November 20, 2024

Paper presented at the Southeastern Archaeological Conference in Williamsburg, VA, November 13 - 16

This paper offers an update on our ongoing efforts to develop more reliable and reproducible methods to decipher the occupational histories of archaeological sites that may have been home to multiple groups of people over the course of several decades.  By “occupational history”, we mean an account, based on archaeological evidence, of variation and change in the houses on a site, along with a complementary account of who lived in them. Our goals include inferring the timing of the construction, modification, and destruction of domestic structures and their occupation by different groups whose members may have drawn on different cultural traditions to face different historical circumstances. We illustrate two of these methods in case study focused on an archaeological site which we call “Site 7” located on Monticello Mountain, at the western edge of the Virginia Piedmont and occupied from the 1740s to around 1790.  

Our work intersects with the theme of this session: boundaries. Inferring occupational histories requires delineating the diachronic boundaries that separate the use of domestic spaces by successive groups of residents in time. It also requires the identification of the synchronic boundaries that separated contemporary use of adjacent domestic spaces by different groups.  And it requires that these two challenges be met simultaneously. They are common in our discipline. Human intuition alone is incapable of the tasks. The help we need comes from quantitative methods.   

Here is how we plan to proceed. First, we provide an outline of the larger research initiative in which our work at Site 7 is embedded. Then we describe what we know about the site from documents, its troubled post-occupational history, and the character of its archaeological record. As we have discussed elsewhere, the adaptive cultural dynamics that helped drive the settlement of the Piedmont by Europeans, enslaved Africans, and their descendants left behind an archaeological record, like the one we find at Site 7, makes the inference of occupational histories challenging (Neiman 2022).

We describe how our methods reveal the broad outlines of Site 7’s occupational history. That history included successive occupations by at least two groups of overseers and enslaved people. The earlier group occupied the site when it was an outlying quarter associated with Peter Jefferson’s Shadwell Plantation. Their successors labored on the home farm of his son Thomas’ Monticello Plantation.

Research Context

Our engagement with occupational history of Site 7 is part of an effort to advance our understanding of patterns of change in the archaeological record and the historical dynamics responsible for them in the early Chesapeake, with a particular focus on Monticello Plantation. Ultimately we aim to clarify how three key developments affected Site 7’s free and enslaved laborers. First, the 1720’s witnessed the demographic expansion into the Piedmont of the slave society that had evolved in the Coastal Plain in the late 17th century (Graham et al. 2007, Morgan 2012: 58-101, Morgan and Nichols 1989). Within two decades, Peter Jefferson had patented thousands of acres on the western edge of the Piedmont, on the leading edge of the settlement wave. He built a house at Shadwell, two miles to the east of Site 7 (Kern, 2005, 2010).  As we will see, Site 7 was initially occupied was an outlying quarter farm during a period of low but increasing settlement density. Second, by the late 1760s, Peter had been dead for a decade and his son Thomas had begun to build a house a half mile west of Site 7, on the summit of the mountain that he had named “Monticello” (Beiswanger 2002: 9-18.  The costly architectural ambitions apparent in the neoclassical design of the house attest to the passing of the frontier and its sequalae: increased population pressure, social inequality, and status competition among local elites.  There was a third critical development. In the early 1790s Jefferson and planters across the Chesapeake responded to a spike in wheat prices on Atlantic markets in the wake of the French Revolution by transitioning from tobacco to wheat production. The shift had important consequences for plantation organization and management strategies, agricultural ecology, and of course the work and domestic lives of enslaved people (Neiman 2008, Walsh 1989, 2012). 

The foundation for our research at Monticello is the Plantation Archaeological Survey, an ongoing project that aims to discover all the archaeological sites and other traces of past human activity on that portion of Jefferson original 5000-acre tract currently owned by the Thomas Jefferson Foundation. Survey. The survey has revealed that Site 7 is one of four domestic sites – the others are Site 3, 8, and 30 -- on the eastern slopes of Monticello Mountain whose occupations date to the third and fourth quarters of the eighteenth century. The survey has also revealed 10 domestic sites eastern slopes that date to the first quarter of the nineteenth century. The different locations of these two groups of sites point to a settlement pattern shift that we believe dates to the 1790’s when the tobacco-to-wheat transition also caused changes in agricultural technology, plantation management strategies, and the domestic and work lives of enslaved people.

