You are on page 1of 11

The Limitations of Traditional Surveying Techniques in a Forested Environment

Author(s): Diana Alexander


Source: Journal of Field Archaeology, Vol. 10, No. 2 (Summer, 1983), pp. 177-186
Published by: Taylor & Francis, Ltd.
Stable URL: http://www.jstor.org/stable/529608
Accessed: 01-12-2016 04:06 UTC
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted
digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about
JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms

Taylor & Francis, Ltd. is collaborating with JSTOR to digitize, preserve and extend access to Journal of
Field Archaeology

This content downloaded from 148.223.96.146 on Thu, 01 Dec 2016 04:06:18 UTC
All use subject to http://about.jstor.org/terms

The Limitations of Traditional Surveying Techniques


in a Forested Environment

Diana Alexander
Simon Fraser University

Burnaby, British Columbia


Canada

Monetary and time constraints have necessitated the continued use of traditional surveying techniques, despite the tendency of non-probabilistic surveys
to yield unrepresentative and biased samples. Unfortunately the lack of information on the extent and nature of this sample bias has made it almost impossible to evaluate the results of such surveys. By examining the results of a
traditional and a probabilistic survey of the same area of NE British ColumbiaS this report quantitatively measures the bias inherent in a traditional, boreal forest survey.

Introduction

comparing the results of a traditional survey with a subThe traditional archaeological approach to site survey surface, probabilistic survey of the same region? this parelies heavily on intuitive concepts to determine potentialper quantitatively measures the degree and form of bias
site locations. 1 Although these concepts have often been that can result from the use of a traditional sampling
gained from intimate familiarity with the primary data, methodology.
The disadvantages of a traditional approach to site surthis ''gut-level" of analysis introduces the possibility
vey are well recognized. The major shortcoming of this
that the resulting conclusions reflect preconceived perapproach is that it is non-probabilistic and incapable of
sonal biases. To overcome this methodological problem
yielding a valid estimate of the risks of error.3 Any atmany archaeologists have resorted to the use of probatempt to determine biases in the procedure is further
bilistic sampling designs.2 Used correctly, this approach
thwarted by the general lack of explicit descriptions of
can provide an unbiased record of site locations. By
the sampling methodology used in traditional surveys.
As a consequence it is almost impossible quantitatively
1. See, e.g.? Gordon R. Willey, Prehistoric Settlement Patterns in
or qualitatively to evaluate or replicate these studies. The
the New World (Wenner-Gren Foundation for Anthropological Reonly means of accurately determining the validity of trasearch: New York 1956); Stuart Struever, "Woodland Subsistence
ditional survey results is by comparing these results with
Settlement Systems in the Lower Illinois Valley,'' in Sally R. Binford
those of a probabilistic survey of the same region.
and Lewis R. Binford, eds., New Perspectives in Archeology (Aldine
Most regional studies, both traditional and probabilisPublishing Company: Chicago 1968) 285-312.
tic,
have been undertaken in areas such as the American
2. See, e.g., David H. Thomas, "Regional Sampling in Archaeology.
Southwest where there is ideal topography and only sparse
A Pilot Great Basin Research Design," University of California Arground cover. It is only in the last few years that probchaeological Survey, Annual Report 1 1 (Los Angeles 1969) 87-100;
Michael B. Schiffer and John H. House, ''The Cache River Archaeabilistic surveys have been attempted in areas with the
ological Project,'' Arkansas Archaeological Survey, Publication in
heavy ground cover typical of the boreal forest found in
Archaeology 8 (1975); Charles A. Reher, Settlement and Subsistence
the
study area.4 The surveying problems encountered in
Along the Lower Chaco River: The C.G.P. Survey (University of New
these forested areas are often very different from those
Mexico Press: Albuquerque 1977); Bruce F. Ball, "Heritage Resources of the Northeast Coal Study Area," Report to the Archaeology
found in more arid regions. Consequently, the results of
Division, Heritage Conservation Branch, Government of British Columbia (1978); Jack D. Nance, "Non-site Sampling in the Lower
Cumberland River Valley, Kentucky," MCJA 5 (1980) 169-191; Brian 3. Hubert M. Blalock Jr. Social Statistics (McGraw-Hill Book Company: New York 1972) 527.
E. Spurling, "The Site C Heritage Resource Inventory and Assessment: Impacts and Mitigation,' Report to British Columbia Hydro
4. See, for example, W. A. Lovis, i'Quartersections and Forests: An

and Power Authority (1980).

