Professional Documents
Culture Documents
DOI 10.1007/s10816-010-9087-7
Abstract The ways and extent to which sampling design influences data collection
and archaeological inference is a constant concern for archaeologists. Yet, spatial
analyses based on anthrosol chemistry have been less willing to concede this
problem and to explore potential solutions. This article reviews the recent literature
on soil sampling for spatial studies and then uses an example from prehispanic
Honduras to examine how both quantitative and qualitative interpretations of soil
chemical patterns can shift when sampling design changes. The results of this study
suggest that the principal challenges to selecting an appropriate sampling design are
in determining the sample size and density, as well as recognizing and adequately
dealing with variation in the soil properties being measured. These findings provide
cautionary tales for spatial studies aimed at using soil chemical data to infer activity
patterns in the archaeological record.
Keywords Soil sampling . Anthrosol chemistry . Spatial analysis . Activity patterns .
Honduras
Introduction
The only material more abundant at an archaeological site than pots and rocks is soil.
It might seem reasonable then to assume that, since archaeologists have developed a
diverse array of sophisticated strategies for sampling where to excavate and which
artifacts to collect, there are probably an equal or greater number of strategies that
inform us how to sample soils and sediments. Unfortunately, nothing could be
further from the truth. Instead, archaeologists investigating soils mostly borrow
sampling strategies that have been developed for survey and excavation (e.g., Orton
E. C. Wells (*)
Department of Anthropology, University of South Florida, 4202 E. Fowler Ave.,
Tampa, FL 33620, USA
e-mail: ecwells@usf.edu
210
Wells
2000). However, sampling designs for site survey and excavation are analytical
strategies that are rather unlike those required for studies of anthrosols (most
generally, soils and earthen surfaces that have been physically or chemically altered
by human activity; see Holliday 2004: 2627), because of the ways and extent to
which anthrosol properties, such as texture and pH, vary over space and time.
Relying, as we have, on strategies developed to detect physical (discrete) traces of
past human activity has resulted in a mixed bag of sampling designs that are variably
productive for understanding how chemical (continuous) traces of human activities
vary across soilscapes. Although some studies have acknowledged interpretive
challenges coincident with variation in the sample matrix (e.g., Kintigh 1988;
Krakker et al. 1983; Redman 1987), the magnitude and significance of the problem
is not well understood, especially in research on anthrosols and prepared surfaces.
Little consideration has been given to exploring the problem, and even less attention
has been paid to resolving practical ways in which it might be addressed. In this
article, I advocate a more critical approach to sampling anthrosols and prepared
surfaces for spatial analysis in which consideration is given to potential interpretive
bias inherent in different sampling plans.
Given the broad reach of soil analysis in spatial studies (see Wells and Terry
2007), I restrict my consideration in this article to small scales, such as those
concerning the archaeology of households and communities. For studies at larger
scales encompassing landscapes, such as archaeological site prospection, sampling
concerns regarding phosphates and heavy metals have been addressed by Crowther
(1997) and Haslam and Tibbett (2004), respectively. I begin with a brief review of
the recent literature on extant sampling schemes, both systematic and unsystematic,
to highlight some of the patterns in the use of design structures, as well as their
deviations. I then draw on an example of my work in Honduras to explore some of
the ways in which sample design (and coincidentally, sample density) can cause
spurious patterns in anthrosol chemical data that can bias our inferences about the
location and nature of past human activities in the archaeological record. I conclude
with a few recommendations for selecting and employing sampling designs for
small-scale archaeological research on anthrosols and prepared surfaces.
Sampling Designs
In the original methods textbook, A Manual of Archaeological Field Methods, first
published in 1949 (and revised several times since), Heizer (1949: 78) encouraged
archaeologists to collect soil samplesfrom every site excavated, though he failed
to mention how or why, or even what constitutes soil. Almost 50 years later, the
seventh edition of this text was published by Hester et al. (1997). The new version
does a much better job of explaining the value of analyzing soils to reconstruct
human activity patterns (Hester et al. 1997: 136), but still does not recommend how
one might go about sampling soils to do so. Specialized volumes, such as Holliday's
(2004) Soils in Archaeological Research and Goldberg and Macphail's (2006)
Practical and Theoretical Geoarchaeology, at least pay some attention to the subject.
Holliday (2004: 3536) provides sage advice along with some general guidelines
about how one's sampling design and sample size ought to vary according to the
211
research question. Goldberg and Macphail (2006: 328333) point out some of the
specific archaeological contexts in which it is more appropriate to collect bulk versus
point samples. Suffice it to say, when collecting soils for chemical studies aimed at
reconstructing activity loci, it is absolutely necessary to collect point samples. In
other words, compositing samples from multiple contexts, even if only a few
centimeters apart, may result in mixed activity residues, which would confound the
purpose of the analysis.
Despite recent advancements and clarifications in soil sampling, archaeologists
interested in obtaining and making sense of soil chemical data have had to rely on
sampling strategies developed for archaeological survey and excavation. In his
classic paper on research design, Binford (1964) proposed sampling as a means of
obtaining an accurate representation of the range of variation within a region. He
even recognized soil as a sample universe (Binford 1964: 432), although he lumped
it into the broader category of ecofacts. Since his paper, archaeologists have come
to understand sample design as a formal procedure that determines how many units
of analysis are to be selected and how this is to be done; thus, the outcome of one
procedure is a sample (in the statistical sense, where a sample is a subset of its
population) composed of units for analysis.