There are several references in the documentary record from the last quarter of the eighteenth century to an overseer’s house that was almost certainly located at Site 7 (Bon Harper 2006). The most definitive is a map and accompanying notes for a survey that Jefferson conducted in 1793 of agricultural fields east of the mansion to help implement the crop transition. The map shows a box, Jefferson’s symbol for a house, just east of a cherry tree and south of the East Road, which ran from Monticello mansion down the spine of mountain to a ford across the river to Shadwell. The notes tell us that “the Overseer’s house” lay three poles from this tree (Jefferson 1793a, 1793b. This same cherry tree turns up in the notes for another survey from 1809 where Jefferson says it is located “on the south edge of the road near the site of the former overseer’s house”, suggesting the site had been abandoned by then (Jefferson 1809).

The Archaeological Record

A first step in understanding how the archaeological record that we have discovered at Site 7 can tell us about its occupational history is to understand how that record has been shaped by the use of the site after its abandonment in the late eighteenth century. The physical traces of this past land use are revealed on high-resolution LiDAR coverage of the site (Figure 1). Among them is the modern incarnation of the East Road which slices through the northern end of the site. The LiDAR also records an earlier version of this road even further north which is likely the location of the road shown on the 1793 survey. LiDAR also reveals four or five long, linear ridges running north-south in the southern half of the site, reflecting the use of the area by Foundation staff in the mid-twentieth century to grow potatoes. This episode brought to a close over a century and a half of cultivation which we suspect began almost immediately after abandonment of the site soon after the transition to plow-based cultivation of wheat. Flat land suitable for plowing is scarce on a mountain. A 60-by-70-foot depression just to the north of the potato-patch ridges is a borrow pit excavated in the 1980s to obtain fill needed for landscape restoration on the Mountaintop in the wake of William Kelso’s Mulberry Row excavations (Kelso 1986, 1997).

Figure 1. LiDAR bare-ground model of Site 7

The shovel test pit survey in which we discovered Site 7 revealed an artifact scatter that was roughly 300 feet north-south by 150 feet east-west. To explore this scatter and understand its occupational history we followed what is now standard protocol for the Monticello Household Archaeology Project. We divided the site into 20-foot grid blocks and then randomly chose a 5-foot quadrat with each one of these blocks.  We excavated additional quadrats within each block when the first quadrat revealed a subsurface feature or an uptick in artifact density. To date we have excavated 134 quadrats across the site. In those quadrats we have found only two subsurface features that unambiguously date to the occupation of the site in the eighteenth century (Figure 2).  One of those is a 6-by-10-foot scatter of cobbles and brick bats at the northern end of the site that we believe marks a hearth or chimney base scattered by later plowing.  The other is a small borrow pit located roughly in the middle of the artifact scatter.

Figure 2. Archaeological plan, Site 7. Source: https://www.daacs.org/sites/site-7/#images 

Given this unpromising record, how to proceed?  Nearly all the information we have about the site lies in the spatial distribution of artifacts in plowzone across it.  We need to focus on spatial pattering. We describe two approaches that we have been working with over the past several years, one based on making maps of the spatial distribution of individual artifact classes, the other based on measuring patterns of similarity in the frequencies of multiple artifact types among assemblages from the 134 quadrats at the site. 

Artifact Distribution Maps

The first approach relies on mapping the spatial distribution of artifact classes that were likely to have been discarded at different times during a site’s occupation.  The goal here is to identify artifact concentrations – spatial zones of high artifact density – that are the result of repeated episodes of artifact discard in the same location.  Decades of research on cross-cultural regularities in discard behavior by ethnoarchaeologists and on spatial pattering on post-contact Chesapeake sites show that artifact concentrations are likely to be located adjacent to the houses in which the artifacts comprising them were used and broken. Artifact concentrations may therefore be useful evidence for house locations and occupation dates, especially when subsurface architectural remains are lacking.   