Example of Probability Sampling in the Eastern Woodlands,'' AmAnt

This content downloaded from 148.223.96.146 on Thu, 01 Dec 2016 04:06:18 UTC
All use subject to http://about.jstor.org/terms

178 Limitations of Traditional Surveying Techniques in a Forested EnvironmentlAlexander


this analysis are not universally applicable but are par-

Table 1. Strata used in the probabilistic survey. After

ticular to a forest regime. Moreover, the probabilistic

Spurling, op. cit. (in note 2) 19.

survey used in this comparison required the excavation


of sampling units, a procedure uncommon both to traditional and probabilistic surveys. Many of the biases
encountered in this analysis, therefore, reflect the use of
surficial, as well as non-probabilistic, surveying tech.

nlques.

Area
Strata

South

(sq.

Upper

km.)

32.50

(valley rim to safeline, south bank)


South

Intermediate

14.25

(intermediate terrace complex, south bank)


South

Methodology

Lower

18.00

(floodplain, south bank)

The surveys used in this comparative analysis were


part of a cultural heritage resource inventory and assessment of the Peace River Valley of NE British Columbia
(FIG. 1). The statistical survey, undertaken during the
summer months of 1977 and 1978, was concerned solely
with the effects of flooding and other impacts to be produced by the proposed construction of the Site C hy-

Islands

9.50

(islands in river)

North

Lower

21.50

(floodplain, north bank)

North Intermediate 25 .75


(intermediate terrace complex, north bank)
North

Upper

29.25

(valley rim to safeline, north bank)

droelectric dam.S This survey used archaeologically


related information to divide the valley into strata prior
to sampling.6 Specifically, seven strata were created on

limited to a survey of the Site One reservoir, the survey

the basis of topography as it related to terrace develop-

was of necessity hurried and superficial. Consequently,

ment and side-of-the-river (TABLE 1). A random sample

emphasis was placed on surveying the north side of the

of units was selected from each strata. These units or

river where there was less ground cover and easy access.

quadrats were 500 m. on a side. Each quadrat was, in

In 1976, the second half of this traditional survey was

turn, subdivided into 100, 50 m. x 50 m. subunits.

confined to the Site C reservoir. Emphasis here was placed

Within each of these subunits a S m. x S m. target area

on examining the previously overlooked terraces on the

was chosen by using a statified, systematic unaligned

south side of the river.

sampling scheme. Finally, a pit 1 m. deep was excavated

Ideally, both surveys should have identical boundaries

in this target area and the matrix was screened and ex-

to ensure that the success of the surveying techniques

amined for cultural items. In broken fields where all Ah

can be assessed without concern for other external fac-

horizons7 had been disturbed, a systematic surface in-

tors, such as faunal and floral communities. To over-

spection was substituted for the test excavation proce-

come the discrepancies in the size of the two study areas,

dure.8 The assumption was that any subsurface materials

those sites discovered in the traditional survey that were

would have some representation on the surface.

outside the boundaries established for the probabilistic

The traditional survey was begun in the summer of

survey were excluded from the analysis. The resulting

1974 by a small research team which undertook an ex-

study area encompassed 210 sq. km.10 The total land

tensive survey of the proposed Site One, Site C and Site

area, however, is reduced to 150.75 sq. km. by dis-

E pondages (FIG. 1).9 Since the original mandate was

counting areas covered by water, steep topography, and


land disturbed by large-scale landslides in historical
times.ll

41 (1976) 364-372; Nance, op. cit. (in note 2); idem, "Regional

For comparative purposes, both the traditional and the

Subsampling and Statistical Inference in Forested Habitats," AmAnt

probabilistic surveys should have had equal access to

44 (1979) 172-176.

money, manpower, and time. Unfortunately this was not

5. Spurling, op. cit. (in note 2) 11.


6.

Ibid.

18.

7. An Ah horizon is a mineral horizon in which organic matter has


accumulated as a result of biological activity.
8. Brian E. Spurling, "The Site C Heritage Resource Inventory and
Assessment: Substantive Results," Report to British Columbia Hydro

the case. Although both were allotted basically the same


amount of time, the traditional survey had only a fifth
of the manpower and a sixth of the money. This circumstance may be a significant source of bias in the comparison because with more manpower and money the
traditional survey might have achieved a more accurate

and Power Authority (1980) 144.

9. Brian E. Spurling, Finola Finlay, and K. R. Fladmark, "Report


on the Peace River Archaeological Survey and Salvage Project," Report to British Columbia Hydro and Power Authority (1976) 20-23.