The essential requirement of a sample, in the way Binford intended, is that it must
be representative of the population (for anthrosol research, this could be either the
soil environment or the human activity area) from which it was drawn, or at least we
must have evidential reason to believe that it is representative. The key assumption
here is that the variances of the values of the characteristics being measured are
equal across all sample units (the so-called homogeneity of variance assumption in
statistics). While we can never or rarely be certain of just how representative a soil
sample is of its population, we can seek to reduce the risk that the sample is not
representative. Binford proposed probabilistic sampling to address this issue,
wherein each unit has a theoretically equal chance of being selected. Such random
samples, then, can be determined to have a certain probability of being
representative of the variation in its associated population. This is an entirely
reasonable approach for sites, features, and artifacts, but a critical problem arises for
soil studies: anthrosol properties between samples do not, in fact, vary in the same
way or to the same extent (Fisher et al. 1998; Zhang et al. 2007). In other words,
variation is often discontinuous, periodic, or stochastic, but rarely homoscedastic. As
such, soil surveys in archaeology have had to rely on systematic sampling strategies,
often referred to as point lattice matrixes or grid (or mesh) sampling (e.g., Entwistle
et al. 2000; Marshall 2001). In this article I use the term lattice (i.e., regularly
spaced array of points in Euclidean space) to refer to these sample designs. Below, I
briefly discuss three of the most common lattices in archaeological sampling
square, staggered, and hexagonal (for illustrative examples, see Kintigh 1988:
687)as I have used them in Central America.
Square Lattice
The square lattice is the most traditional approach to systematic sampling for spatial
studies employing anthrosol chemical data. This approach involves the selection of
units from regular (equal) intervals on evenly spaced transects throughout the sample
212
Wells
frame, or universe, with only the first unit being randomly placed. This design has
the advantage of distributing the sample frame more evenly across the population
than would be obtained by simple random or judgmental sampling. At the Late
Classic (ca. AD 650850) civic-ceremonial center of Palmarejo in northwest
Honduras (Fig. 1), my colleagues and I used a 2-m square lattice to collect 188
units to study activity areas from the site's main plaza, which measures roughly 30
30 m (Wells et al. 2007). Due to patterns of modern vegetation cover, we were
unable to select units from every point on the lattice. Thus, our sampling fell short of
the roughly 225 units we might have been able to collect had there been no
obstruction by vegetation. Samples were collected from an earthen surface (onto
which plaster had been added in some areas), roughly 0.15 m below the modern
ground surface. We chose the square lattice in this case because of the shape of the
space to be sampled. Our sample intensity of 2-m intervals was selected based on
previous experience working in modern plaza spaces in this region, which showed
that these areas are used for a wide range of small-scale activities (Wells and
Urban 2002). We therefore decided that we would need a relatively high sample
density to capture evidence for localized activities. Finally, we used the square lattice
N
magnetic
0 5
20
METERS
Fig. 1 Palmarejo, northwestern Honduras, showing the square lattice on the plaza (upper left) and the
rectangular lattice on the patio (lower right)
213
because, again based on previous work, we expected that activities would not be
systematically distributed throughout the space. Therefore, systematic unit intervals
would probably not systematically miss activity areas.
A variation on the square lattice is the rectangular lattice, where the distance
between rows is not the same as the sample-unit spacing within a row (Kintigh 1988:
687). At Palmarejo (see Fig. 1), we used a rectangular lattice with 5-m intervals in a
large elite patio composed of compacted soil (i.e., a prepared earthen surface),
collecting 136 units. We chose the rectangular lattice because of the shape of the
space, which is roughly 70 m east-to-west50 m north-to-south. Since we needed to
sample such a large space, and our budget would not allow us to process the
approximately 875 units that we might have collected from a grid with 2-m intervals,
we chose 5-m intervals. Again, as with the plaza, modern vegetation cover prevented
us from covering the entire space.
Since we considered these anthrosols to be ideal for chemical research (they are
silty clays with a median pH of 7.0), we processed them with a mild acid extraction
procedure developed by Lewis et al. (1993), which combines dilute hydrochloric
and nitric acids. This is one of the strongest of the weak acid extraction procedures,
which we felt was desirable in this case due to the deep and variable substrates and
our need to dissolve powdery (recently formed) carbonates, poorly crystallized
oxides and hydroxyoxides, and native metals. We analyzed the extracts for a range
of chemical elements using inductively coupled plasma-atomic emission spectroscopy (ICP-AES). Although sample density varied between the plaza and the patio,
and the substrates were not the same (the plaster-enriched plaza vs the earthen patio),
we were able to compare the locations of activity areas within each space through
geostatistical interpolation of the chemical data using spherical semivariogram
models and kriging (Wells et al. 2007). We found that the patio showed little
evidence for activities, while the plaza appears to have been used extensively.
Staggered Lattice
For a staggered lattice, the spacing of sample units within a row is equal to the
distance between rows, but with units in adjacent rows offset one half of this interval
(Krakker et al. 1983: 427). The units can be systematically offset or they can be
arbitrary, in which the distance between rows does not equal the spacing between
sample units along the row (Kintigh 1988: 687). At the Classic (ca. AD 250850)
Maya city of Piedras Negras, Guatemala (Fig. 2), my colleagues and I used a
combination of a 2-m staggered lattice, 2-m square lattice, and opportunistic
sampling to investigate activity patterning on a compact, earthen surface in an elite
residential courtyard referred to as the U-sector of the South Group (Wells et al.
2000; see also Parnell et al. 2002). The staggered lattice followed a rectangular
sample frame on a terrace alongside the exterior of two buildings, the square lattice
was used for the evenly sized house floors, and opportunistic samples were taken
from specific features (mainly suspected middens). These strategies provided us with
30 units for analysis. Other areas of the site were sampled using 5-m square lattices
(Parnell et al. 2001). We used a combination of sampling designs in this case
because of the complexity of the layout of the buildings, which resisted a one-sizefits-all sampling approach.
