EDXRF MURR 2010 0.91 9% 0.60 0.77 1.12 12% 0.58 0.76NAA MURR 2010 0.72 28% 0.97 0.98 0.92 8% 0.94 0.97LA-ICP-MS Chataigner and Gratuze 2014a 0.74 26% 0.95 0.97pXRF This study 1.27 27% 0.95 0.97WDXRF Keller et al., 1996 0.34 66% 0.59 0.77 0.44 56% 0.62 0.79NAA Blackman et al., 1998

Mean: 34% 0.80 0.89 26% 0.78 0.88Nb EMPA Frahm 2010

EDXRF MURR 2010 0.86 14% 0.97 0.98 0.92 8% 0.95 0.97NAA MURR 2010LA-ICP-MS Chataigner and Gratuze 2014a 0.93 7% 0.99 0.99pXRF This study 1.07 7% 0.99 0.99WDXRF Keller et al., 1996 0.69 31% 0.91 0.95 0.75 25% 0.88 0.94NAA Blackman et al., 1998

Mean: 17% 0.96 0.98 13% 0.94 0.97Rb EMPA Frahm 2010


Mean: 5% 0.98 0.99 7% 0.96 0.98Sr EMPA Frahm 2010

EDXRF MURR 2010 1.02 2% 0.99 0.99 0.59 41% 1.00 1.00NAA MURR 2010 0.91 9% 0.97 0.98 0.51 49% 0.92 0.96LA-ICP-MS Chataigner and Gratuze 2014a 1.71 71% 0.99 0.99pXRF This study 0.58 42% 0.99 0.99WDXRF Keller et al., 1996 1.29 29% 0.97 0.98 0.73 27% 0.99 0.99NAA Blackman et al., 1998

Mean: 28% 0.98 0.99 40% 0.98 0.99Ti EMPA Frahm 2010 1.05 5% 0.98 0.99 1.02 2% 0.99 0.99

EDXRF MURR 2010 0.68 32% 0.76 0.87 0.61 39% 0.65 0.81NAA MURR 2010LA-ICP-MS Chataigner and Gratuze 2014a 1.02 2% 0.96 0.98pXRF This study 0.95 5% 0.96 0.98WDXRF Keller et al., 1996 0.96 4% 0.83 0.91 0.88 12% 0.99 0.99NAA Blackman et al., 1998

Mean: 11% 0.88 0.94 15% 0.90 0.94Zn EMPA Frahm 2010

EDXRF MURR 2010 0.45 55% 0.53 0.73 0.30 70% 0.24 0.49NAA MURR 2010 0.67 33% 0.69 0.83 0.49 51% 0.37 0.61LA-ICP-MS Chataigner and Gratuze 2014a 0.57 43% 0.32 0.57pXRF This study 0.56 44% 0.32 0.57


Table 3 (continued )

pXRF e This Study LA-ICP-MS e Chataigner and Gratuze, 2014a

Slope Correlation Slope Correlation

Value Diff R2 r Value Diff R2 r

WDXRF Keller et al., 1996NAA Blackman et al., 1998 0.91 9% 0.49 0.70 0.79 21% 0.37 0.61

Mean: 35% 0.51 0.71 47% 0.33 0.57Zr EMPA Frahm 2010 1.01 1% 0.98 0.99 0.89 11% 0.98 0.99


Mean: 9% 0.97 0.99 13% 0.97 0.98Overall mean: 17% 0.90 0.94 21% 0.87 0.93


test, reproducibility is assessed relative to the mean values fromtest-stand measurements in Section 5.

6.3. - Materials and methods

Four groups of operators with different pXRF instrumentsanalyzed the same specimens tested in Section 5. In addition tothe XL3t GOLDDþ analyzer, the others were a Niton XL2, XL2GOLDD, and XL3t. Important instrument parameters are summa-rized in Table 4. Certain variables were not under participants’control. When each group chose the FP-based “mining” mode,many instrument variables thought significant (e.g., tube current,voltage, filters; Shackley, 2011:25e30) could not be altered, andthe instruments’ operating conditions were, in many respects,more alike than different. Each group selected three SRMsfrom Section 5.3 for calibration. The specimens were measuredthree times each, but the measurement times were chosen byeach group. Two groups decided to change operators betweeneach specimen, whereas the other two changed operators for eachmeasurement.

Each instrument affected how participants physically tookmeasurements. Two instruments (the XL3t and XL3t GOLDDþ) hadcameras to aid aligning specimens with the measurement window.Thus, operators of those two instruments had feedback that theothers did not. Additionally, the XL2 and XL2 GOLDD are about0.25 kg heavier than the others. Group 4 used the XL3t GOLDDþwith a test stand, but the other three used the instrument in hand.The XL2 also had an additional safety feature enabled: not only didthe trigger need to be held, but also a button on the back of theinstrument had to be pressed simultaneously. That is, one had toalign the instrument and specimen while exerting forces in oppo-site directions. Unsurprisingly, Group 1 reported this as a consid-erable hindrance.

In many respects, the situation of Group 1 represents a worstcase scenario for acquiring data that reproduce the test-standdataset from Section 5: the analyzer was heavier and held byhand, the detector was the prior generation, therewas no camera toaid specimen orientation, the operators had considerably less “in-hand” experience and changed between measurements, and twobuttons had to be simultaneously pressed, increasing physicalinstability in the tube-specimen-detector arrangement.

6.4. - Results

Fig. 6 shows the results for Rb and Sr. Despite all the differences,the R2 and r values for Rb are consistently high: 0.95e0.99 and0.97e0.99, respectively. The values for Sr, though, improve steadily

from Group 1 (R2 ¼ 0.87, r ¼ 0.93) to 4 (R2 and r ¼ 0.99). Slopes ofthe regression lines also markedly improved between Group 1(Rb ¼ 0.76, Sr ¼ 0.77) and 4 (Rb ¼ 1.02, Sr ¼ 0.98). Therefore, theGroup 4 values, on average, differ by just 2% from the test-standdata in Section 5. These correlations are similar to those betweenEDXRF and NAA datasets from MURR: r ¼ 0.99e1.00 in Glascock(2011) for Mexican obsidians and r ¼ 0.95e0.99 in Frahm (2013)for Southwest Asian obsidians.

Table 4 lists these values for all elements. Like Sr, the mean R2

and r values steadily increase from Group 1 to 4. The mean valuesfor Group 1 are misleading due to two poorly measured elements.Nb occurs in the obsidians near detection limits for that instrument(10e40 ppm), so analytical challenges (e.g., handheld use with aheavier instrument, no camera, pressing two buttons simulta-neously) strongly affected its measurement. Also, while Ti can bedetected with the “main” X-ray filter, its quantification at lowconcentrations requires the “low” filter, which Group 1 did not use(to shorten their analyses). If Nb and Ti are excluded, Group 1’smean R2 value increases to 0.89 and r to 0.94.