10. Spurling, op. cit. (in note 8) 133.


11. Ibid.

This content downloaded from 148.223.96.146 on Thu, 01 Dec 2016 04:06:18 UTC
All use subject to http://about.jstor.org/terms

Journal of Field ArchaeologylVol. 10, 1983 179

Figure 1. The study area.

possibility that any biases would have been emphasized

ing aboriginal sites. Relying on personal experience and


intuition, many archaeologists have developed, perhaps

rather than reduced.

unconsciously, a list of criteria for locating sites. Some

account of site distribution. There is, however, a strong

Table 2 records the relative productivity of the two

of the variables included in this category are proximity

sampling techniques in terms of the number of sites located. The figures in this table are based on information
provided by Spurlingl2 and by site forms. The section

to water, certain ecotones and major river confluences,


shelters and protection from the elements and a view
for the observation of game and strangers. The assumed

of this paper concerned with the traditional survey uses

importance of these variables in determining site location

a figure of 133 aboriginal sites. The discrepancy between

has been expressed by Brose,l3 Jochim,l4 and Williams,

this number and the 109 prehistoric sites in the table


results from two factors: l) some 'shistorical" sites have

Thomas, and Bettinger. 1 5

an "aboriginal" component, and 2) the boundaries of

as a basis for determining survey strategy. That is, the

Regrettably, this iicommon knowledge" is often used

the study area have changed slightly since the original

archaeologist concentrates his surveying efforts in those

calculations were made.

portions of the study area where he expects to find sites.

The aim of this study was to test whether or not the

Discovery of sites in these predicted, high-density lo-

traditional, surficial survey produced biased results. It

cations may reflect differential treatment of such areas

was assumed that there were three major factors influ-

rather than prehistoric settlement pattems. Unfortunately

encing the discovery of sites. The first is the nature of

these predictions are rarely made in print and it is often

the survey. The traditional survey depended heavily on


surface exposure for location of cultural material. Vari-

impossible to evaluate their effect on the survey results.


Finally, unbiased results cannot be achieved when

ables such as ground cover and changes in topography,


therefore, were expected to influence the detection of

13. David S. Brose, "Locational Analysis in the Prehistory of North-

sites in this survey. Obviously this emphasis on exposure

east Ohio," in Charles C. Cleland, ed., Cultural Change and Continuity (Academic Press, Inc.: New York 1976) 3-18.

can bias the results and provide an inaccurate record of


the prehistoric settlement patterns, with disproportionate
representation of late sites and sites in disturbed areas.
The second factor is the "common knowledge" as-

14. Michael A Jochim, ffiIunter-Gatherer Subsistence and Settlement:


A Predictive Model (Academic Press, Inc.: New York 1976) Chapter
4.

similated by the surveyors and used as a basis for locat-

15 Leonard Williams, David H. Thomas, and Robert Bettinger,


"Notions to Numbers: Great Basin Settlements as Polythetic Sets,"
in Charles L. Redman, ed., Research and Theory in Current Anthro-

12. Ibid. 88.127.

pology (Wiley: New York 1973) 226-227.

This content downloaded from 148.223.96.146 on Thu, 01 Dec 2016 04:06:18 UTC
All use subject to http://about.jstor.org/terms

180 Limitations of Traditional Suseying Techniques in a Forested Environment/Alexander


Table 2. Relative productivity of
the two surveying techniques in

terms of the number of sites located


by each technique.

temporal changes are ignored. Using ethnographic data

tances from these features were noted since it was rea-

and historical documentation it is often possible to re-

soned that different results may be produced when vertical

construct the aboriginal settlement pattern of the proto-

distance was considered. Sites in the vertical category

historic period. This knowledge may unduly influence

were within both the horizontal and vertical ranges. 16

the surveyor to survey more intensively those areas oc-

Interval distances, though arbitrary in themselves,

cupied during that period. With time, however, the set-

might have provided a more sensitive test. The low cell

tlement patterns may not only change within the same

frequencies characteristic of the probabilistic data, how-

environment, the environment itself, as reflected in the

ever, necessitated the use of Fisher's Exact Test. This

topography and vegetation, may be vastly altered. The

test disallows the use of 2 x R tables that would result

effect of these changes on site location should be care-

from a series of interval distances.

fully considered when undertaking any survey.

The first objective of this analysis was to test for sta-

Table 3 provides a summary of the variables selected

tistical independence between the variables and the lo-

for comparative analysis. Since some of the chosen vari-

cation of 133 aboriginal sites uncovered by traditional

ables require a measurement of the distance to an envi-

surveying techniques. Three non-parametric tests of in-

ronmental feature (e.g., proximity to a river confluence),

dependence can be used with nominal data in a one-

it was necessary to define a range that would approxi-

sample design: the binomial test, the chi-square test and

mate this distance. The range was selected somewhat

16. Although most measurements in this study are metric, vertical

arbitrarily, but it should reflect the surveyor's concept

distance is recorded in feet since all available maps have non-metnc

of "near" and "far". Both horizontal and vertical dis-

contour lines.