214
Wells
N
magnetic
0 1
METERS
Fig. 2 Piedras Negras, western Guatemala, showing the staggered lattice (supplemented by opportunistic
sampling) on the terrace (lower left) and the square lattice on the patio (upper right)
We processed the anthrosols using the Mehlich II dilute acid extraction technique
(Mehlich 1978) for phosphate analysis and analyzed them with a portable
colorimeter (according to the procedures outlined in Terry et al. 2000). We followed
this approach because we wanted the results with us in the field to help us make
decisions about where to place test units for excavation. In the laboratory, we also
used diethylenetriaminepentaacetic acid to extract heavy metals (Lindsay and
Norvell 1978), which were analyzed with ICP-AES. Mixing sampling designs and
interval distances made the resulting data difficult to analyze spatially. We used a
linear variogram model and kriging to interpolate chemical concentrations between
our sampling units. The result was a map of concentrations depicted as isopleth
lines. Since our sample lattices and intervals were different for each part of the
space, the larger distances between sample units have much higher variances
associated with them and, thus, are less reliable interpolations. Still, we found
possible evidence for a range of activities around the buildings, including evidence
for food preparation and consumption and a variety of craft activities (Wells et al.
2000: 454459).
Hexagonal Lattice
A hexagonal lattice is formed by the placement of sample units at the vertices of
equilateral triangles (Kintigh 1988: 687; Krakker et al. 1983: 427), such that
sampling units are equidistant from each of their six nearest neighbors. At El
Coyote, a large, Late Classic (ca. AD 650850) civic-ceremonial center in northwest
Honduras (Fig. 3) roughly 20 km southwest of Palmarejo, I collected 530 units from
215
N
magnetic
05
20
METERS
Fig. 3 El Coyote, northwestern Honduras, showing the hexagonal lattice (supplemented by a square
lattice and opportunistic sampling) on the plaza and patio
a 5-m hexagonal lattice placed over the lime-plastered surface of the site's main
plaza and adjacent residential patio spaces (Wells 2004). The plaza measures roughly
100 m north-to-south 50 m east-to-west, while the patio spaces are much smaller,
encompassing 2020 m. While the study area is composed of differently sized
spaces, I used a single lattice system (supplemented in certain areas with a square
lattice and opportunistic sampling) because I wanted specifically to be able to
compare the activities in all spaces. The work at Piedras Negras taught me that
comparison would not be possible if I used different lattice designs and sample
intervals. I decided to use a hexagonal lattice over a square or staggered lattice
because I had such a large space to cover (a total area of about 15,000 m2) and a
relatively meager budget for chemical analysis. Using the hexagonal lattice, I could
sample the space with about half as many samples as I could have using a square
216
Wells
lattice, since the mean distance to the nearest lattice point is reduced compared with
a square lattice of the same density.
To process the samples, I used the Burton and Simon (1993) extraction method
that calls for dilute hydrochloric acid (1 M), again because of my small budget (this
is a very inexpensive technique that works well with calcareous soil; e.g., Middleton
and Price 1996). Samples were characterized with ICP-AES, and the resulting data
were analyzed with principal components analysis and discriminant function
analysis. The results allowed me to differentiate three groups of anthrosols: those
high in phosphorus (P) and low in potassium (K) in the plaza, those high in K and
low in P in the patios, and more complex soils representing a range of activities that
took place outside of formally defined spaces (Wells 2004: 7880). The data were
also analyzed spatially using an exponential variogram model and kriging, which
allowed me to produce image maps of the interpolated chemical concentrations
(Wells 2004: 74, 76).
Inferential Bias
It has long been recognized in archaeology that different sampling designs can lead
to inferential bias. In other words, the selection of how to sample and how much to
sample impacts the data we collect and, thus, the inferences we make about the
archaeological record. This problem has been addressed to varying degrees for
shovel-test sampling (Champion et al. 1996; Kintigh 1988; Krakker et al. 1983;
Lightfoot 1986; Nance and Ball 1986; Shott 1985, 1987, 1989; Wobst 1983) and
auger sampling (Casteel 1970; Howell 1993; Price et al. 1964; Reed et al. 1968;
Stein 1986)approaches to sub-surface prospection that are relevant to anthrosol
chemical surveys. These papers address a wide range of subjects in sampling theory
regarding how to model the probability of finding archaeological sites or features
but, on the whole, focus on issues of reliability and validity, both in a statistical
sense. Rather than retread this intellectual ground, here I wish to add to these
insights by discussing how our inferences about activity patterns can be biased
through different kinds and scales of parameter variation in soils. Variation in soil
properties is quite unlike variation in site or artifact distribution and so needs to be
considered separately. I argue that knowing the sources of variation in the soil
properties being measured is critically important for selecting an appropriate
sampling design and density.
In a recent article, Haslam and Tibbett (2004) discuss sampling strategies with
regard to soil chemistry, specifically variation in heavy metals across space. They
argue that conventional (probabilistic and semi-probabilistic) quantitative modeling
of soil chemical data is insufficient for understanding chemical variability in soils
because of the underlying assumptions about the independence of measured
observations in spatial distribution. They rightly point out that soils closer together
tend to have similar properties (and, thus, should not be treated as completely
independent variables), while those farther apart increasingly become completely
different, from a classical statistical perspective. In other words, soil should be
treated as a regionalized variable (Webster and Oliver 1990). Haslam and Tibbett
(2004) urge archaeologists to use geostatistics, such as variogram modeling (e.g.,
217
218
Wells
models) to determine the spatial dependence of soil properties, and they employed
ordinary kriging to produce prediction maps of the spatial distribution of these
properties. This enabled them to interpolate non-sampled locations in an attempt to
study variation in land use practices at eighteenth-century Greaulin in northwest
Scotland. Their work is significant because it highlights the importance of identifying
the spatial covariance of elements and the need for marshalling several lines of
physical and chemical evidence when prospecting and interpreting past activity areas.
