2 Sampling Strategies in Archaeological Survey
Learn how archaeological survey sampling designs connect research questions to the areas examined, the observations collected, and the conclusions those observations can support.
Connect the question to the design
Archaeological survey rarely examines every place or artifact at equal intensity. A sampling design specifies what area or material is being studied, which parts will be examined, how they will be selected, and what conclusions the observations can support.
The research question guides the design. Estimating site density across a region calls for a different sample than locating resources within a construction corridor or documenting artifact distribution at a known site.
Define units and the
Define the study boundary and units before selecting a sample. A is the spatial element selected for observation; it may be a grid cell, parcel, transect, shovel-test location, or block of land. The selected unit need not be the same as the unit recorded: a team might select a grid cell, walk transects inside it, and record artifacts or site observations at smaller scales.
The is the defined set of units from which the sample is selected, such as all grid cells within a mapped study boundary. Note excluded or inaccessible areas and the reason for each exclusion. If the frame omits particular landforms or properties, the survey cannot automatically support conclusions about them.
When archaeological materials cross unit boundaries, use consistent rules for assigning or recording them. This helps avoid omissions and double counting.
Choose a coverage strategy
Coverage strategies determine which places are examined and how. They may be used separately or combined to suit the research question.
: Examine the entire defined area at a specified intensity, often by systematic pedestrian transects. Full coverage describes where surveyors go; it does not guarantee that every buried site or artifact will be detected.
: Select units randomly from the frame. When every unit has a known, nonzero selection probability, the design can support statistical estimates for the defined population. A random sample may nevertheless cluster spatially by chance.
: Select units or walk transects at regular intervals, usually with a randomized starting point. This provides even spatial coverage and is practical in the field, but a regular pattern may interact with repeating spatial patterns in the archaeological record.
: Divide the frame into meaningful groups, such as landforms, vegetation zones, or predicted site-potential classes, and sample within each. This ensures that important but less extensive zones are represented. Sampling may be proportional to stratum area or deliberately denser in selected strata; disproportionate designs require appropriate weighting for overall estimates.
: Select groups of units first, then sample units within those groups. This can reduce travel and field costs, but observations from nearby units may be similar, affecting precision.
: Prioritize locations expected to contain archaeological materials, using existing records, landscape knowledge, or predictive models. This can be effective for discovery, but selection probabilities are generally unknown, so results should not be treated as a statistically representative sample of the whole area.
These approaches can be combined. For example, a project may stratify a region by landform, randomly select blocks within each landform, and then walk regular transects through each selected block.
Account for and bias
Selection is only one part of survey design. varies with vegetation, ground visibility, erosion, sediment cover, terrain, artifact size, surveyor experience, and field methods. A place that was walked may still have a low chance of yielding an observation.
Surface inspection and subsurface testing reveal different evidence. Surface finds alone may not establish a site's depth, integrity, or boundaries.
Bias can enter when units are chosen for convenience, inaccessible areas are silently omitted, transect spacing changes, or surveyors concentrate on promising ground without recording the change. Differences in visibility or effort between units can also introduce bias.
Record the and boundaries, unit dimensions, selection method, transect spacing, areas not surveyed, ground visibility, land cover, weather, survey intensity, crew, and any procedural changes. Document the location and basis of artifact collections and tests. These records help later researchers distinguish “no materials observed” from “not examined” or “examined under poor visibility.”
Match conclusions to the design
Inference must match the design. A probability sample with a clearly defined frame and documented selection probabilities can support estimates about that frame, subject to sampling error and imperfect . It does not by itself justify claims about omitted areas, unrecorded material classes, or the entire archaeological past.
A purposive survey can identify promising locations, but ordinarily cannot estimate regional site frequency without additional assumptions or a probability-based design. Even a full-coverage pedestrian survey may miss buried, obscured, or inconspicuous remains.
For example, suppose a study area contains a terrace of hectares and a floodplain of hectares, for a total of hectares. A stratified design could randomly select of each, surveying hectares of terrace and hectares of floodplain. This represents both landforms proportionally.
If instead the project surveys of the terrace and of the floodplain to investigate a terrace-focused question, it should report the unequal sampling fractions. It should not treat the raw combined find count as an unbiased estimate for the whole area.
Keep conclusions within the evidence
A sampling strategy connects a research question to the places examined and the claims made from the results. Define the frame and units, choose a coverage strategy suited to the goal, document selection and field conditions, and account for uneven sampling and . Conclusions should not extend beyond the population and evidence that the design can reasonably represent.