The increasing reproducibility from Group 1 to 4 cannot beexplained by participants’ analytical choices (i.e., correction, cali-bration, measurement times, etc. weremore alike than different) orscientific knowledge. Technical variables alone (e.g., detector, filtercomposition) cannot explain these differences either. Instead,reproducibility seems to reflect how users physically interactedwith the instrument. It is higher with an internal camera, a teststand, or a lighter instrument. It is lower with handheld use,heavier instruments, and other physical complications. This impliesthat high reproducibility between handheld and test-stand use inSection 5 was a result of practical “know-how” concerning hand-held operation and, in turn, that the skills needed to acquirehandheld data directly compatible with benchtop measurementsare learned rather than intuitive or instinctive. The implication isthat novices, if possible, should use a test stand, but handheld datacan be reproducible with “in hand” experience.

7. - Case study: Pokr Arteni, Armenia

The Arteni volcanic complex (Fig. 7a) has two eruptive centers:Mets (“Big”) Arteni and Pokr (“Little”) Arteni. The complex wasformed by a series of rhyolitic eruptions, and both centers pro-duced high-quality obsidian and voluminous perlite deposits(Karapetian et al., 2001). Fission-track and KeAr dating of theobsidians yielded consistent ages between 1.2 � 0.1 and 1.4 � 0.2Ma (Komarov et al., 1972;Wagner andWeiner, 1987; Oddone et al.,2000; Chernyshev et al., 2006). Based on prior sourcing studies(e.g., Badalyan et al., 2004, 2007, 2010; Cherry et al., 2010;


Chataigner and Gratuze, 2014b), Arteni was one of the most usedobsidian source complexes in Armenia. In a summary of largelyunpublished work, Badalyan et al. (2004) report that Arteniobsidian constitutes at least half of the assemblage at sites as faras 60 km and that Pokr Arteni obsidian was markedly preferredover Mets Arteni obsidian (a ratio of 30:1 among Neolithicthrough Iron Age artifacts). Therefore, Pokr Arteni is one of themost archaeologically important obsidian sources in the SouthCaucasus, and there is great interpretive potential in elucidatinglinks between its geochemical composition and manifestation onthe landscape.

7.1. - Previous characterization efforts

Keller et al. (1996) concluded that three Pokr Arteni obsidianspecimens constituted a coherent chemical group: Arteni 1A.Blackman et al. (1998), however, analyzed five specimens from theeastern flanks of Pokr Arteni and reported considerable composi-tional variability:

Fig. 5. Correlations between the pXRF measureme

The five Pokr Arteni samples displayed a much wider range ofconcentrations . Three of the five Pokr Arteni source samplesappear to form a single composition group with the remainingtwo samples . at considerable variance from this group andfrom each other. Examination of the Pokr Arteni data show that,although the absolute concentrations vary, their relative pro-portions remain constant, indicating the close relationshipamong the Pokr Arteni samples [but]. the two outlying sourcesamples do not appear to be part of a single variable composi-tional group. (210e211)

Although the Pokr Arteni specimens were collected from asingle locality, the variability amongst these five specimens exceedsthat of any other obsidian source in their dataset. Such variabilitywould seem to necessitate obsidian-bearing eruptions during verydifferent volcanic periods in order to permit sufficient time forconsiderable magmatic evolution to occur within the chamber.Field observations at Pokr Arteni were not consistent with thepresence of such compositionally disparate obsidians.

nts from handheld and test-stand operation.

Table 4Correlations between the data from four groups/instruments and the test-standmeasurements from Section 5. Important instrument parameters and operating procedures arealso summarized here. The “trimmed mean” for Groups 1 reflects the removal of two poorly measured elements (Ti and Nb) for the reasons discussed in Section 6.4.

Group 1 Group 2 Group 3 Group 4

Instrument: Model Niton XL2 Niton 2XL GOLDD Niton XL3t Niton XL3t GOLDDþDetector Si PN SDD Si PN SDDTube (kV) 45 45 50 50Camera? No No Yes YesMass (kg) 1.53 1.53 <1.3 <1.3

Operation: Use mode Handheld Handheld Handheld Test standCorrection FP FP FP FPCalibration 3 SRMs 3 SRMs 3 SRMs 3 SRMsDuration 20 s/filter 20 s/filter 20e30 s/filter 20 s/filterUser change Each measurement Each specimen Each measurement Each specimenOther notes Rear button required e e e

Slope R2 r Slope R2 r Slope R2 r Slope R2 rElements: Ca 0.59 0.78 0.88 0.92 0.78 0.88 1.55 0.90 0.95 1.19 0.98 0.99

Fe 0.74 0.84 0.92 1.12 0.96 0.98 0.98 0.99 0.99 0.99 1.00 1.00Mn 0.78 0.89 0.94 0.94 0.93 0.96 0.90 0.91 0.95 0.84 0.92 0.96Nb 0.23 0.14 0.37 0.14 0.47 0.69 2.80 0.93 0.96 1.34 0.98 0.99Rb 0.76 0.96 0.98 0.92 0.98 0.99 1.23 0.95 0.97 1.02 0.99 0.99Sr 0.77 0.87 0.93 0.81 0.93 0.96 0.95 0.98 0.99 0.98 0.99 0.99Ti 1.19 0.27 0.52 2.64 0.82 0.91 0.93 0.98 0.99 0.96 0.99 0.99Zn 0.80 0.91 0.95 0.90 0.83 0.91 1.22 0.95 0.97 0.88 0.97 0.98Zr 1.15 0.96 0.98 0.96 0.98 0.99 1.31 0.99 0.99 1.12 1.00 1.00Mean: 0.74 0.83 0.85 0.92 0.95 0.98 0.98 0.99Trimmed mean: 0.89 0.94


Recently, Chataigner and Gratuze (2014a) analyzed three PokrArteni specimens with LA-ICP-MS and identified two distinctgroups: Arteni 2 and 3.

The latter two groups [i.e., Arteni 2 and 3] contain obsidian thatoriginates from both the Pokr Arteni and Aragats flow, this isprobably due to an error in sampling, as these flows are difficultto distinguish in the field. Therefore a new systematic samplingand a new set of analyses would be necessary. (36)

It should be noted that they identified variability in Pokr Arteniobsidian much smaller than that reported by Blackman et al. (1998)but attributed it to an error in field sampling. Their sampling area orareas are not elucidated, so it is unknown how these locationscompare to those in earlier studies.