A Case Study from Honduras
To explore some of the ways in which sampling design influences interpretations of
ancient human activity patterning as inferred by anthrosol chemistry, I draw on my
work from El Coyote (Fig. 4; Wells 2004). El Coyote was the capital of a regional
settlement hierarchy during the Late and Terminal Classic periods, ca. AD 7001000
(Wells 2003). As mentioned previously, the site is composed of a large, central plaza,
surrounded by pyramidal buildings on the east and range-type structures on the west,
north, and south. There is an elite residential zone with a ball court located
immediately to the south of the plaza and smaller residences to the southwest. This
case study focuses on the main plaza, where the soil underlying the plaza surface can
be characterized as a relatively thin (ca. 0.250.50 m) mollic epipedon that formed
on a limestone substrate. There is no evidence for occupation or use of the plaza
after the site was abandoned in the Postclassic period, around AD 1300.
Excavation of the plaza surface consisted of 40 22-m test units (varying in depth
from 0.251.50 m) arranged in a hexagonal 10050-m lattice. While artifact density
was light (an average of 58 ceramic shards/m3 soil) compared to residential areas at
the site, the evidence suggests that food consumption, marked by the presence of
large ceramic plates and serving dishes, deer bones, and other faunal remains, was
one of the primary activities in the plaza (Wells 2007). Soils were point-sampled
from the center of each test unit at the level of the plaza surface, resulting in 40 units.
In addition, samples were collected between excavation loci at regular 10-m
intervals, yielding an additional 74 units across the surface of the plaza. Finally,
samples were taken opportunistically during the excavations of two low platforms
located in the middle of the plaza; this work yielded another 50 units. In sum,
hexagonal lattice, square lattice, and judgmental strategies were employed to collect
a total of 164 sample units from the plaza.
Soil samples were predominantly soft, compact sandy clays and ranged in color
(moist) from gray (Munsell designation 10YR 4/15/1) to dark grayish brown
(10YR 3/24/2). The mean pH value for all samples was 7.2 (s.d.=0.3). For each
sample for chemical study, approximately 0.5 kg was selected from an area of
roughly 0.1 m in diameter with a corrosion-resistant, stainless steel sample scoop,
which was cleaned with bottled water between sample collections. Samples were
placed directly into sterilized polyethylene bags and sealed for transport back to the
field lab for processing, where they were air-dried and sieved in a 2-mm2-mesh
plastic screen to remove organic debris and clastic materials larger than sand.
Additional sample preparation and analysis procedures were outlined previously.
To consider how different sampling designs might influence my analyses and
interpretations of the chemical data, the overall dataset of 164 units was repeatedly
219
Fig. 4 El Coyote, northwestern Honduras, showing the main plaza from which the samples for this study
were collected
sampled (with replacement) to create seven new (smaller) datasets, each of which
reflects a different sampling strategy: random (n=50), judgmental (n=20), square
high-density lattice (n=74), square low-density lattice (n=40), staggered highdensity lattice (n=39), staggered low-density lattice (n=13), and linked-nested
hexagonal lattice (n=113). By using repeated sampling with replacement from a
finite population, my samples are statistically independent (i.e., the selection and
resulting composition of any given sample has no effect on the selections and
220
Wells
Staggered
4.09 (0.07) 1.88 (0.06) 4.28 (0.29)
high density
Random
Judgmental
Na
Sr
Mn
Zn
1.95 (0.18)
1.90 (0.16)
1.95 (0.19)
1.93 (0.22)
1.92 (0.18)
2.03 (0.16)
1.97 (0.18)
1.91 (0.18)
Ti
0.80 (0.58) 2.01 (0.05) 3.83 (<0.01) 1.65 (0.12) 3.24 (<0.01) 1.31 (0.25) 0.36 (0.92) 1.21 (0.29) 3.86 (<0.01) 2.75 (0.01) 3.31 (<0.01) 2.44 (0.02)
Mg
Data are reported as mean (standard deviation) of base-10 logarithm concentrations (ppm). ANOVA is reported as F (significance) where df=7, 504
ANOVA
Square low
density
Fe
Square high
density
Ca
All data
Ba
A1
Sample
Design
Table 1 Descriptive Statistics and ANOVA Results for Each Element across Sample Designs
222
Wells
Element
Significance
Ca
0.01
0.00
Sr
0.00
Zn
0.07
Ca
0.00
Mn
0.00
Sr
0.00
Ti
0.02
Mn
0.08
Linked-nested hexagonal
Sr
0.01
Ca
0.01
0.01
Mn
0.02
Zn
0.03
223
Fig. 5 Boxplots of selected elemental base-10 logarithm concentrations (ppm) for each sampling design:
A) all data, B) square high density, C) square low density, D) staggered high density, E) staggered low
density, F) random, G) judgmental, H) linked-nested hexagonal. Outliers (open circles) are defined as 1.5
midspread and extreme values (asterisks) are defined as 3.0 midspread
from 13 to 40 units. Still, these findings suggest that, in order to select the most
appropriate sample lattice design, one ought to take into account the different kinds
of variation in soil properties.