Our analysis of artifact concentrations we initially focus on ceramic ware types manufactured in vessel shapes associated with eating and drinking. These activities probably took place in spaces shared by members of a given household and generated refuse with similar spatial patterns of discard. The focus on ware types having the same function – the consumption of food and drink – has two analytical advantages. First, the social character of these activities makes it more likely that type frequencies will be affected by fashion trends, making make them useful chronological indicators. Second, by focusing on ware types with a similar function, we lessen the risk that we confound change over time in the location of eating and drinking with synchronic variation in in the location of different activities also mediated by ceramics, for example the bulk processing, storage, and preparation of food.   

One way to differentiate these two groups of ware types is by measuring vessel wall thickness. The idea here is that thicker-walled ceramics are more likely to be used in bulk processing, storage, and preparation of food because they better resist the mechanical shocks entailed by these activities. We summarized variation in shed thickness across the ware types found at the site (Figure 3). We assigned ware types with median sherd thickness less than Delft’s to a “thin” group that we think is more likely to be used in eating and drinking.     

Figure 3. KDEs of sherd thickness values by ceramic ware type. The vertical line is the median. Individual sherd values represented by open circles

We then dividend the “thin” ware types into an early and a late group. The goal is to measure change over time in the locations of concentrations that would indicate change in the locations of discard and perhaps the locations of houses on the site. The early group is comprised of ware types whose peak popularity in our region fell during the middle of the eighteenth century. The late group includes type whose peak popularity occurred in the last quarter of the eighteenth century.  By mapping several types within each group, we can assess the extent to which the high-density zones for each type fall in the same locations, as we would expect if our hypotheses about their use in shared space and similar discard pathways hold. 

Making Maps

Recall that the data to make our maps consist of type counts in 134 five-foot quadrats scattered across the site. Our goal is to estimate from them the continuous surface of densities from which they were sampled. To do that we use “generalized additive models” or GAMs (Wood 2017). GAMs are a generalization of regression models that can accommodate complex curved (non-linear) relationships between independent variables (the grid coordinates of the quadrats) and the variable to be predicted from them (the artifact density). GAMs can follow complex density surfaces in the data because they are built from many small simple curves glued together. The math behind the gluing process yields a predicted density surface that represent a sweet spot between a bumpy surface that would be required to fit the data perfectly but also would capture lots of random noise, and an overly smooth surface that would miss the major trends around which the actual counts fluctuate randomly. Because GAMs can explicitly model errors from predictions using probability distributions that are appropriate for count data, their results can be evaluated using a variety of statistical tools. The two we will rely on here are the amount of variation in the raw data that are presented by the predicted surface: the “proportion of deviance accounted for” in GAM lingo. Second, a statistical test of the null hypothesis that the predicted surface could have arisen from sampling a flat surface of constant densities. This offers some protection against offering substantive explanations for spatial patterns that cannot be distinguished from noise. The proportion of deviance accounted for values for the GAMs that follow range from .94 to .46. All p-values are less than .001. These numbers should inspire confidence.         

Our group of early ceramics includes Delft, White Salt Glaze, and Whieldon. We fit a GAM for the sum of their counts, as well as GAMs for the counts of the individual types. The summary map shows two major concentrations (Figure 4). One of them lies at the northern end of the site coincident with the remains of the chimney base or hearth (C1).  The second lies 150 feet to the south and contains two peaks (C2). They are separated a density hole caused by the 1980’s borrow pit. Both major concentrations and the hole appear on the three individual maps. The 150-foot distance separating the C1 and C2 implies they represent refuse deposition from two different structures.  A conservative inference is that there were at least two houses at Site 7 in the middle of the eighteenth century.

Figure 4. GAMs of sherd counts in 5-foot quadrats for three early thin-bodied ware types and their sums              

The three types in our group of late ceramics are Creamware, Chinese Porcelain, and Pearlware (Figure 5). The northern concentration (C1) evident on the maps of the early ware types does not appear on the summary maps or any of the individual maps for the late types. It has been replaced by a new concentration to the south (C3). We infer from this shift that the house associated with the hearth had been abandoned by the time Creamware appeared at the site, that is by the early 1770’s. The house that replaced it was likely the house that Jefferson mapped as an Overseer’s House.