7.2. - Hypotheses regarding previous findings

There are three main possibilities to explain the compositionalvariations of Pokr Arteni obsidian reported by Blackman et al.(1998):

Hypothesis #1. The variation reflects different but related, possiblycomagmatic, eruptions. This explanation is favored by Blackman andcolleagues. Glass Buttes in Oregon is an example of a volcaniccomplex with dual eruptive centers that, between the two of them,reportedly created obsidians with seven (or more) compositionsduring a series of eruptions (Ambroz et al., 2001).

Hypothesis #2. A single eruption produced obsidian with a variablecomposition. Most obsidian suitable for tools is homogeneouswithin the limits of analytical uncertainties (Hughes and Smith,1993), but there are a few exceptions. Bowman et al. (1973a,1973b) discovered that obsidian from Borax Lake in Californiaexhibited a continuous range of compositions, suggestive of themixing of two magmas with distinct compositions in varied pro-portions (see also Shackley,1998). Thus, mixing and fractionation ina magma chamber over time could result in compositional het-erogeneity in Pokr Arteni obsidian (Mahood and Hildreth, 1983;Macdonald et al., 1987; Mahood and Stimac, 1990).

Hypothesis #3. The variability reflects the sampling of an anthro-pogenic context with obsidians from multiple origins. Blackman(1984) analyzed “geological source samples . from a source be-tween the city of [H]razdan and the northwestern tip of LakeSevan” (23). This does not fit any source (Fig. 1). Among the sixspecimens collected there, Blackman (1984) identified twogeochemical groups: five specimens belonged to the “Lake Sevan I”source and one belonged to “Lake Sevan II.” Blackman et al. (1998),after reexamination of these data, found that five of the specimenswere Gutansar obsidian and the sixth was Pokr Arteni obsidian.Pokr Arteni obsidian cannot occur in secondary deposits in theHrazdan basin, so human transport is the only explanation. Indeed,Pokr Arteni obsidian has been reported at sites along the HrazdanRiver (Frahm et al., 2014a). An archaeological, not geological,context was treated as a source in Blackman (1984), and this mightbe responsible for variable “Pokr Arteni” obsidian.

7.3. - Methods and materials

The instrument and methods were the same as Section 3. ANiton pXRF XL3t GOLDDþ, fully corrected and calibrated, was usedto analyze 55 Pokr Arteni specimens, collected from four loci (Map/KML File 1), at least three times each. A portion of the specimenswas also analyzed with EMPA using the procedures documented inFrahm (2012b).

7.4. - Results

Supplementary Table S1 and Fig. 8a summarize the pXRF mea-surements. These data suggest that two similar obsidian composi-tions exist on one continuum, likely due to geochemical evolutionof the magma. A similar trend and the same group membership inthe EMPA data (Fig. 8b) attest this trend is real, not a result ofmeasurement uncertainties. The “gap” between Groups 1 and 2might suggest that specimens of intermediate compositions werenot, or could not be, sampled for this study. Obsidian correspondingto this discontinuity may be deeply buried, altered (e.g., hydrated),or otherwise inaccessible. For example, it is possible that the twoobserved compositions at Pokr Arteni might correspond to a zoned

Fig. 6. Correlations of Rb and Sr data between each group/instrument and the test-stand data.


flow that folded during its emplacement, exposing two apparentlydistinct obsidian compositions at the surface and burying obsidianof intermediate compositions. Alternatively, magmatic dikes ofobsidian in perlitic tuff are apparent at Pokr Arteni (Fig. 7d). It couldbe that zoned or evolved magma filled the dikes, producing

obsidian with slightly different compositions in adjacent ones. Athird possibility is these compositional groups reflect high vari-ability in the abundance of mineral inclusions in Pokr Arteniobsidian. Fig. 7e is a scanning electronmicroscope (SEM) image of aPokr Arteni specimen (AR.2009.68.4) that contains inclusions of

Fig. 7. (a) Simplified geological map of the Arteni volcanic complex and (b and c) the obsidian deposits associated with Pokr Arteni. Digital elevation data from Google Earth inaccordancewith their terms of use; geologicalmap based on Karapetian et al. (2001: Fig. 5) aswell asfield observations. The two compositional groups of Pokr Arteni obsidianmight bedue to (1) zoned or evolvedmagma yielding slightly different compositions within dikes into perlitic tuff (d; highlighted by dotted white lines) or (2) high variability in the abundanceof mineral inclusions, such as zircon [(Zr1ey, REEy)SiO4] in this SEM image (e) of a Pokr Arteni obsidian specimen (AR.2009.68.4). Field photographs and SEM image by the author.


zircon, which has high concentrations of Zr and, frequently, rareearth elements (REEs) such as Hf. Priorities for future field seasonsinclude elucidating the spatial distributions of these two obsidiancompositions on the landscape and the geological reasons for thesegroups.

Using commensurate elements, Fig. 9a plots the pXRF data andLA-ICP-MS mean values for Arteni 2 and 3 from Chataigner andGratuze (2014a). For the two groups, mean values for the pXRFdata closely coincide with the LA-ICP-MSmeans. This indicates that

Fig. 8. (a) pXRF and (b) EMPA data for Pokr Arten

the two Pokr Arteni compositions in Chataigner and Gratuze(2014a) reflect reality. Similarly, Fig. 9b plots the pXRF data withWDXRF data from Keller et al. (1996). For comparable elements, thepXRF Group 1 mean closely coincides with the mean for Arteni A1.This establishes that Keller and colleagues found one Pokr Artenicomposition because their three specimens correspond only toGroup 1 in the pXRF data.

Blackman et al. (1998), though, noted markedly greater vari-ability than Chataigner and Gratuze (2014a). Fig. 10 shows the NAA

i obsidian specimens and their best-fit lines.

Fig. 9. Comparison of the pXRF data for Pokr Arteni to (a) the LA-ICP-MS data from Chataigner and Gratuze (2014a) and (b) the WDXRF from Keller et al. (1996).


measurements from Blackman et al. (1998) with the pXRF data forobsidian sources in Armenia and northeast Turkey. The cluster ofthree specimens in Blackman et al. (1998) agrees fairly well withPokr Arteni. One of the outliers fits Mets Arteni best, while theother falls nearest to two Tsakhkunyats sources. This is consistentwith Fig. 6 in Blackman et al. (1998), a plot of Ta versus Sc: one ofthe “Pokr Arteni” outliers falls in the Mets Arteni cluster, and theother liesbetween twoTsakhkunyats sources.Although the specimennear the Tsakhkunyats sources cannot be conclusively attributed toany of them, it almost certainly did not originate from Pokr Arteni.