One kind of variation that is important can be described as homoscedastic, in
which there is an equal distribution of variation across parameter estimates. Many
classical probability statistical tests require so-called homogeneity of variance as a
basic assumption, so as to allow each parameter an equal chance of contributing to
the model, for example, that produced by way of regression analysis or ANOVA.
224
Wells
Staggered
High Density
Square
High Density
All Data
Random
90
90
90
90
80
80
80
80
70
70
70
70
60
60
60
60
50
50
50
50
40
40
40
40
30
30
30
30
20
20
20
20
10
10
10
20
30
40
50
10
10
20
30
40
50
10
10
Square
Low Density
Linked-Nested
Hexagonal
20
30
40
50
10
Staggered
Low Density
90
80
80
70
70
60
60
60
50
50
50
40
40
40
40
30
30
30
30
20
20
20
20
20
30
40
50
40
50
70
10
10
30
Judgmental
90
10
20
60
50
10
10
20
30
40
50
10
10 20 30 40 50
10
20
30
40
50
Fig. 6 Kriged image maps showing the distribution of extractable soil P in base-10 logarithm
concentrations (ppm) for each sample design. Darker hues correspond to higher concentrations of P
Square
High Density
Staggered
High Density
400000
250000
300000
250000
200000
200000
150000
100000
150000
250000
200000
150000
400000
300000
200000
50000
10
15
20
25
0
0
30
100000
50000
50000
5
Lag Distance
Linked-Nested
Hexagonal
10
15
20
25
0
0
30
10
15
20
25
0
0
30
Judgmental
1200000
1000000
100000
100000
120000
100000
80000
60000
10
15
20
25
30
600000
400000
200000
20000
Lag Distance
800000
40000
50000
50000
5
30
140000
150000
0
0
25
Staggered
Low Density
Variogram
150000
20
Square
Low Density
Variogram
Variogram
200000
15
Lag Distance
200000
250000
10
Lag Distance
160000
300000
Lag Distance
350000
Variogram
500000
300000
100000
100000
0
0
600000
350000
Variogram
Variogram
350000
Variogram
400000
0
0
Random
300000
450000
Variogram
225
10
15
20
Lag Distance
25
30
0
0
10
15
20
Lag Distance
25
0
0
10
15
20
25
Lag Distance
Fig. 7 Experimental (nodal line) and theoretical (smooth line) variogram models of extractable soil P in
base-10 logarithm concentrations (ppm) for each sample design: all data (linear), square high-density
(Gaussian), staggered high-density (spherical), random (linear), linked-nested hexagonal (exponential),
square low-density (exponential), staggered low-density (linear), judgmental (linear)
Recommendations
My research into sampling design permits three recommendations for sampling, with
the proviso that the methods reported here need to be extended to test datasets from
other archaeological sites where anthrosols were produced from different soil-forming
factors in different soil environments. First, it is essential that one understands the
properties of the soils under analysis before selecting a sampling design. The extent to
which soils can store chemical information about past human activities depends on
several properties, including texture and density/porosity (Holliday and Gartner 2007;
226
Wells
Oonk et al. 2009), which are straightforward to measure in a field laboratory. Soil
texture represents the soil size fractions (sand, silt, and clay) of the mineral
component of soils that vary in size from stone to gravel to powder. Clayey soils are
very effective at trapping cations in clay interlayers and at adsorbing ions to clay
particle surfaces. Sandy soils, on the other hand, are more conducive to water filtration
and erosion and allow cations to migrate more readily in soil bodies both vertically and
horizontally. Soil texture is obviously related to density (the mass of soil per unit
volume) and porosity (the portion of the soil volume occupied by air and water). Clayey
soil tends to exhibit lower bulk density and higher porosity than sandy soil. To
determine soil texture, the pipette (Tan 2005: 169170) and hydrometer (Tan 2005:
162166) techniques are most common. Both techniques involve mixing the sample
with a dispersion agent (e.g., sodium metaphosphate), shaking or stirring the mixture,
and then allowing the resulting soil suspension to settle. Gravity settles the coarse
sand particles quickly (within 40 s), followed by the silt (after 2 h) and then clay
(after 8 h). For measuring density and porosity (see Tan 2005: 175188), one can
weigh standard volumes of oven-dried samples (to determine bulk density) and then
measure the moisture content of the soil using the gravimetric method (to determine
particle density). Comparing the two results (particle density minus bulk density,
divided by particle density, multiplied by 100) yields the percentage of pore space.
Understanding soil properties can be done most efficiently by taking a random
sample of soils from throughout the study area, which can be treated as the entire
sample frame or else subdivided into sampling separate strata from which a
consistent number of samples can be collected. There is no good rule of thumb for
the number of units to obtain for the pilot study sample, but it seems reasonable,
based on what is known about intra-pedon variability of soil properties (Entwistle et
al. 2007; Marshall 2001), that at least one sample per 5 m2 should be collected for
small-scale spaces and 20 m2 for larger sample universes. By using a pilot study to
understand variation in soil properties across the study area, it can be determined if
the sample density should be increased because of poor soil conditions or a high
degree of variability in soil properties over small distances. The use of geostatistical
interpolation at this stage of the research has been suggested as a productive means
to determine the optimal sampling interval by solving the kriging equations for several
sampling intensities and plotting the maximum kriged variance against the sample
spacing (Atkinson 1996; Burgesse et al. 1981; McBratney and Webster 1981).
Second, if sampling density is less than 1% of the study universe (e.g., the low
density samples in the El Coyote case), it may be a good idea to use a random
sampling strategy. This is especially the case when the target soil properties vary
significantly across the sample frame. Judgmental sampling should always be done
with caution, however. The resulting soil chemical patterns may not be representative of the range of ancient human activities represented in the study space. It is
also important to keep in mind that any data acquired from randomly obtained
samples are not amenable to kriging and other kinds of spatial analysis that involve
interpolation, because the data violate many assumptions about variance. Based on
my own experience using judgmental sampling, I recommend that it is best used in
combination with other sampling strategies.