This content downloaded from 148.223.96.146 on Thu, 01 Dec 2016 04:06:18 UTC
All use subject to http://about.jstor.org/terms

I
Journal of Field Archaeology/Vol. 10, 1983 181
the G-test (the likelihood ratio version of chi-square).l7

Table 3. Variables selected for analysis

In this case the binomial test was impractical because of

DOMINANT GROUND COVER ON SITES

large frequencies which made computation unwieldy. The


chi-square test ';is the traditional way of analyzing such

l
l l

cases, but . . . G has general theoretical advantages

over X2, as well as being computationally simpler for


tests of independence.''l8 Moreover, ;;G appears to fol-

Cultivated Field Natural Cover

low the X2 - distribution a bit more closely. " t9 For these

reasons, the G-test was employed using the following


formula:

Pasture Broken Prairie Forest

oj

ln

i)

where o = observed frequency

Evergreen

Deciduous

e = expected frequency
k = number of cells

Variable

Range
Horizontal Vertical

The observed frequencies were easily calculated. FirstS

(m

(ft.)

the dominant ground cover on each site was noted and

recorded. Similarly it was determined whether or not

Near Terrace Edge 100

each site was within the ranges outlined for the remain-

Near Powerline or Road 100 200

ing variables. The total number of sites belonging to each


category constituted the observed frequencies. For example, 20 sites were within 100 m. (horizontally) of a

secondary river drainage while 113 sites were outside


this range (TABLE 4).

Near Present Confluence 500 200


Near Remnant C::onfluence 500 200
Near

Peace

River

100

100

Near Secondary Drainage 100 100


Near

Bog

100

200

The expected frequencies were based on information


provided by Spurling,20 Blood,2l and Thurber Consultants Ltd.22 In the four tests comparing types of ground
cover, the expected frequencies were based on the relative proportions of each type below the floodline. Since
there is little difference in land use and vegetation below
the floodline and between the floodline and the safeline,
it was assumed that this estimate was representative of
the entire study area. A similar assumption was made in

Table 4. Example of the data used in calculating

probabilities for the traditional survey. This table shows how


the probability that secondary river drainages are associated
with archaeological sites was determined for the entire study
area.
Near a Secondary River Drainage
+

17. l;t. Sokal and F. l;tohlf, Biometrv (W. H. Freeman: San Francisco

Number

oi

20

113

33

of

1969) Chapter 16.


18.

Ibid.

550.

19.

Ibid.

560.

20. Spurling, op. cit (in notes 2 and 8).

21. Donald A. Bloods ''Peace l;tiver Site C Hydroelectric Develop-

Sites

ei

4.6

128.4

G = 30.0086; df = 1; p < .001

(df = degrees of freedom; p = probability that archaeological sites are associated with secondary river
drainages. )

ment, Environmental and Socioeconomic Assessment, Wildlife Subl;teport," Report to British Columbia Hydro and Power Authority
(1977).

22. Thurber Consultants Ltd ., " Development Proposals and l;tesource Inventories, Sites C and E Hydroelectric Development Proposals, Lower Peace l;tiver Environmental Study, Vol. I," l;teport to
British Columbia Hydro and Power Authority (1976); idem, ''Peace

calculating the amount of land near bogs. Maps of the


Peace River Valley provided additional information on

l;tiver Site C Hydroelectric Development, General Land Use Studies,"

the proportion of land located near the Peace River, sec-

l;teport to British Columbia Hydro and Power Authority (1978).

ondary drainages, and river confluences. In the latter two

This content downloaded from 148.223.96.146 on Thu, 01 Dec 2016 04:06:18 UTC
All use subject to http://about.jstor.org/terms

182 Limitations of Traditional Surveying Techniques in a Forested EnvironmentlAlexander


cases, only the seven largest hibutaries of the Peace River

Table 5. Example of the data used in calculating

were classified as secondary drainages.

probabilities for the probabilistic survey. Each of these tables

It was impossible to make an accurate estimate of the


proportion of land covered by roads and powerlines or
near terrace edges. Consequently the proportions were

assumed to equal 50% of the total area. In reality, the


land near roads and terrace edges is less than half the

shows how the probability that secondary river drainages are


associated with cultural material was calculated for two of
the quadrats sampled in the survey. These probabilities were
then combined with others to provide the combined
probability for all quadrats studied.

total area. Since tests showed a dependence on the pres-

Near a Secondary River Drainage

ence of roads and terrace edges, these proportions provide a conservative estimate.

