68 THE FLORIDA ANTHROPOLOGIST 2004 VOL. 57(1-2)
Table 3. Grouping of sediment samples based on subjective similarities in particle size curves of Figure 7.
wansitonal
Area 2, East wall, on bedrock (120-114 cm below surface)
broken (cf. Figures 13-15). Presumably, when very finely
comminuted, these particles contribute (along with the friable
carbon aggregates) to the black coating found on many of the
other particle types. It is also apparent from microscopic
examination that the charcoal fragments represent several
different wood species, which might be identified by a
paleoethnobotanist.
6) Oolitic bedrock fragments. Oolitic bedrock fragments are
weathered from the Miami Limestone bedrock underlying the
cultural deposits. They disappear in size fractions smaller
than fine sand (i.e., 39 diameter), at which point they are
almost all abraded or dissolved away, freeing their included
quartz and other mineral grains.
In addition to these six categories, miscellaneous particle
types were observed in some of the ten samples; these include
microfaunal bone fragments and nearly intact skeletal ele-
ments (notably several examples of millimeter-sized, conical
animal teeth, in one instance still embedded in a jaw frag-
ment). These should be identifiable by a vertebrate zoologist
or zooarchaeologist.
Discussion of Particle Size Analysis statistics
Appendix 2 summarizes the weight percent distribution of
the seven size fractions for each of the ten samples. The
corresponding unmodified cumulative particle size curves are
displayed in Figure 6 (note that “+5q” as plotted on this graph
is equivalent to “finer than +49”).
For non-natural, anthropogenic sediments such as these, a
fundamental analytical problem involves the coarsest size
fraction (here greater than —1q in diameter), which invariably
comprises non-sedimentary, ecofactual components such as
faunal bone, shell, as well as rock fragments and concretions.
While it is important to isolate and study this size fraction in
order to understand depositional processes, the fact that the
majority of the samples’ weight is concentrated here results in
very skewed (i.¢., unrepresentative) particle size distributions,
such as Figure 5. The size distribution curves are,
sedimentologically speaking, incompatible; they cannot be
compared with one another in terms of the sand-size fractions,
and it is the size distribution of sand in a sample that is even
more important in understanding how it was deposited.
To allow valid comparisons among samples, a further
mathematical transformation was made to the particle size
analysis statistics: each sample’s coarsest fraction (i.e., > -1@)
was subtracted from its total weight and the cumulative
weights of the remaining size fractions were recalculated to
100% and replotted. The results, presented in Figure 7, are
directly comparable and show two general groups of size
curves (again, note that “+5q” as plotted on this graph is
equivalent to “finer than +4@”). Samples FS 1030, FS 1031,
FS 1034, FS 1035, FS 1036 and FS 1037 appear to constitute
one group characterized by a lack of sediment particles in the
very coarse to coarse sand size fractions (-!9 to +19).
Samples FS 379, FS 313, FS 329, and FS 1033 constitute
another group having substantial weight in those size frac-
tions. This is interpretable in terms of where the samples were
collected, as summarized in Table 3.
Note that sample pair FS 1034 and FS 1036 are essentially
identical in their unmodified particle size distributions, as are
two other pairs: FS 1035 and FS 1030; FS 329 and FS 313.
It is not assumed that the subjective groupings of Table 2,
based on visual comparison of sediment particle size distribu-
tions among these ten samples, indicate an actual stratigraphic
correlation, i.e., that Group 2 sediments represent one
depositional event and Group | another. However, as a
working hypothesis this idea was further investigated using a
fundamentally different means of comparing the samples’ size
fractions: automatic classification of the samples into “natural”
groups on the basis of multivariate statistical analysis of the
component particle categories described previously. This
involved hierarchical linkage analysis of sample size fractions
(cases) by particle types (variables). No assumptions had to be
made about either the number or members of the clusters.
This method is described in Doran and Hodson (1975: 176) as
“single-linkage cluster analysis,” and is the only procedure