The third hypothesis appears correct: the “Pokr Arteni” speci-mens in Blackman et al. (1998) came from an archaeologicalcontext. This hypothesis is supported by fieldwork at a Palaeolithicsite very near the base of Pokr Arteni (Glauberman et al., 2013).Analysis of surface-collected artifacts revealed they had five origins,including one of the Tsakhkunyats sources. Surface deflation in thissemi-arid environment has produced a desert pavement of arti-facts, a palimpsest from innumerable occupations such that, if onesampled preferentially for variability in appearance, one wouldlikely collect obsidians from multiple sources. A similar case haspreviously been debated in this journal (Church 1995; Hoard et al.,1992), establishing the importance of directly sampling thegeological outcrops, a task well suited to pXRF analyses.

Fig. 10. Comparison of the “Pokr Arteni” obsidian from Blackman et al. (1998) to data from obBlackman et al. (1998) appear to have originated from obsidian sources other than Pokr Acollecting specimens from an anthropogenic context.

8. - Conclusions

Evaluated with respect to laboratory-based techniques andvariance among their datasets, pXRF measurements exhibit accu-racy and high reproducibility. The data from handheld and test-stand use can be directly compatible. Even in the “worst case sce-nario” for reproducibility e comparing the data from (1) a heavierinstrument with an earlier detector technology and no built-incamera operated in handheld mode by novice users who mustpress two buttons simultaneously and (2) an SDD- and camera-equipped instrument used in a stand e the datasets are wellcorrelated (r ¼ 0.88e0.98) for many elements, and after intercali-bration based on regression analysis, the measurements would belargely compatible. Instances of elements with poor correlation tolab-based XRF but high correlation to other techniques (e.g., NAA,LA-ICP-MS) support the conclusion of Heginbotham et al. (2010)that the “FP with standards” approach to calibration yields moreaccurate data than traditional empirical schemes.

Based on the results here, there is no evidence to claim thatpXRF is inaccurate, unreproducible, or otherwise inadequate forsource characterization. Nor is there a basis to assert that a prioriknowledge of source compositions is essential before pXRF can beused. The development of ruggedized, handheld instruments is

sidian sources across Armenia and northeastern Turkey. The two “outliers” reported byrteni (Mets Arteni and perhaps one of the Tsakhkunyats sources), most likely due to


simply one of the latest developments in an analytical techniquethat has been utilized in archaeology since the 1950s (e.g., Chengand Schwitter, 1957) and obsidian sourcing since the 1960s (e.g.,Jack and Heizer, 1968). The aforementioned XRF-based NorthAmerican obsidian sourcing laboratories have, since the 1980s,analyzed more than 70,000 artifacts and geological specimens(Skinner, personal communication). It would be surprising if astate-of-the-art pXRF analyzer were less effective for source char-acterization than instruments manufactured in earlier decades.

It is true that NAA and ICP-MS techniques can presentlymeasuremore elements at their typical levels in obsidians (e.g., 28 elementsroutinely quantified in obsidian using NAA in Glascock et al., 2007).There might be instances in some parts of the world where traceelements currently below the minimum detection limits of pXRF(e.g., Eu at the ppm-level or lower) are necessary to discern subtledifferences between certain sources. This, however, is not the casefor Armenia (Frahm et al., 2014a) or neighboring areas such asAnatolia (Forster and Grave, 2012; Frahm, 2013) and the Aegeanregion (Frahm et al., 2014b; Mili�c, 2014). Furthermore, pXRFsensitivity and element range have improved markedly in recentyears, and such advances are expected to continue. Already theXL3t GOLDDþ instrument used in this study is capable of quanti-fying up to 34 elements, and elements such as Zr, Sr, Rb, and Nbhave detection limits circa 3 ppm in a SiO2 matrix with measure-ment times of 60 s per filter.

The case study from Pokr Arteni highlights the advantages ofpXRF, specifically analyzing more specimens than would beotherwise practical, affordable, or legal and studying spatial re-lationships while in the field. Previous studies of Pokr Arteniobsidian resulted in incomplete or erroneous insights into itscomposition. Keller et al. (1996) captured only a small part of thecompositional variability due to small specimen numbers,Blackman et al. (1998) reported spurious variation due to samplingan anthropogenic context, and Chataigner and Gratuze (2014a)attributed real variability to sampling error. Such issues can beaverted via use of a portable instrument that can analyze directlythe outcrops (or nodules in secondary contexts) and stamp eachanalysis with spatial coordinates through a wireless link to a GPSunit. Rather than a case of using pXRF as “an end in itself” (Shackley,2010:17), its use to systematically characterize obsidian sources is aproblem-driven solution to recognizing compositional heteroge-neity and elucidating field relationships. For example, if intra-source variability is indeed more common than widely held andis compositionally recorded in artifacts, identifying the spatialconstraints of heterogeneities might enable differentiation ofquarrying areas. Ultimately, combined field and laboratory use ofpXRF enables better characterization of obsidian sources thanpreviously possible or practical.

Acknowledgments

I am indebted to dozens of individuals whose contributionsmade this study possible and practical. Pavel Avetisyan, Director ofthe Institute for Archaeology and Ethnography, National Academyof Sciences, Armenia is thanked for his on-going support of theprojects in which I am involved. My co-directors on the ObsidianResources and Landscapes of Palaeolithic Armenia project are BorisGasparyan, Institute of Archaeology and Ethnography and Daniel S.Adler, Old World Archaeology Program, University of Connecticut.Fellow members of the Hrazdan Gorge Palaeolithic Project(directed by Adler, Gasparyan, and Benik Yeritsyan, Institute ofArchaeology and Ethnography), particularly Keith Wilkinson, Bev-erly Schmidt-Magee, Philip Glauberman, and Yannick Raczynski-Henk, are thanked for their contributions to our shared researchaims. I collected a portion of the Armenian obsidian specimens in