Cultural

Continuing with the earlier example, the expected frequency of sites near a secondary river drainage was cal-

Material

82

86

culated with the use of maps. The study area was


estimated to contain 35 km. of secondary drainages. This
figure was multiplied by 200 m. which allowed for a

100 m. corridor along both sides of each drainage. In

near a secondary drainage. The expected frequencies,


4.6 and 128.4, were attained by multiplying the total

Cultural
Material

number of sites by the proportion of the study area inside


and outside the given range. Next, the observed and
expected frequencies were presented in a series of tables

88

Near a Secondary River Drainage

total, an area of seven sq. km. of land was located near


a secondary drainage. This area was subtracted from the
total study area to give a figure of 203 sq. km. NOT

84

p (for a single quadrat) = .9107

+
-

1
7

12

16
20

p (for a single quadrat) = .4654


Combined p (for all 75 quadrats) = .7964

(TABLE 4) and the data were tested, with the G-test, for
statistical independence between each variable and the
occurrence of cultural material. Finally the exact prob-

regarding this assumption of independence has serious

ability was determined for each test value. In this ex-

consequences. Blalock notes that "Instead of having sig-

ample, G = 30.00 which, with one degree of freedom,

nificance at the 0.05 level, the true level (as obtained by

provided a probability less than .001.

correct cluster-sample formulae) may be as high as

It should be noted that some of the probabilities re-

.50."25 Moreover, there is a general lack of formulae

lating to ground cover are conditional probabilities. For

for the analytical treatment of data arising from complex

example, the observed and expected frequencies of sites

sampling designs.

with a dominant ground cover of pasture or broken land

To avoid these complications pits or subunits were

were based on the amount of cultivated farmland, NOT

used as the basic elements of investigation. With this

the total available land surface in the study area.

approach standard statistical techniques could be used

For economical and practical reasons, the archaeo-

since the analysis was of a random sample of pits within

statistical survey used a complex sampling design in

each quadrat rather than a cluster sample of sites within

which the primary sampling unit was the quadrat. If the

the total study area. Consequently, the presence or ab-

site is used as the basic element of investigation, then

sence of cultural material was noted for each excavated

the sampling units comprise clusters of individuals.

pit and related to the selected variables. When surface

Cluster samples are not uncommon in archaeology.23

reconnaissance replaced excavation it was noted whether

Unfortunately "Standard statistical literature has been

or not cultural material was located in the subunit. The

developed almost entirely in terms of independent ob-

data were then treated as though the material was found

servations obtained by simple random sampling."24 Dis-

in the pit. Using this method all 75 quadrats were tested

23. Thomas, op. cit. (in note 2) 87-100; Richard G Matson and

rat was independently examined to determine the number

William D. Lipe, "Regional Sampling: A Case Study of Cedar Mesas

of pits with cultural material within 100 m. (horizontally)

Utah," in James W. Mueller, ed., Sampling in Archaeology (Uni-

of a secondary drainage and, conversely, how many pits

individually for each variable. For example, each quad-

versity of Arizona Press: Tuscon 1975) 128-131; James M. Mueller,


The Use of Sampling in Archaeological Sarvey, SAA Mem 28 (1974)
29,66.

in the quadrat were outside the 100 m. range and contained artifacts (TABLE 5).

24. Leslie Kish, "Confidence Intervals for Clustered Samples,"


American Sociological Review 22 (1957) 154.

25. Blalock, op. cit. {in note 3) 527.

This content downloaded from 148.223.96.146 on Thu, 01 Dec 2016 04:06:18 UTC
All use subject to http://about.jstor.org/terms

Journal of Field Archaeology/Vol. 10, 1983 183


The data from the probabilistic survey generally prohibit the use of the G-test. Theoretically there were 100
pits in every quadrat; in practice, however, it was often
impractical or impossible to excavate every pit. Factors

such as steep terrain, creeks, rivers, and fields in crop


often considerably reduced the sample size. Moreover,

the frequency of pits with cultural material was usually


very low (TABLE 5).

In the case of 2 x 2 tables where the expected frequencies are small, it is possible to use an exact probability test known as Fisher's Exact Test.26 If we label
the cells and margins of a 2 x 2 table as follows,

Results

Tables 6 and 7 illustrate the final results of this comparative analysis. The basis statistical data are presented
in Table 6 while Table 7 provides a verbal interpretation
of these statistics. In the latter, .05 was used as the level
of statistical significance for the traditional survey data.
Because the Fisher's Exact Test, as used in this study,
was a one-tailed test and the interest was in a two-tailed
test similar to the G-test, the significance level was halved
to .025 for the probabilistic survey data. 28
Clearly, the traditional survey produced different results. For example, the data from the probabilistic survey
suggest that site locations are independent of bogs, secondary river drainages and remnant river confluences,

A+B

while the traditional survey indicates a dependence on


the absence of bogs and the presence of secondary drain-

C+D

ages and remnant river confluences. Similarly, sites are


associated with broken land rather than pasture, prairies

A+C B+D N

rather than forest, and deciduous rather than evergreen


forest in the traditional survey, while the probabilistic

we can obtain the probability of getting exactly these


frequencies. This probability is given by the formula:

survey implies that cultural material is independent of


these factors.