Figure 5. GAMs of sherd counts in 5-foot quadrats for three late thin-bodied ware types and their sums.

The summary map also shows a second new concentration (C4) just south of C3. Only 50 feet separate them, leaving open the possibility that they are the result of deposition from the same structure – presumably the overseer’s house. However, a source in a second new structure adjacent to it cannot be ruled out. The summary map as well as the Creamware and Chinese Porcelain maps show relatively high artifact densities extending roughly 150 feet south of the C4 concentration, suggesting that the structure responsible for the C2 concentration from the mid-eighteenth century remained occupied. Or perhaps it had been replaced by a second structure in the same location.  The C3 concentration appears on the Pearlware map along with a high-density zone near the C4 concentration. However, the high-density zone to the south of C4 does not. Pearlware is the latest type in the group. Its near absence may represent abandonment of the southern zone or the inability of its occupants to acquire stylish and more costly ceramics.

Based on the GAMs of ceramic ware types, we tentatively conclude that there were at least two houses on the site in the middle of the eighteenth century when it was part of Peter Jefferson’s Shadwell Plantation. His 1756 probate inventory contains three groups of enumerations of agricultural tools and enslaved people that represent outlying quarter farms.  Site 7 must be one of them (Kern 2010). The two concentrations (C1 and C2) from this period probably represent an overseer’s house and housing for enslaved people respectively.  By the 1770s, the northern house, represented by the hearth foundation, had been replaced by a new structure that is probably the overseer’s house that appears in Jefferson’s surveys and other documents. Whether its construction was accompanied by new slave housing is not clear. However, occupation of the southern half of the site, presumably by enslaved laborers, continued. Site 7 was now the center of the home-farm quarter of Thomas Jefferson’s Monticello Plantation.

Correspondence Analysis

The outline of an occupational history for Site 7 has emerged from our GAM-based analysis of the spatial distribution of thin-bodied ware types. But we counsel skepticism about results based on a single kind of data and a single method. Would a different analytical approach yield similar results or offer additional illumination? To find out, we shift our focus from modeling spatial patterns in counts for single ware types to modeling spatial patterns in the proportions of multiple ware types in the ceramic assemblages from each excavated quadrat.

We start with a method that archaeologists have relied on to infer the chronological order of multiple assemblages based on their composition for over a century: frequency seriation. The model behind the method is simple: type proportions (or relative frequencies) in assemblages follow a trajectory over time of monotonic increase to a maximum, followed by monotonic decrease (lenticular or “battleship-shape” curves). Ecologists invoke a similar model when they assume species abundance follows “Gaussian response curves” as a function of some ecological gradient, for example elevation. In the archaeological case, the temporal order of the assemblages is unknown. The frequency seriation method relies on finding an order for the assemblages in which the type proportions best fit the battleship-shaped curves specified by the model (Dunnell 1970).

The frequency seriation method is the starting front for our analysis of variation in quadrat assemblages. The expectation that it might use informative results depends on a key assumption: that during the occupation of the site,; the locations into which the streams of refuse generated by different households changed over time, for example, when one house was abandonned and a new one built in a different location. In other words, in the analysis of plowzone assemblages, the seriation clock ticks when the location of artifact deposition moves. The GAM analysis of individual artifact distributions offers grounds to suspect this is true in the case of Site 7.    

Correspondence analysis (CA) is a multivariate statistical method that offers a way to solve the seriation problem (De Leeuw 2007, Greenacre 2017). Consider a data matrix of counts of a set of mutually exclusive types in a set of assemblages. CA aims to estimate the position or score for each assemblage along an underlying dimension of variation or gradient. The scores are estimated in a way that ensure the distances them along the dimension offer the best possible one-dimensional picture of the pattern of similarity among the assemblages based on their type proportions.