this study with my collaborators Khachatur Meliksetian and SergeiKarapetian, Institute of Geological Sciences, National Academy ofSciences, Armenia. My sampling and surveying at the Arteni vol-canic complex was assisted by Suren Kesejyan. Additional South-west Asian obsidian specimens related to this study were collectedby George “Rip” Rapp (University of MinnesotaeDuluth), TuncayErcan (Directorate of Mineral Research and Exploration, Turkey),John Whittaker and Kathryn Kamp (Grinnell College), Albert Har-utyunyan (State Engineering University, Armenia), Ruben Badalian(Institute of Archaeology and Ethnography, Armenia), NinoSadradze and Givi Maisuradze (Institute of Geology, Georgia),Robert L. Smith (United States Geological Survey), James F. Luhr andIvan P. Savov (Global Volcanology Program, Smithsonian Institu-tion), and M. James Blackman (Smithsonian Institution). NAA andEDXRF data for a portion of these specimens were provided byMichael Glascock, Archaeometry Laboratory, University of MissouriResearch Reactor. Mexican obsidians also used in inter-techniquecalibration were provided by Magen Coleman, who also collectedcomparative NAA and EDXRF data as part of her Master’s researchat the University of Missouri. The pXRF instrument that I used forthe analyses in Sections 3, 4, 5, and 7 is owned by the ArchaeologyDepartment at the University of Sheffield with funding secured byRoger C.P. Doonan. The specimens used in Sections 5 and 6 weremade by Vassilis Kilikoglou, National Centre for Scientific Research“Demokritos,” Greece. Liev Frahm assisted with their measurementin the test stand. For their analyses in Section 6, the datasets werecollected as a hands-on exercise by participants in the Marie Curietraining course “Introduction to pXRF in Archaeology” held at theUniversity of Sheffield in January 2013 and co-taught by me andDoonan, who indulged my suggestion that we conduct such a testas a component of our course. The course participants are sincerelythanked for their contribution to this study: Mainardo Asinelli,Lydia Avlonitou, Olivier Bonnerot, Andrea Ceglia, Andreas Char-alambous, Elisavet Charalambous, Artemi Chaviara, AnastasiaCholakova, Ioannis Christodoulakis, Maria Dikomitou-Eliadou,William Gilstrap, Keith Haylock, Francesca Licenziati, ChristinaMakarona, Roberta Mentesana, Noémi Müller, Frederik Rade-makers, Jessica Slater, Michael Smith, and Lente Van Brempt. Two ofthe four pXRF instruments that course participants used in Section6 are owned by the University of Sheffield, and the other two wereloaned for the course by Niton UK, thanks especially to John Hurleyand Ken Granger. This research was also supported by theDepartment of Anthropology, Department of Earth Sciences, Elec-tron Microprobe Laboratory, and Institute for Rock Magnetism atthe University of Minnesota. Further support was provided by theMarie Curie Network FP7-PEOPLE-2010-ITN: New ArchaeologicalResearch Network for Integrating Approaches to Ancient MaterialStudies (NARNIA), which is coordinated by Vasiliki Kassianidou,Archaeological Research Unit, University of Cyprus. Doonan wasleader for NARNIA Work Package 7 and also provided insightfulcomments on a manuscript draft that ultimately became Section 6here. The editor and three anonymous reviewers are thanked fortheir helpful comments and critiques that improved the clarity ofthe final manuscript.

Appendix A. Supplementary data

Supplementary data related to this article can be found at http://dx.doi.org/10.1016/j.jas.2014.05.003.

References

Adler, D.S., Yeritsyan, B., Wilkinson, K., Pinhasi, R., Bar-Oz, G., Nahapetyan, S.,Bailey, R., Schmidt, B.A., Glauberman, P., Wales, N., Gasparian, B., 2012. TheHrazdan Gorge Palaeolithic Project, 2008-2009. In: Avetisyan, P., Bobokhyan, A.



http://refhub.elsevier.com/S0305-4403(14)00163-0/sref1




(Eds.), Archaeology of Armenia in Regional Context, Proceedings of the Inter-national Conference Dedicated to the 50th Anniversary of the Institute ofArchaeology and Ethnography Held on September 15–17, 2009 in Yerevan,Armenia. NAS RA Gitutyn Publishing House, Yerevan, pp. 21e37.

Ambroz, J.A., Glascock, M.D., Skinner, C.E., 2001. Chemical differentiation of obsidianwithin the glass buttes complex, Oregon. J. Archaeol. Sci. 28 (7), 741e746.

Badalian, R., Bigazzi, G., Cauvin, M.-C., Chataigner, C., Jrbashyan, R., Karapetyan, S.G.,Oddone, M., Poidevin, J.-L., 2001. An international research project on armenianarchaeological sites: fission-track dating of obsidians. Radiat. Meas. 34 (1e6),373e378.

Badalyan, R., Chataigner, C., Kohl, P., 2004. Trans-Caucasian Obsidian: the Exploi-tation of the Sources and their Distribution. In: Sagona, A. (Ed.), A View from theHighlands: Archaeological Studies in Honour of Charles Burney, pp. 437e465.Ancient Near Eastern Studies.

Badalyan, R., Lombard, P., Avetisyan, P., Chataigner, C., Chabot, J., Vila, E.,Hovsepyan, R., Willcox, G., Pessin, H., 2007. New data on the late prehistory ofthe Southern Caucasus. The excavations at Aratashen (Armenia): preliminaryreport. In: Lyonnet, B. (Ed.), Les Cultures du Caucase (VIeeIIIe millénaires avantnotre ère). Leurs relations avec le Proche-Orient. CNRS éditions, pp. 37e61.

Badalyan, R., Harutyunyan, A., Chataigner, C., Le Mort, F., Chabot, J., Brochier, J.-E.,Balasescu, A., Radu, V., Hovsepyan, R., 2010. The Settlement of Aknashen-Khatunarkh, a neolithic site in the Ararat Plain (Armenia): excavation results2004e2009. Tüba-Ar 13, 187e220.

Bartholomew, D.J., 1996. The Statistical Approach to Social Measurement. AcademicPress, San Diego.

Blackman, M., 1984. Provenance Studies of Middle Eastern Obsidian from Sites inHighland Iran. In: Lambert, J.B. (Ed.), Archaeological Chemistry III, pp. 19e50.

Blackman, M.J., Badalian, R., Kikodze, Z., Kohl, P., 1998. Chemical characterization ofCaucasian Obsidian Geological Sources. In: Cauvin, M.-C., Gourgaud, A.,Gratuze, B., Arnaud, N., Poupeau, G., Poidevin, J.L., Chataigner, C. (Eds.), L’obsi-dienne au Proche et Moyen-Orient: Du Volcan à l’Outil, BAR International Se-ries, pp. 205e231.

Bohrnstedt, G.W., Knoke, D., 1988. Statistics for Social Data Analysis, second ed. F. E.Peacock Publishers, Itasca, IL.

Bowman, H., Asaro, F., Pearlman, I., 1973a. Composition variations in obsidiansources and the archaeological implications. Archaeometry 15 (1), 123e127.

Bowman, H., Asaro, F., Pearlman, I., 1973b. On the uniformity of composition inobsidians and evidence for magmatic mixing. J. Geol. 81 (3), 312e327.

Brandt, S., Johnson, L.M., Chase, A., 2013. A Comparative Study of Multiple BrukerTracer III Calibrated Instruments Using Identical Analytical and OperationalProcedures: Implications for Past, Present, and Future Obsidian Studies. In: 78thAnnual Meeting, Honolulu, Hawaii, 3e7 April 2013. Society for AmericanArchaeology.

Burley, D.V., Sheppard, P.J., Simonin, M., 2011. Tongan and Samoan volcanic glass:pXRF analysis and implications for constructs of ancestral Polynesian society.J. Archaeol. Sci. 38 (10), 2625e2632.