Third, for sampling densities over 1% of the study universe (e.g., the highdensity samples in the El Coyote case), the selection of sampling strategy should be
227
restricted to a lattice matrix. For very large sample sizes relative to the size of the
space being sampled, lattices appear to be less sensitive to outliers or batches of
samples with generally high concentrations (compared to other batches). In this
article, I offer the linked-nested hexagonal strategy as an alternative to the standard
lattices (i.e., square and staggered grids) for use when sampling density meets or
exceeds 1% of the sample universe, because standard lattices may bias inferences
about activity patterning. Still, lattices of any kind should be used with caution. For
systematically obtained samples, the square lattice produces very different results
from the staggered lattice. Both strategies incorrectly assume that soil chemical
behavior is patterned at regular intervals across any given space, which is only
sometimes the case. Thus, systematic sampling using a lattice may, in some cases,
systematically miss evidence for activity loci.
With applications in both geostatistics (Atkinson 1996; Yfantis et al. 1987) and
archaeology (Kintigh 1988; Krakker et al. 1983), it has been demonstrated that a
hexagonal lattice is more efficient than a square lattice. I would add to this point that
the hexagonal lattice has the additional advantage of having a smaller maximum lag
distance on adjacent sample unit transects, thus decreasing the maximum kriging
variance when using kriging to interpolate the data (see Burgesse et al. 1981).
Moreover, regular and closely spaced intervals lend themselves especially well to
contour plotting, since the interpolation between data points is more closely
constrained and, therefore, more representative of the underlying trend in the data
than if collected in a random or opportunistic fashion (Entwistle et al. 2007: 397).
If recent trends in archaeological research are any indication, anthrosol chemistry
will soon become a standard part of archaeologists' analytical toolkits, akin to
radiocarbon dating or chemical provenance studies of artifacts. A pressing issue is
sampling, for which this article opens a dialog that needs to be extended. Sampling
archaeological soils for chemical study in spatial analysis is a complex problem that
can be addressed productively through the accumulation of collective experience in
various test cases, such as those presented in this article.
Acknowledgments I would like to thank Sandra L. Lpez Varela, Christopher D. Dore, and Manuel R.
Palacios-Fest for inviting me to participate in the original symposium in which a preliminary draft of this
paper was presented at the 2006 Annual Meeting of the Society for American Archaeology in San Juan,
Puerto Rico, and for all their hard work on its subsequent expansion and publication. Research at El
Coyote was conducted with the permission and assistance of the Instituto Hondureo de Antropologa e
Historia. I am exceedingly grateful to Patricia A. Urban and Edward M. Schortman for allowing me to
conduct this research and for their support throughout the project. Funding for my research was provided
by the National Science Foundation (BCS-0108742) and the Wenner-Gren Foundation for Anthropological
Research (GR. 6810). Soil analysis was conducted with the support and advice of James H. Burton and
T. Douglas Price at the Laboratory for Archaeological Chemistry at the University of Wisconsin, Madison.
Lpez Varela, Dore, Karla L. Davis-Salazar, and three anonymous reviewers read drafts of this manuscript and
provided very useful comments that helped improve the arguments in this paper.
References
Atkinson, P. M. (1996). Optimal sampling strategies for raster-based geographical information systems.
Global Ecology and Biogeography Letters, 5(4/5), 271280.
228
Wells
Bethell, P., & Mt, I. (1989). The use of soil phosphate analysis in archaeology: A critique. In J.
Henderson (Ed.), Scientific analysis in archaeology and its interpretation, institute of archaeology
(pp. 129). Los Angeles: University of California.
Binford, L. R. (1964). A consideration of archaeological research design. American Antiquity, 29(4), 425
441.
Boekhold, E. E., & Van der Zee, S. (1992). Significance of soil chemical heterogeneity for spatial behavior
of cadmium in field soils. Soil Science Society of America Journal, 56(3), 747754.
Burgesse, T. M., Webster, R., & McBratney, A. B. (1981). Optimal interpolation and isarithmic mapping
of soil properties: IV, sampling strategy. Journal of Soil Science, 32(3), 643659.
Burton, J. H., & Simon, A. W. (1993). Acid extraction as a simple and inexpensive method for
compositional characterization of archeological ceramics. American Antiquity, 58(1), 4559.
Campbell, J. B., & Edmonds, W. J. (1984). The missing geographic dimension to soil taxonomy. Annals of
the Association of American Geographers, 74(1), 8397.
Casteel, R. W. (1970). Core and column sampling. American Antiquity, 35(4), 465467.
Champion, T., Cuming, P., & Shennan, S. J. (1996). Planning for the past, vol. 3. Decision-making and field
methods in archaeological evaluation. London: English Heritage and University of Southampton.
Cook, S. R., Clarke, A. S., & Fulford, M. G. (2005). Soil geochemistry and detection of early roman
precious metal and copper alloy working at the roman town of calleva atrebatum (Silchester,
Hampshire, UK). Journal of Archaeological Science, 32(5), 805812.
Crowther, J. (1997). Soil phosphate surveys: critical approaches to sampling, analysis and interpretation.
Archaeological Prospection, 4(2), 93102.
Entwistle, J. A., Abrahams, P. W., & Dodgshon, R. A. (2000). The geoarchaeological significance and
spatial variability of a range of physical and chemical soil properties from a former habitation site, isle
of skye. Journal of Archaeological Science, 27(4), 287303.