When we know or suspect on other grounds that the type proportions are chronologically sensitive, the underlying dimension identified by CA is often correlated with time and the assemblage scores follow the underlying temporal order.  As part of the same computation, CA estimates a score for each type on the same underlying dimension, which represent the location of the peak in its popularity curve. If the scores of the assemblages do track the underlying chronological gradient, then when we sort the assemblages on their scores and plot the type proportions, we should see the battleship-shaped curves (Smith and Neiman 2007). 

However, CA does not just estimate a single underlying gradient. To fully account for all the variation (“inertia” in CA lingo) in all the type proportions among all the assemblages requires estimating as many underlying dimensions as there are types, minus 1. The CA algorithm also estimates these additional dimensions and the assemblage scores along them in such a way that, for example, the assemblage scores on the second dimension, when plotted against the scores on the first offers the best-possible two-dimensional picture of the overall pattern of similarity, while the scores on the third dimension when plotted against the scores on the first and second offer the best three-dimensional picture, and so forth. Each dimension carries with it an estimate of “inertia”, a measure of how much of the overall pattern of similarity among the assemblages it represents or accounts for. If a single gradient does underly the overall pattern of similarity, then the first CA dimension will account for a very large proportion of the total inertia, while the successive dimensions will account for much less.  We can plot the successive inertia values and compare their pattern of fall off to what we might expect if the type frequencies followed no underlying gradient, based on the “broken stick” model (Jackson 1993). While not a formal statistical test, the comparison helps us make more reliable inferences about how many CA dimensions are required to identify the gradient(s) underlying the data, while the remainder measure statistical noise. 

Bayesian Estimation

We can increase the chances of successfully identifying underlying gradients in our data if the samples from each quadrat are large enough to allow accurate estimates of type frequencies. However, many of our samples are quite small – over half of them contain fewer than 20 sherds. In what follows we try to minimize the effects of sampling error by estimating type proportions in each quadrat based not only on we found in it but what we found in the quadrats around it. The idea here is that quadrats in the same neighborhood are likely to have similar assemblages. If that’s right, then it would make sense to “borrow strength” from neighboring quadrats in computing estimates for a given quadrat. Bayes theorem is the foundation for the math required to do this (Robertson 1996). The strength that we are borrowing can be formulated as a prior probability distribution that summarizes what we know about probable values for the type proportions based on the neighboring quadrats. We estimate the parameters of this distribution based on quadrats in the neighborhood. With data on the type proportions and sample size from a given quadrat, we use a second probability distribution to compute the likelihood of that the assemblage was derived from a larger population with a range of different proportions for the different types. By combining the prior, which in our case is modeled using a Dirichlet distribution, and the likelihood, modeled by a multinomial distribution, we can produce a posterior estimate of type proportions. We estimate the Dirichlet prior using the EM algorithm (Morgan 2024).  We use these posterior estimates in what follows.

A final methodological question is: how big should the neighborhood be? In other words, what is the distance between pairs of quadrats over which their assemblages tend to be similar. To answer it, we compute a “variogram”: a plot that summarizes how the distances between pairs of assemblages based on their type proportions increase as the geographical distances between those same pairs of quadrats increase (Lloyd and Atkinson 2020). This plot reveals that at Site 7 inter-assemblage distance values level off as the geographical distance between quadrat pairs approaches 50 feet (Figure 6). At Site 7 the neighborhoods from which we borrow strength have a radius of 50 feet. 

Figure 6. Variogram for proportions of thin-bodied ware types in quadrat assemblages. The vertical axis is the chi-square distance between assemblages in pairs of quadrats. This is the distance measure used by correspondence analysis. The horizontal axis is the geographical distance between pairs of quadrats.

CA of Thin-Bodied Ware Types

We begin our analysis of type proportions by focusing on the same thin-bodies type that we used for our GAM maps, augmented by a few thin-bodied types that were not represented by large enough samples to make statistically reliable maps, which might contribute to this analysis (Staffordshire Mottle Glaze, Jackfield, and Nottingham). We plotted of the proportions of inertia (variation) in the posterior estimates of type proportions that are accounted for by the successive dimensions identified by CA, along with the proportions expected under the broken stick model (Figure 7). 