These conclusions agree with predictions made earlier


in this paper. The traditional survey is heavily dependent
upon surface exposures for the location of sites. Broken

p (a + b)! (c + d)! (a + c)! (b + d)!

fields provide almost continuous exposure. Similarly,

N! a! b! c! d!

deciduous forests have fewer exposures than evergreen


forests because of thicker humic deposits and heavier

where a, b, c, and d are cell frequencies and N is the

underbrush. Forested areas, in turn, have less exposed

population size.

ground than the prairies, especially on dunes and breaks.


These exposures greatly facilitate the discovery of ar-

To determine the exact probability for each variable


in the probabilistic survey, it was necessary to combine
the probabilities for the 75 quadrats. When a joint overall
statistical analysis is not possible, a series of separate
significance tests on different sets of data may be combined.27 The actual computation is based on the fact that
- 2 ln P is distributed as chi-square with 2 k degrees
of freedom (k = the number of separate tests and probabilities). In other words, the combined probability is
equal to - 2 E ln P; where two or more cells contained
a zero, the probability was recorded as one. The natural
log of P is, in this case, zero and has no effect on the
combined probability.

Finally these combined probabilities were interpreted


and compared with the results derived from the traditional survey data.

chaeological sites in a surface survey. Consequently the


data from the traditional survey indicate that sites are
dependent on the presence of broken land, prairies, and
evergreen forest even though the probabilistic survey
suggests that the opposite is true.

The other differences ensue from prejudicial assumptions made by the surveyors. Since the entire survey area
could not be observed, they concentrated their surveying
efforts in locations considered to be the most productive.
In this case the "good spots" were located away from
bogs and near remnant river confluences and secondary
drainages. This configuration reflects lack of concern for
past hunting-and-gathering practices which necessitated
following game into areas away from the rivers and into
marshland.

The traditional and probabilistic surveys did, in some


cases, produce similar results. Both suggest that archaeological sites are associated with the presence of culti-

26.

Ibid.

287.

vated fields, terrace edges, the Peace River (if vertical

27. Sokal and Rolhf, op. cit. (in note 17) Section 17.1; R. A. Fisher,

Statistical Methods for Research Workers, 12th rev. ed. (Oliver and
Boyd: Edinburgh 1954) Section 21.1.

28. Sokal and Rohlf, op. cit. (in note 17) 593-598.

This content downloaded from 148.223.96.146 on Thu, 01 Dec 2016 04:06:18 UTC
All use subject to http://about.jstor.org/terms

184 Limitations of Traditional Surveying Techniques in a Forested EnvironmentlAlexander


Table 6. Results of the study.

Traditional Survey Probabilistic Survey

Variable

Range*

Cultivated/Natural
Pasture/Broken
Prairie/Forest
100

Bog

Road

(H)

200

100

146.0884

Road

Terr.

Edge

(V)

200

(V)

100

Peace

R.

100

Peace

R.

Sec.

Drain.

Sec.

Drain.

(H)

200
100

Pres.

Conf.

Pres.

Conf.

(V)
(H)

+200
500
+

Remn.

Confl.

Remn.

Confl.

22.1505

(H)

200

4.9946
2.6368

(V)

500
+200

.9136

(H)

.271

5.7657

.828
.864
.864

<.001
<.001

97.3438

<.001

41.3875

.004

41.3875

.004

<.001

11.1080

.796

.026

12.4348

.712

.108

16.1257

.111

16.1257

.111

.368

5.0491

(V)

.763

79.5340

<.001

30.0086

(V)

.341

4.9319

11.5420

<.001

1.2510

13.5684

79.5340

<.001

<.001

11.5420

<.001

61.8826

2E1nP
82.4227

.006

<.001

16.9732

(H)

<.001

<.001

13.9372

20.0663

<.001

7.5650

13.9372

(H)

<.001

227.2463

Deciduous/Evergreen
Bog

38.1873

.025

5.0491

.025

4.8498

.771

4.1428

.838

(p = probability that an association between archaeological sites and any variable occurred by chance; -2E1np = combined
probability for all quadrats)

* 100 (H) = within 100 m. horizontally


+ 200 (V) = within 100 m. horizontally and 200 feet vertically

distance is ignored), and roads and powerlines. They

vey, as is reflected in the inflated probabilities produced

also suggest that sites are independent of present river

by the traditional survey data.

confluences. These similarities require explanation.