Entwistle, J. A., McCaffrey, K. J. W., & Dodgshon, R. A. (2007). Geostatistical and multi-elemental
analysis of soils to interpret land-use history in the Hebrides, Scotland. Geoarchaeology: An
International Journal, 22(4), 391415.
Fisher, E., Thornton, B., Hudson, G., & Edwards, A. C. (1998). The variability in total and extractable soil
phosphorus under a grazed pasture. Plant and Soil, 203(2), 249255.
Goldberg, P., & Macphail, R. I. (2006). Practical and theoretical geoarchaeology. Malden: Blackwell
Publishing.
Goovaerts, P. (1999). Geostatistics in soil science: state-of-the-art and perspectives. Geoderma, 89, 145.
Hammond, L. C., Pritchett, W. L., & Chew, V. (1958). Soil sampling in relation to soil heterogeneity. Soil
Science Society of America Proceedings, 22(6), 548552.
Haslam, R., & Tibbett, M. (2004). Sampling and analyzing metals in soils for archaeological prospection:
a critique. Geoarchaeology, 19(8), 731751.
Heizer, R. F. (1949). A manual of archaeological field methods. Millbrae: National Press.
Hester, T. R., Shafer, H. J., & Feder, K. L. (1997). Field methods in archaeology (7th ed.). Mountain
View: Mayfield.
Holliday, V. T. (2004). Soils in archaeological research. Oxford: Oxford University Press.
Holliday, V. T., & Gartner, W. G. (2007). Methods of soil p analysis in archaeology. Journal of
Archaeological Science, 34(2), 301333.
Howell, T. L. (1993). Evaluating the utility of auger testing as a predictor of subsurface artifact density.
Journal of Field Archaeology, 20(4), 475484.
Hutson, S. R., & Terry, R. E. (2006). Recovering social and cultural dynamics from plaster floors: chemical
analyses at ancient chunchucmil, Yucatan, Mexico. Journal of Archaeological Science, 33(3), 391
404.
Kintigh, K. W. (1988). The effectiveness of subsurface testing: a simulation approach. American Antiquity,
53(4), 686707.
Kitanidis, P. K. (1997). Introduction to geostatistics: applications in hydrogeology. Cambridge:
Cambridge University Press.
Kitanidis, P. K., & Shen, K.-F. (1996). Geostatistical interpolation of chemical concentration. Advances in
Water Resources, 19(6), 369378.
Kozar, B., Lawrence, R., & Long, D. S. (2002). Soil phosphorus and potassium mapping using a spatial
correlation model incorporating terrain slope gradient. Precision Agriculture, 3(4), 407412.
Krakker, J. J., Shott, M. J., & Welch, P. D. (1983). Design and evaluation of shovel-test sampling in
regional archaeological survey. Journal of Field Archaeology, 10(4), 469480.
Lark, R. M. (2003). Two robust estimators of the cross-variogram for multivariate geostatistical analysis of
soil properties. European Journal of Soil Science, 54(1), 187202.
229
Lewis, R. J., Foss, J. E., Morris, M. W., Timpson, M. E., & Stiles, C. A. (1993). Trace element analysis in
pedo-archaeology studies. In J. E. Foss, M. E. Timpson, & M. W. Morris (Eds.), Proceedings of the
1st international conference on pedo-archaeology, special publication 93-03 (pp. 8188). Knoxville:
University of Tennessee Agricultural Experiment Station.
Lightfoot, K. G. (1986). Regional surveys in the eastern United States: the strengths and weaknesses of
implementing subsurface testing programs. American Antiquity, 51(3), 484504.
Lindsay, W. L., & Norvell, W. A. (1978). Development of a DTPA test for zinc, iron, manganese, and
copper. Soil Science Society of America Journal, 42(3), 421428.
Lloyd, C. D., & Atkinson, P. M. (2004). Archaeology and geostatistics. Journal of Archaeological
Science, 31(2), 151165.
Marshall, A. (2001). Functional analysis of settlement areas: prospection over a defended enclosure of iron age
date at the bowsings, guiting power, Gloucestershire, UK. Archaeological Prospection, 8(2), 79106.
McBratney, A. B., & Webster, R. (1981). The design of optimal sampling schemes for local estimation and
mapping of regionalized variables, II: program and examples. Computers and Geosciences, 7(4), 335365.
Mehlich, A. (1978). New extractant for soil test evaluation of phosphorus, potassium, magnesium, calcium,
sodium, manganese, and zinc. Communications in Soil Science and Plant Analysis, 9(6), 477492.
Meul, M., & Van Meirvenne, M. (2003). Kriging soil texture under different types of nonstationarity.
Geoderma, 112(34), 217233.
Middleton, W. D., & Price, T. D. (1996). Identification of activity areas by multi-elemental
characterization of sediments from modern and archaeological house floors using inductively coupled
plasma-atomic emission spectroscopy. Journal of Archaeological Science, 23(5), 673687.
Nance, J. D., & Ball, B. F. (1986). No surprises? The reliability and validity of test pit sampling. American
Antiquity, 51(3), 457483.
Oonk, S., Slomp, C. P., & Huisman, D. J. (2009). Geochemistry as an aid in archaeological prospection
and site interpretation: current issues and research directions. Archaeological Prospection, 16, 3551.
Orton, C. (2000). Sampling in archaeology. Cambridge: Cambridge University Press.
Parnell, J. J., Terry, R. E., & Golden, C. (2001). Using in-field phosphate testing to rapidly identify
middens at Piedras Negras, Guatemala. Geoarchaeology: An International Journal, 16(8), 855873.