Carmines, E.G., Zeller, R.A., 1979. Reliability and Validity Assessment. QuantitativeApplications in the Social Sciences. Sage Publications.

Cecil, L.G., Moriarity, M.D., Speakman, R.J., Glascock, M.D., 2007. Feasibility of field-portable XRF to identify obsidian sources in central Peten, Guatemala. In:Glascock, M.D., Speakman, R.J., Popelka-Filcoff, R.S. (Eds.), ArchaeologicalChemistry: Analytical Techniques and Archaeological Interpretation, ACSSymposium Series No. 968. American Chemical Society, Washington, DC,pp. 506e521.

Chataigner, C., Badalian, R., Bigazzi, G., Cauvin, M.-C., 2003. Provenance studies ofobsidian artefacts from armenian archaeological sites using the fission-trackdating method. J. Non-Crystalline Solids 323, 172e179.

Chataigner, C., Gratuze, B., 2014a. New data on the exploitation of obsidian in thesouthern caucasus (Armenia, Georgia) and eastern Turkey, Part 1: sourcecharacterization. Archaeometry 56, 25e47.

Chataigner, C., Gratuze, B., 2014b. New data on the exploitation of obsidian in thesouthern caucasus (Armenia, Georgia) and eastern Turkey, Part 2: obsidianprocurement from the Upper palaeolithic to the late Bronze age. Archaeometry56, 48e69.

Cheng, C.F., Schwitter, C.M., 1957. Nickel in ancient Bronzes. Am. J. Archaeol. 61,351e365.

Chernyshev, I.V., Lebedev, V.A., Arakelyants, M.M., 2006. KeAr dating of Quaternaryvolcanics: methodology and interpretation of results. Petrology 14, 62e80.

Cherry, J.F., Faro, E.Z., Minc, L., 2010. Field survey and geochemical characterizationof the southern armenian obsidian sources. J. Field Archaeol. 35 (2), 147e163.

Church, T., 1995. Comment on “neutron activation analysis of stone from the Cha-dron formation and a Clovis site on the Great Plains” by Hoard et al. (1992).J. Archaeol. Sci. 22, 1e5.

Clark, B.J., Mayo, K., 2012. Freshmen Sourcing Obsidian? Using PXRF in the Intro-ductory Archaeology Classroom. In: Society for American Archaeology, 77thAnnual Meeting. Memphis, Tennessee, 18e22 April 2012.

Conrey, R.M., Goodman-Elgar, M., Bettencourt, N., Seyfarth, A., Van Hoose, A.,Wolff, J.A., 2014. Calibration of a portable X-ray fluorescence spectrometer inthe analysis of archaeological samples using influence coefficients. Geochem.Explor. Environ. Anal. http://dx.doi.org/10.1144/geochem2013-198.

Craig, N., Speakman, R.J., Popelka-Filcoff, R., Glascock, M.D., Robertson, J.,Shackley, M.S., Aldenderfer, M., 2007. Comparison of XRF and PXRF for analysisof archaeological obsidian from southern Perú. J. Archaeol. Sci. 34, 2012e2024.

Craig, N., Speakman, R.J., Popelka-Filcoff, R., Aldenderfer, M., Blanco, L., Vega, M.,Glascock, M.D., Stanish, C., 2010. Macusani obsidian from southern Peru: a

characterization of its elemental composition with a demonstration of itsancient use. J. Archaeol. Sci. 37, 569e576.

Dybowski, D., 2012. pXRF/WDXRFS inter-unit data comparison of Arizona obsidiansamples. J. Ariz. Archaeol. 2, 1e7.

Forster, N., Grave, P., 2012. Non-destructive PXRF analysis of museum-curatedobsidian from the Near East. J. Archaeol. Sci. 39, 728e736.

Frahm, E., 2007. An Evaluation of Portable X-Ray Fluorescence for Artifact Sourcingin the Field: Can a Handheld Device Differentiate Anatolian Obsidian Sources?Session: Sourcing Techniques in Archaeology. In: Geological Society of AmericaAnnual Meeting, Denver, Colorado, 28e31 October.

Frahm, E., 2010. The Bronze-age obsidian industry at Tell Mozan (ancient Urkesh),Syria. Ph.D. dissertation. Department of Anthropology, University of Minnesota.Published online through the University of Minnesota’s Digital Conservancy:http://purl.umn.edu/99753.

Frahm, E., 2012a. Evaluation of archaeological sourcing techniques: reconsidering andre-derivinghughes’ four-foldassessmentscheme.Geoarchaeology27 (2),166e174.

Frahm, E., 2012b. Non-Destructive sourcing of Bronze-age near eastern obsidianartefacts: redeveloping and reassessing electron microprobe analysis forobsidian sourcing. Archaeometry 54 (4), 623e642.

Frahm, E., 2013. Validity of “off-the-shelf” handheld portable XRF for sourcing NearEastern obsidian chip debris. J. Archaeol. Sci. 40, 1080e1092.

Frahm, E., Doonan, R., Kilikoglou, V., 2014b. Handheld portable X-ray fluorescenceof aegean obsidians. Archaeometry 56, 228e260.

Frahm, E., Schmidt, B., Gasparyan, B., Yeritsyan, B., Karapetian, S.,Meliksetian, Kh, Adler, D.S., 2014a. Ten seconds in the field: rapid armenianobsidian sourcing with portable XRF to inform excavations and surveys.J. Archaeol. Sci. 41, 333e348.

Gauthier, G., Burke, A.L., 2011. The effects of surface weathering on the geochemicalanalysis of archaeological lithic samples using non-destructive polarized energydispersive XRF. Geoarchaeology 26 (2), 269e291.

Glascock, M.D., 1999. An inter-laboratory comparison of element compositions fortwo obsidian sources. Int. Assoc. Obsidian Stud. Bull. 23, 13e25.

Glascock, M.D., 2011. Comparison and contrast between XRF and NAA: used forcharacterization of Obsidian sources in Central Mexico. In: Shackley, M.S.(Ed.), X-Ray Fluorescence Spectrometry (XRF) in Geoarchaeology. Springer,pp. 161e192.

Glascock, M.D., Ferguson, J.R., 2012. Report on the Analysis of Obsidian SourceSamples byMultiple Analytical Methods. Archaeometry Laboratory, Universityof Missouri. January 2012.

Glascock, M.D., Braswell, G., Cobean, R.H., 1998. A Systematic approach to Obsidiansource characterization. In: Shackley, M.S. (Ed.), Archaeological ObsidianStudies: Method and Theory. Society for Archaeological Sciences, pp. 15e65.

Glascock, M.D., Speakman, R.J., Neff, H., 2007. Archaeometry at the University ofMissouri research reactor and the provenance of obsidian Artefacts in NorthAmerica. Archaeometry 49 (2), 343e357.