Ignoring vertical distance, both surveys suggest that

Contrary to popular opinion, the sites in this study area

sites are associated with the Peace River, probably re-

are not associated with present river confluences. This

flecting aboriginal settlement patterns. The traditional

statement is supported by the results of both surveys,

survey is not biased in favor of this association as ex-

although it was anticipated that the traditional survey

pected, except in those areas less than 100 feet above

data would suggest the opposite since the surveyors were

the Peace River. In this case the traditional survey data

working under the presupposition that sites cluster at

conflict with that of the probabilistic survey and suggest

confluences. This result is particularly surprising when

that site locations are independent of proximity to the

one notes that this survey, contrary to the probabilistic

Peace River. This lack of association again suggests an

survey, suggests that remnant river confluences are as-

avoidance of recently formed floodplains.

sociated with sites. A closer inspection of the land in the

As expected, the traditional survey suggests an asso-

vicinity of the present river confluences, however, re-

ciation between cultivated fields and archaeological sites.

veals that most of this area is floodplain of relatively

The probabilistic survey, however, also indicates an as-

recent origin. Because of their temporally short exis-

sociation between cultural material and type of cover.

tence, the surveyors expected these floodplains to con-

The difference is that, in the latter case, sites are asso-

tain few sites. In this case, therefore, the negative effect

ciated equally with both natural and cultivated fields. In

on site density created by the presence of recent flood-

other words the one-to-one correlation suggested by the

plains outweighed the importance assigned to the prox-

traditional survey does not exist and the data are there-

imity of a river confluence. In other words, the traditional

fore biased.

survey produced the correct results for the wrong rea-

Since terrace edges are frequently associated with ex-

sons. Sites .vere not discovered near present confluences

posures, the traditional survey was expected to be biased

because those areas were generally avoided during sur-

in favor of an association between sites and terrace edges.

This content downloaded from 148.223.96.146 on Thu, 01 Dec 2016 04:06:18 UTC
All use subject to http://about.jstor.org/terms

Journal of Field ArchaeologylVol. 10, 1983 185


Table 7. Interpretation of the results.

for farming and has 202 prehistoric sites.30 Almost all


roads and powerlines are also on the north side in those

TRADITIONAL SURVEY

areas used for farming and habitation. In short, there is

Sites are associated with:


Cultivated land rather than natural cover

Broken land rather than pasture


Prairie rather than forest
Evergreen rather than deciduous forest

an indirect association between sites and roads and powerlines caused by similarities in prehistoric and extant
settlement patterns.
In summary, the traditional survey is 1) biased in fa-

Absence of bog

vor of cultivated fields, broken land, prairies, evergreen

Presence of roads and powerlines

forest, remnant river confluences, and secondary river

Presence of terrace edge

drainages, and 2) biased against bogs, present conflu-

Presence of Peace River ignoring vertical distance

ences, and the Peace River below 100 m. vertical. Evi-

Presence of secondary drainage

dence also suggests that three assumptions about site

Presence of remnant river confluence

association hold true: 1) sites are associated with the Peace

Sites are independent of:

River, 2) sites are associated with terrace edges, and

Presence of Peace River if less than 100 m. above river

3) sites are associated with current settlement patterns as

Presence of present river confluence

reflected in the occurrence of roads and powerlines. Con-

PROBABILISTIC SURVEY
Cultural material is associated with:
Presence of cultivated land and natural cover
Presence of roads and powerlines

sideration of vertical distance had little effect on the outcome. Moreover, individual testing of the strata used in

the probabilistic survey failed to produce contrary results.

Presence of terrace edge

Comparison of Results with Other Research

Presence of Peace River


Cultural material is independent of:

David Brose provides the only other similar compar-

Broken fields and pasture

ison between two different surveying techniques.3l A1-

Prairie and forest

though the explanation of his methodology is unclear and

Evergreen and deciduous forest


Bogs

incomplete, it would appear that he has compared a tra-

ditional survey to a probabilistic cluster sample. Since

Secondary drainages

his study area, NE Ohio, has heavy ground cover similar

Present river confluences

to that of the Peace River region one would expect sim-

Remnant river confluences

ilar results. In fact, he concludes that "traditional survey

Level of statistical significance


= .05 for the traditional survey

methods overrepresent sites in topographic zones, such

= .025 for the probabilistic survey

as lake shores or river bluffs, where archaeologists, both


amateur and professional, tend to think of prehistoric
occupations as having been concentrated. At the same

time, floral communities presently underexploited by our


The probabilistic survey indicates that this association is

agricultural technology, such as elm-ash swamps or thin

real. In addition the probability, < .001, is no greater

mixed oak forests, have yielded too little prehistoric site

than suggested by the probabilistic survey. The associ-

data to traditional surveys."32

ation noted in the traditional survey data, therefore, does


not reflect bias, but reality.