Parnell, J. J., Terry, R. E., & Nelson, Z. (2002). Soil chemical analysis applied as an interpretive tool for
ancient human activities in Piedras Negras, Guatemala. Journal of Archaeological Science, 29(4),
379404.
Price, J. C., Hunter, R. G., & McMichael, E. V. (1964). Core drilling in an archaeological site. American
Antiquity, 30(2), 219222.
Redman, C. L. (1987). Surface collection, sampling, and research design: a retrospective. American
Antiquity, 52(2), 249265.
Reed, N. A., Bennett, J. W., & Porter, J. W. (1968). Solid core drilling of monks mound: technique and
findings. American Antiquity, 33(2), 137148.
Robertson, G. P., Crum, J. R., & Ellis, B. G. (1993). The spatial variability of soil resources following
long-term disturbance. Oecologia, 96(4), 451456.
Shott, M. J. (1985). Shovel-test sampling as a site discovery technique: a case study from Michigan.
Archaeological Prospection, 12(4), 457468.
Shott, M. J. (1987). Feature discovery and the sampling requirements of archaeological evaluations.
Journal of Field Archaeology, 14(3), 359371.
Shott, M. J. (1989). Shovel-test sampling in archaeological survey: comments on nance and ball, and
lightfoot. American Antiquity, 54(2), 396404.
Snedecor, G. W., & Cochrane, W. G. (1980). Statistical methods (7th ed.). Ames: Iowa State University
Press.
Stein, J. K. (1986). Coring archaeological sites. American Antiquity, 51(3), 505527.
Stein, A., & Ettema, C. (2003). An overview of spatial sampling procedures and experimental design
studies for ecosystem comparisons. Agriculture, Ecosystems & Environment, 94(1), 3147.
Tan, K. H. (2005). Soil sampling, preparation, and analysis (2nd ed.). Boca Raton: CRC Press.
Terry, R. E., Hardin, P. J., Houston, S. D., Nelson, S. D., Jackson, M. W., Carr, J., et al. (2000).
Quantitative phosphorus measurement: a field test procedure for archaeological site analysis at Piedras
Negras, Guatemala. Geoarchaeology: An International Journal, 15(2), 151166.
Usowicz, B., & Kossowski, J. (2001). Spatial variation of soil moisture and sampling strategy. In J.
Blahovec & M. Libra (Eds.), Proceedings of the international conference on physical methods in
agriculture: Approach to precision and quality (pp. 319323). Prague: Czech University of Agriculture.
Webster, R., & Oliver, M. A. (1990). Statistical methods in soil and land resource survey. Oxford: Oxford
University Press.
230
Wells
Welch, B. L. (1951). On the comparison of several mean values: an alternative approach. Biometrika, 38(34),
330336.
Wells, E. C. (2003). Artisans, chiefs, and feasts: Classic period social dynamics at El Coyote, Honduras,
Ph.D. dissertation, Arizona State University, Tempe.
Wells, E. C. (2004). Investigating activity patterns in prehispanic plazas: weak acid-extraction ICP/AES
analysis of anthrosols at classic period El Coyote, Northwest Honduras. Archaeometry, 46(1), 6784.
Wells, E. C. (2007). Faenas, ferias, and fiestas: Ritual finance in ancient and modern Honduras. In E. C.
Wells & K. L. Davis-Salazar (Eds.), Mesoamerican ritual economy: Archaeological and ethnological
perspectives (pp. 2965). Boulder: University Press of Colorado.
Wells, E. C., Novotny, C., & Hawken, J. R. (2007). Predictive modeling of soil chemical data by ICP-OES
reveals the uses of ancient Mesoamerican plazas. In M. D. Glascock, R. J. Speakman, & R. S.
Popelka-Filcoff (Eds.), Archaeological chemistry: Analytical techniques and archaeological interpretation (pp. 210230). Washington, DC: American Chemical Society.
Wells, E. C., & Terry, R. E. (2007). Introduction to the special issue: advances in geoarchaeological
approaches to anthrosol chemistry, part II: activity area analysis. Geoarchaeology: An International
Journal, 22(4), 387390.
Wells, E. C., Terry, R. E., Hardin, P. J., Parnell, J. J., Houston, S. D., & Jackson, M. W. (2000). Chemical
analyses of ancient anthrosols in residential areas at Piedras Negras, Guatemala. Journal of
Archaeological Science, 27(5), 449462.
Wells, E. C., & Urban, P. A. (2002). An ethnoarchaeological perspective on the material and chemical
residues of communal feasting at El Coyote, Northwest Honduras. In P. Vandiver, M. Goodway, & J.
Mass (Eds.), Materials issues in art and archaeology VI, MRS proceedings vol. 712 (pp. 193198).
Warrendale: Materials Research Society.
Wobst, H. M. (1983). We cant see the forest for the trees: Sampling and shapes of archaeological
distributions. In J. A. Moore & A. S. Keene (Eds.), Archaeological hammers and theories (pp. 37
85). New York: Academic Press.
Yfantis, E. A., Flatman, G. T., & Behar, J. V. (1987). Efficiency of kriging estimation for square,
triangular, and hexagonal grids. Mathematical Geology, 19(3), 183205.
Young, F. J., & Hammer, R. D. (2000). Defining geographic soil bodies by landscape position, soil
taxonomy, and cluster analysis. Soil Science Society of America Journal, 64(3), 989998.
Zhang, C., Jordan, C., & Higgins, A. (2007). Using neighborhood statistics and GIS to quantify and
visualize spatial variation in geochemical variables: an example using Ni concentrations in the
topsoils of Northern Ireland. Geoderma, 137(34), 466476.