Figure 7. Fall off in the proportion of inertia (variation among assemblages in type proportions) accounted for by successive CA dimensions. The dashed line represents expectations under the "broken stick" model.

The first-dimension accounts of 60% of the inertia in the type proportions far more than expected.  WE see a dramatic drop for the second dimension to only 11%, far less than expected. Succeeding dimensions account for even less. They probably represent statistical noise.  We conclude that the first CA dimension and the scores of the assemblages and types on it offer an accurate picture of the gradient that underlies variation in ware-type proportions among assemblages (Figure 8).

Figure 8. Plots of quadrat assemblage scores (left) and ware type scores (right) in the first two CA dimensions. The analysis only considers thin-bodied ware types.

Is the gradient chronological?  Recall that the score of each type estimates the location of its maximum proportion along the underlying dimension. This leads to the expectation that if the first dimension does capture time, the early thin-bodied type scores should fall at one end of it and the later thin-bodied type scores should fall at the other. This is exactly the pattern we see. It implies that assemblages with negative dimension-1 scores are early and assemblages with positive dimension-1 scores are late – the zero value on the scale marks the location of a hypothetical average assemblage.

The plot of the assemblage scores on the first two dimensions suggest that the assemblages might fall into three clusters, based on their dimension-1 scores. In contrast there is no evidence for structure in the dimension-2 scores, as we might expect if the variation on this dimension is noise. If the Dimension-1 clusters are real, what might they represent? They are comprised of assemblages that were, on average, deposited during the same periods of time. The gaps between them may be the result of deposition ending at one location at the site or beginning at another.

To measure these pulses of deposition more accurately, we computed a histogram in which the locations of the bars on the x axis are the dimension-1 scores for each assemblage and the heights of the bars represent the number of artifacts in in that assemblage. The bar heights jump around (Figure 9). We can smooth away some of this noise with the help of kernel density estimation (KDE), a technique that replaces the discrete bars of the histogram with a smooth curve (Baxter et al. 1997).  The KDE indicates two peaks at opposite ends of dimension 1. The early peak, on the left, is separated from the rest of the assemblages by a gap. We then see a decelerating increase in the size of assemblages, which ends with a rapid increase that forms a late peak. Based on changes in slope of the KDE, we divided the assemblages into four phases. We then mapped the locations of the phases on the site (Figure 9).

Figure 9. Left: Histogram of assemblage scores on CA dimension 1, weighted by assemblage size. Vertical dashed lines are phase boundaries, based on change in slope of the KDE for the assemblage scores. Right: Phase assignments for quadrats.

The spatial distribution of phases supports out hypotheses about the site’s occupational history based on the ware-type distribution maps. Phase 4 matches the artifact concentrations (C3 and C4) that we suggested mark the location discard from the overseer’s house that Jefferson mapped. Phase1, at the northern end of the site, matches the location of the cobble hearth that probably heated its predecessor.  The vast majority of the Phase-2 and Phase-3 quadrats lie in the southern half of the site that we suggested was the zone occupies by enslaved laborers. Their intermediate dimension-1 scores are what we would expect if artifact discard into these areas spanned the successive periods of occupation of the two overseer’s houses. However, other processes may be at work. For example, the line of Phase-2 quadrats that separate Phase-1 from Phase-4 quadrats in the northern half of the site is probably the result of mixing artifacts from two distinct periods of deposition.

Time Averaging

How might we evaluate this hypothesis? We need to devise a measure that will scale with the amount of time over which an assemblage accumulated. In other words, we need a measure of “time averaging”: how dispersed in time were the original discard events responsible for the artifacts that the assemblage contains (Lucas 2012:106-109, Perreualt 2019:25, 61-62). The method we have devised relies on CA.  While we do not know when individual artifact in an assemblage was discarded, in theory the CA type scores are an estimate of the expected time artifact of that type were discarded, under the assumption that trajectory of change in the type’s proportions is symmetrical around the location of it maximum proportion. In other words, CA type scores scale with the average discard date. Assemblages that are more time averaged should have more types whose scores are further away from their assemblage scores than less time-averaged assemblage.  To summarize this dispersion, we compute a weighted average of the squared deviations of the type scores from the assemblage scores, where the weights are the type proportions. We call this measure the interassemblage variance and use its square root, the interassemblage standard deviation, as our measure of time averaging.