Although these conclusions support the results of this


current research, caution must be exercised in making

As anticipated, roads and powerlines appear to be a

any generalizations. First, Brose's probabilistic survey

prerequisite for the discovery of sites in the traditional

is still a surficial reconnaissance and suffers from many

survey. They are also associated with cultural material

of the restrictions inherent in such surveys. Moreover, a

in the probabilistic survey. These results seem to reflect

critical evaluation of Brose's paper indicates that he in-

similarities in past and present land-use patterns. The

correctly used a chi-square test on a cluster sample; his

south side of the river is largely untouched at present

results, therefore, must be viewed with skepticism.

and has little evidence of prehistoric occupations; only

20 aboriginal sites have been located on the south side.29


The north side, on the other hand, is used extensively

30. Ibid.
31. Brose, op. cit. (in note 13) 3-18.

29. Spurling, op. cit. (in note 2) 19, Table 58, Sheets 1-3.

32. Ibid. 11.

This content downloaded from 148.223.96.146 on Thu, 01 Dec 2016 04:06:18 UTC
All use subject to http://about.jstor.org/terms

186 Limitations of Traditional Surveying Techniques in a Forested EnvironmentlAlexander


In fact, the traditional and statistical approaches are

Conclusions

Although the probabilistic approach to site survey may


produce more satisfactory results, it is not always cost-

effective. Hence, monetary and time constraints make it


impossible to avoid the use of traditional surveying techniques. There should be restrictions, however, on the
use of information derived from traditional surveys. "The
validity of inferences based on archaeological data are

inextricably tied to the quality of that data. If they are


biased, then without correction, all inferences based on

that data must be biased."33 There can be little doubt


that the traditional survey used in this analysis produced
biased results. Any similarities between the results of the

not at variance, but complementary. "The distinction


. is a false dichotomy the issue is one of reducing
bias, not of using one procedure instead of another."39

A combination of both methodologies, identified by


Mueller as the "Archaeo-statistical Approach",40 permits statistically valid sampling procedures while still

making use of archaeological or archaeologically related


data. With the correct methodology, archaeological surveys can do more than find sites; they can produce meaningful results.
Acknowledgments

traditional and probabilistic surveys would appear to be

I would like to extend thanks to Bruce Ball, Brian

the result of chance rather than design. In other words,

Hayden, Jack Nance, and Brian Spurling for helpful sug-

the traditional survey fails to provide a representative

gestions and criticisms of an earlier draft of this paper.

sample and is consequently a biased procedure. The implications are obvious. The nonrepresentative sample

39. Read, op. cit. (in note 33) 49.

produced by traditional sampling techniques cannot be

40. Mueller, op. cit. (in note 23) 3.

legitimately used in the analysis of areal trends in archaeology. Perhaps the greatest danger is that the idio-

syncratic biases contained in these studies will not be


eliminated and will consequently become an indelible
part of the archaeological record.34

It is not the intent of this paper to demonstrate that all


traditional surveys are worthless. Indeed, many situations simply do not require probability sampling, as both
statisticians and some archaeologists are quick to point

Diana Alexander is a graduate student in archaeology


at Simon Fraser University. Her primary interests are
the prehistory and ethnohistory of Athabaskan cultures,
hunter-gatherer subsistence and settlement patterns,
and the application of statistical and sampling
techniques in archaeology.

out.35 If the aim of a project is to find "the oldest, the


latest, the biggest, the most important or quintessential
example of our quest with minimal effort",36 then "archaeological prospecting" is probably the most efficient

surveying technique.37 Indeed, probabilistic surveys can


also fail to provide a representative sample. "Unless
proper statistical precautions are taken, any project that

does not produce a 100% inventory of all sites is likely


to produce a biased or weighted sample that does not
accurately reflect the true population of sites in the survey area."38
33. Dwight W. Read, "Regional Sampling," in Mueller, ed., op.
cit. (in note 23) 48.
34.

Ibid.

48.

35. Kent V. Flannery, The Early Mesoamerican Village (Academic


Press, Inc.: New York 1976) 132-136; Michael B. Schiffer, Alan P.
Sullivan, and Timothy C. Klinger, "The Design of Archaeological
Surveys," WA 10 (1978) 1-2.

36. Ezra B. W. Zubrow and John W. Harbaugh, "Archaeological


Prospecting: Kriging and Simulation," in Ian Hodder, ed., Simulation
Studies in Archaeology (Cambridge University Press: Cambridge 1978)
110.

37.

Ibid.

38. Mueller, op. cit. (in note 23) 4.

This content downloaded from 148.223.96.146 on Thu, 01 Dec 2016 04:06:18 UTC
All use subject to http://about.jstor.org/terms

You might also like