Glauberman, P.J., Gasparyan, B., Wilkinson, K., Frahm, E., Raczynski-Henk, Y.,Adler, D.S., 2013. Barozh 12: A New Middle Palaeolithic Open-Air Site on theEdge of the Ararat Plain, Armenia. In: The Role of the Southern Caucasus onEarly Human Evolution and Expansion: Refuge, Hub or Source Area? Tbilisi,Georgia, 15e20 October 2013.

Goldstein, J.I., Newberry, D.E., Echlin, P., Joy, D.C., Fiori, C., Lifshin, E., 1981. ScanningElectron Microscopy and X-ray Microanalysis: a Text for Biologists, MaterialsScientists, and Geologists. Plenum Press, New York.

Goodale, N., Bailey, D., Jones, G., Prescott, C., Scholz, E., Stagliano, N., Lewis, C., 2012.pXRF: a study of inter-instrument performance. J. Archaeol. Sci. 39, 875e883.

Hancock, R.G.V., Carter, T., 2010. How reliable are our published archaeometricanalyses? Effects of analytical techniques through time on the elementalanalysis of obsidians. J. Archaeol. Sci. 37 (2), 243e250.

Heginbotham, A., Bezur, A., Bouchard, M., Davis, J.M., Eremin, K., Frantz, J.H.,Glinsman, L., Hayek, L.-A., Hook, D., Kantarelou, V., Germanos Karydas, A.,Lee, L., Mass, J., Matsen, C., McCarthy, B., McGath, M., Shugar, A., Sirois, J.,Smith, D., Speakman, R.J., 2010. An evaluation of inter-laboratory reproduc-ibility for quantitative XRF of historic copper alloys. In: Mardikian, P., et al.(Eds.), Metal 2010. Proceedings of the International Conference on Metal Con-servation, Charleston, South Carolina, USA, October 11e15, 2010. ClemsonUniversity Press, pp. 244e255.

Hoard, R.J., Holen, S.R., Glascock, M.D., Neff, H., Elam, J.M., 1992. Neutron activationanalysis of stone from the Chadron Formation and a Clovis site on the GreatPlains. J. Archaeol. Sci. 19, 655e665.

Hughes, R.E., 1998. On reliability, validity, and scale in obsidian sourcing research.In: Ramenofsky, A.F., Steffen, A. (Eds.), Unit Issues in Archaeology: MeasuringTime, Space, and Material. University of Utah Press, pp. 103e114.

Hughes, R.E., Smith, R.L., 1993. Archaeology, Geology, and Geochemistry of ObsidianProvenance Studies. In: Stein, J.K., Linse, A.R. (Eds.), Effects of Scale onArchaeological and Geological Perspectives, GSA Special Paper, volume 283.Geological Society of America, Boulder, Colorado, pp. 79e91.

Jack, R.N., Heizer, R.F., 1968. “Finger-printing” of some Mesoamerican obsidian ar-tifacts. In: Contributions of the University of California Archaeological ResearchFacility, vol. 5, pp. 81e100.

Jia, P., Doelman, T., Chen, C., Zhao, H., Lin, S., Torrence, R., Glascock, M.D., 2010.Moving sources: a preliminary study of volcanic glass artifact distributions innortheast China using PXRF. J. Archaeol. Sci. 37, 1670e1677.

Jia, P., Doelman, T., Torrence, R., Glascock, M.D., 2013. New pieces: the acquisitionand distribution of volcanic glass sources in northeast China during the Holo-cene. J. Archaeol. Sci. 40, 971e982.






































































































http://dx.doi.org/10.1144/geochem2013-198





















http://purl.umn.edu/99753






























































































Kandel, A.W., Gasparyan, B., Frahm, E., Taller, A., Weissbrod, L., 2013. Latest Resultsfrom Aghitu-3, an Early Upper Paleolithic Cave Site in Armenia. In: HugoObermaier Society for Quaternary Research and Archaeology of the Stone AgeMeeting, Vienna, 2e6 April.

Kaplan, A., 1964. The conduct of inquiry: methodology for behavioral science.Chandler Publishing Company, San Francisco.

Karapetian Jr., S.G., bashian, R., Mnatsakanian, A.Kh, 2001. Late collision rhyoliticvolcanism in the north-eastern part of the Armenian Highland. J. Volcanol.Geotherm. Res. 112, 189e220.

Keller, J., Djerbashian, E., Pernicka, E., Karapetian, S., Nasedkin, V., 1996. Armenianand Caucasian obsidian Occurrences as sources for the neolithic trade: volca-nological Setting and chemical characteristics. In: Archaeometry 94: the Pro-ceedings of the 29th International Symposium on Archaeometry, Ankara, 9e14May 1994, pp. 69e86.

Keller, J., Seifried, C., 1990. The Present Status of obsidian source Identification inanatolia and the near east. PACT 25: volcanology and archaeology. In: Pro-ceedings of the European Workshops of Ravello, vol. 25, pp. 57e87.

Kellett, L.C., Golitko, M., Bauer, B.S., 2013. A provenance study of archaeologicalobsidian from the Andahuaylas region of southern Peru. J. Archaeol. Sci. 40,1890e1902.

Kilikoglou, V., Vekinis, G., Maniatis, Y., Day, P.M., 1998. Mechanical performance ofquartz-tempered ceramics: part I, strength and toughness. Archaeometry 40,261e279.

Komarov, A.N., Skovorodkin, N.V., Karapetian, S.G., 1972. Determination of the age ofnatural glasses according to tracks of uranium fission fragments. Geochimia N6,693e698 (in Russian).

Lemonnier, P., 1992. Elements for an Anthropology of Technology. In: Anthropo-logical Papers Museum of Anthropology, vol. 88. University of Michigan.

Leroi-Gourhan, A., 1993. Gesture and Speech. Translated from Le Geste et la parole[1964] by A.B. Berger. Massachusetts Institute of Technology Press.

Liritzis, I., 2008. Assessment of Aegean obsidian sources by a portable ED-XRFanalyser:grouping, provenance and accuracy. In: Facorellis, Y., Zacharias, N., Polikreti, K.,Vakoulis, T., Bassiakos, Y., Kiriatzi, V., Aloupi, E. (Eds.), Archaeometry Studies in theAegean: Reviews and Recent Developments. Archaeopress, Oxford, pp. 399e406.

Liritzis, I., Zacharias, N., 2011. Chapter 6: portable XRF of archaeological artifacts:current research, potentials and limitations. In: Shackley, M.S. (Ed.), X-rayFluorescence Spectrometry (XRF) in Geoarchaeology. Springer, pp. 109e142.

Macdonald, R., Davies, G.R., Bliss, C.M., Leat, P.T., Bailey, D.K., Smith, R.L., 1987.Geochemistry of high-silica peralkaline rhyolites, Naivasha, Kenya rift valley.J. Petrol. 28, 979e1008.