A simple way to evaluate our hypothesis that the Phase 2 and 3 assemblages are more time averaged that the Phase 1 and 4 assemblages is to plot the interassemblage standard deviation estimates against the assemblage scores (Figure 10). The standard deviations are a quadratic function of the assemblage scores, with the Phase 2 and 3 assemblages having the highest values. Our expectations are met.

Figure 10. Plot of assemblage standard deviations, an estimate of time averaging, against assemblage CA dimension-1 scores.

Discussion

While our focus here has been Site 7, our results have more general implications for our attempts to write occupational histories using archaeological data from plowzone contexts Understanding spatial patterning in plowzone data depends on realizing that the data are the result of events played out over time. In the absence of multiple stratified deposits, it is tempting to regard spatial patterns as a synchronic snapshot of a moment in the past. We hope to have shown that the spatial patterns at Site 7 are the outcome of change over time in the location of artifact discard by different residential groups. The analysis of spatial data depends on chronological inference.

When a site has been occupied by multiple groups, the temporal boundaries we are able to draw for each group may not align. At Site 7, we can identify quadrats that contain comparably time-averaged assemblages generated by overseers (Phases 1 and 4). However, the quadrat assemblages generated by enslaved laborers  (Phases 2 and 3) are much more time averaged. They probably accumulated over the course of both phases of overseer occupation. We can identify a boundary between the assemblages generated by overseers when Site 7 was a Shadwell quarter and when it was the center of the Monticello home farm. But we cannot identify the analogous boundary of assemblages generate by enslaved laborers because there was little change in the locations they discarded artifacts.  

This finding has important implications for future research on the artifact assemblages recovered at the site. The four phases are the basis for the aggregating the 134 quadrat assemblages into counting units with the larger sample sizes required for further analysis, both within the site and with similarly defined units form other sites. Topics that we aim to explore include change and synchronic among enslaved and free households in participation in local markets and the larger consumer economy.

The occupational history of Site 7 makes possible those comparisons, but also constrains them. Comparing the assemblage generated by enslaved laborers to either of the overseer assemblages individually will confound change and inequality. The more informative comparison requires lumping the two overseer assemblages together. Similar issues must be addressed in the analysis of Site 7 assemblages and those from other Monticello sites whose occupations, like Site 7’s, seem to have ended around 1790, but began around 1770 not 1750.  Our ability to infer occupational histories from archaeological data is an essential foundation for future insights into patterns of change in the daily lives of a site’s residents and the long-term economic and social dynamics responsible for them.                                                

References   

Baxter, Michael J., Christian C. Beardah, and Richard VS Wright. 1997 Some archaeological applications of kernel density estimates. Journal of Archaeological Science 24, no. 4: 347-354.

Beiswangwer, William L. 2002. Monticello in Measure Drawings. Thomas Jefferson Foundation, Charlottesville VA.

Bon Harper, Sara 2006. Site 7: Background. https://www.daacs.org/sites/site-7/#background

De Leeuw, Jan 2007, Correspondence Analysis of Archeological Abundance Matrices. UCLA: Department of Statistics, UCLA. Retrieved from https://escholarship.org/uc/item/3m18x7qp

Dunnell, Robert C. 1970. Seriation method and its evaluation. American Antiquity 35 (3):305-319.

Graham, Willie, Carter L. Hudgins, Carl R. Lounsbury, Fraser D. Neiman, and James P. Whittenburg 2007. Adaptation and innovation: Archaeological and architectural perspectives on the seventeenth-century Chesapeake. The William and Mary Quarterly 64 (3): 451-522.

Greenacre, Michael. 2017 Correspondence analysis in practice. Chapman and Hall/CRC.

Jackson, Donald A. 1993. Stopping rules in principal components analysis: a comparison of heuristical and statistical approaches. Ecology 74, 2204–2214.

Jefferson, Thomas 1793a.  Monticello: east roads and fields (plat), 15 October 1793, by Thomas Jefferson. N194; K167b [electronic edition]. Thomas Jefferson Papers: An Electronic Archive. Boston, Mass. : Massachusetts Historical Society, 2003. http://www.thomasjeffersonpapers.org/

Jefferson, Thomas 1793b.  Monticello: survey notes, page 4 of 20, 1793, by Thomas Jefferson. N196; K167d [electronic edition]. Thomas Jefferson Papers: An Electronic Archive. Boston, Mass.: Massachusetts Historical Society, 2003. http://www.thomasjeffersonpapers.org/

Jefferson, Thomas 1809. Monticello: survey notes, page 4 of 4, 1809, by Thomas Jefferson. N213; K168b [electronic edition]. Thomas Jefferson Papers: An Electronic Archive. Boston, Mass.: Massachusetts Historical Society, 2003. http://www.thomasjeffersonpapers.org/

Kelso, William M. 1986. Mulberry Row: Slave Life at Thomas Jefferson's Monticello. Archaeology 39(5): 28-35.

Kelso, William M. 1997. Archaeology at Monticello: Artifacts of Everyday Life in the Plantation Community. Monticello Monograph Series. Thomas Jefferson Foundation, Charlottesville.

Kern, Susan 2010.  The Jeffersons at Shadwell. Yale University Press, New Haven.

Kern, Susan 2005. The Material World of the Jeffersons at Shadwell. The William and Mary Quarterly, 62(2): 213-242.

Lloyd, Christopher D., and Peter M. Atkinson. 2020.  "Geostatistics and spatial structure in archaeology." In Archaeological Spatial Analysis, edited by Mark Gillings, Piraye Hacıgüzeller, and Gary Lock, pp. 93-117. Routledge,  London.

Lucas, Gavin 2012 Understanding the archaeological record. Cambridge University Press.

Morgan, Martin 2024. DirichletMultinomial: Dirichlet-Multinomial Mixture Model Machine Learning for Microbiome Data. R package version 1.48.0, https://mtmorgan.github.io/DirichletMultinomial/

Morgan, Philip D. 2012. Slave counterpoint: Black culture in the eighteenth-century Chesapeake and Lowcountry. UNC Press. Chappel Hill.

Morgan, Philip D., and Michael L. Nicholls 1989. Slaves in Piedmont Virginia, 1720-1790. The William and Mary Quarterly. 46 (2): 211-251

Neiman, Fraser D.  2008. "The lost world of Monticello: an evolutionary perspective." Journal of Anthropological Research 64, no. 2: 161-193.

Neiman, Fraser D. 2020. The Organization of Enslaved Households at Site 6: Clues from Monticello and… Australia? Monticello Blog. https://www.monticello.org/exhibits-events/blog/the-organization-of-enslaved-households-at-site-6-clues-from-monticello-and-australia/

Neiman, Fraser D. 2022. Deciphering the Occupational History of an Eighteenth-century Domestic Site at Monticello from Plowzone Evidence. DAACS Conversations, https://www.daacs.org/research/playlists/daacs-conversations/

Nenadic Oleg, Greenacre Micheal 2007. “Correspondence Analysis in R, with two- and three-dimensional graphics: The ca package.” Journal of Statistical Software20(3), 1-13. http://www.jstatsoft.org.

Perreault, Charles. 2020.  The quality of the archaeological record. University of Chicago Press,

Robertson, Ian G. 1999 Spatial and Multivariate Analysis, Random Sampling Error, and Analytical Noise: Empirical Bayesian Methods at Teotihuacan, Mexico. American Antiquity 64:137–152

Walsh, Lorena S. 1989 Plantation Management in the Chesapeake, 1620–1820. The Journal of Economic History. 49(2):393-406.

Walsh, Lorena S. 2012.  Motives of Honor, Pleasure, and Profit: Plantation Management in the Colonial Chesapeake, 1607-1763. UNC Press, Chappel Hill.

Wood, Simon N. 2017. Generalized additive models: an introduction with R, Second Edition. Chapman and Hall/CRC.