Mahood, G.A., Hildreth, E.W., 1983. Large partition coefficients for trace elements inhigh-silica rhyolites. Geochim. Cosmochim. Acta 47, 11e30.

Mahood, G.A., Stimac, J.A., 1990. Trace-element partitioning in pantellerites andtrachytes. Geochim. Cosmochim. Acta 54, 2257e2276.

Mark, H., Workman, J., 2003. Statistics in Spectroscopy, second ed. Elsevier Aca-demic Press.

Mauss, M., 1973. Techniques of the Body. Econ. Soc. 2, 70e88.McCoy, M., Mills, P., Lundblad, S., Rieth, T., Kahn, J., Gard, R., 2011. A cost surface

model of volcanic glass quarrying and exchange in Hawai’i. J. Archaeol. Sci. 38,2547e2560.

Millhauser, J., Rodríguez-Alegría, E., Glascock, M.D., 2011. Testing the accuracy ofportable X-ray fluorescence to study Aztec and Colonial obsidian supply atXaltocan, Mexico. J. Archaeol. Sci. 38, 3141e3152.

Mili�c, M., 2014. PXRF characterisation of obsidian from central Anatolia, the Aegeanand central Europe. J. Archaeol. Sci. 41, 285e296.

Nazaroff, A., Prufer, K., Drake, B., 2010. Assessing the applicability of portable X-rayfluorescence spectrometry for obsidian provenance research in the Maya low-lands. J. Archaeol. Sci. 37, 885e895.

Oddone, M., Bigazzi, G., Keheyan, Y., Meloni, S., 2000. Characterisation of armenianObsidians: implications for raw material supply for Prehistoric artifacts.J. Radioanal. Nucl. Chem. 243 (3), 673e682.

Reed, S.J.B., 1996. Electron Microprobe Analysis and Scanning Electron Microscopyin Geology. Cambridge University Press.

Rodríguez-Alegría, E., Millhauser, J.K., Stoner, W.D., 2013. Trade, tribute, andneutron activation: the colonial political economy of Xaltocan, Mexico.J. Anthropol. Archaeol. 32, 397e414.

Scharlotta, I., 2010. Groundmass microsampling using laser ablation time-of-flightinductively coupled plasma mass spectrometry (LA-TOF-ICP-MS): potential forrhyolite provenance research. J. Archaeol. Sci. 37, 1929e1941.

Shackley, M.S., 1998. Intrasource chemical variability and secondary depositionalprocesses: lessons from the american southwest. In archaeological obsidianstudies: Method and Theory. In: Shackley, M.S. (Ed.), Advances in Archaeolog-ical and Museum Science 3. Plenum Press, pp. 82e102.

Shackley, M.S., 2005. Obsidian: Geology and Archaeology in the North AmericanSouthwest. University of Arizona.

Shackley, M.S., 2010. Is there reliability and validity in portable X-ray fluorescencespectrometry (PXRF)? The SAA Archaeological Record. Novemb. issue, 17e20.

Shackley, M.S., 2011. An introduction to X-Ray Fluorescence (XRF) analysis inarchaeology. In: Shackley, M.S. (Ed.), X-Ray Fluorescence Spectrometry (XRF) inGeoarchaeology. Springer, pp. 7e44.

Shackley, M.S., 2012. Portable X-ray Fluorescence Spectrometry (pXRF): the Good,the Bad, and the Ugly (online essay). Archaeol. Southwest Mag. 26 (Spring issue).

Sheppard, P., Irwin, G., Lin, S., McCaffrey, C., 2011. Characterization of New Zealandobsidian using PXRF. J. Archaeol. Sci. 38, 45e56.

Speakman, R.J., 2012. Evaluation of Bruker’s Tracer Family Factory Obsidian Cali-bration for Handheld Portable XRF Studies of Obsidian. Report Prepared forBruker AXS, Kennewick, WA.

Speakman, R.J., Shackley, M.S., 2013. Commentary: Silo science and portable XRF inarchaeology: a Response to Frahm. J. Archaeol. Sci. 40, 1435e1443.

Speakman, R.J., Little, N.C., Creel, D., Miller, M.R., Iñañe, J.G., 2011. Sourcing ceramicswith portable XRF spectrometers? A comparison with INAA using Mimbrespottery from the American Southwest. J. Archaeol. Sci. 38, 3483e3496.

Taylor, J.R., 1996. Introduction to Error Analysis: the Study of Uncertainties inPhysical Measurements, second ed. University Science Books.

Taylor, B.N., Kuyatt, C.E., 1994. Guidelines for Evaluating and Expressing the Un-certainty of NIST Measurement Results. NIST Technical Note 1297.

Tykot, R.H., Lai, L., Tozzi, C., 2011. Intra-site Obsidian subsource patterns at Con-traguda, Sardinia (Italy). In: Turbanti-Memmi, I. (Ed.), Proceedings of the 37thInternational Symposium on Archaeometry. Springer, pp. 321e328.

Tykot, R.H., Freund, K.P., Vianello, A., 2013a. Source analysis of Prehistoric ObsidianArtifacts in Sicily (Italy) using pXRF. In: Armitage, R.A., Burton, J.H. (Eds.),Archaeological Chemistry VIII, pp. 195e210.

Tykot, R.H., White, N.M., Du Vernay, J.P., Freeman, J.S., Hays, C.T., Koppe, M.,Hunt, C.N., Weinstein, R.A., Woodward, D.S., 2013b. Advantages and disadvan-tages of PXRF for archaeological ceramic analysis: prehistoric pottery distri-bution and trade in NW Florida. In: Armitage, R.A., Burton, J.H. (Eds.),Archaeological Chemistry VIII, ACS Symposium Series, vol. 1147. AmericanChemical Society, pp. 233e244.

Wagner, G.A., Weiner, K.L., 1987. Deutches Archaeologisches Institut Demircihoyuk.Die Ergebenisse des Ausgrabungen 1975e1978. In: Manfred Korfmann (Ed.),Naturwissenschaftliche Untersuchungen, Band II. Verlag Philipp von Zabern,Mainz, Germany, pp. 26e29.

Williams, P.R., Dussubieux, L., Nash, D., 2012. Provenance of Peruvian Wari obsidian:comparing INAA, LAeICPeMS, and portable XRF. In: Liritzis, I., Stevenson, C.(Eds.), Obsidian and Ancient Manufactured Glasses. University of New MexicoPress, pp. 75e85.

























































































































































Characterizing obsidian sources with portable XRF ...members.peak.org/~obsidian/pdf/frahm_2014_2.pdf · Characterizing obsidian sources with portable XRF: accuracy, reproducibility,

Documents