Reassessing the Chronological Boundaries of Monticello’s Mulberry Row, Charlottesville, Virginia

November 20, 2024

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

This project stemmed from an interest in better understanding households within and between sites at Monticello. At a symposium on “Communities, Coalescence, and the Chesapeake,” colleagues of ours looked at how we can do this at plow zone sites (O'Connor and Neiman 2024). With this paper, we aim to examine a similar question on stratified sites, using one legacy project – Building o – as the pilot run. We will provide a brief overview of this site and the subsequent analyses since its excavation. To explore that overarching goal of better defining households, we re-examine the methods used to identify chronological boundaries and affiliated households. This paper is the first step in a longer process, for now using ceramic ware type and other ceramic attributes to investigate how we can distinguish between occupation periods at stratified legacy sites like Building o.   

Site Background

Monticello, located just east of Charlottesville, Virginia, was once home to Thomas Jefferson, his family, and hundreds of free and enslaved individuals. From 1770, when Jefferson moved to Monticello, until his death in 1826, he divided the core of the five-thousand-acre plantation into four quarter farms, with Monticello Mountain comprising the home farm quarter and satellite quarters of Tufton, Lego and Shadwell (Figure 1).

Figure 1: Map of the quarter farms a part of Thomas Jefferson's 5000-acre plantation. Map by author.

This paper focuses on one site called Building o, located on the Monticello home farm, situated along Mulberry Row, a thousand-foot-long road just south of the mansion house. Mulberry Row was once a bustling street of domestic and industrial activity where enslaved and free workers lived and labored (Figure 2). Building o gets its name from a document that guided much of the Monticello Department of Archaeology’s excavations throughout the 1980s (Figure 3, Jefferson 1796).

Figure 2: 3-D Rendering of Mulberry Row

Figure 3: 1796 Mutual Assurance Plat. Left: Original created by Thomas Jefferson. Right: Transcription with highlighted portions related to Building o.

Jefferson drew the plat in 1796 as part of an insurance policy he took out with the Mutual Assurance Company of Richmond. The policy application required a description of not only each insured building but also all uninsured buildings in proximity. The required information included the buildings’ functions, materials used in construction, and distance between structures.

Jefferson’s interpretation of these instructions produced the meticulous map of where his mansion stood in relation to all the buildings along Mulberry Row. Jefferson differentiates the insured buildings from the uninsured buildings by using a capital letter whereas the uninsured buildings are lowercase.

Modern researchers refer to the different segments of Mulberry Row using the information in this document, such as the letter designations – for example “Building o” – and functional names – like “a servant’s house”.        

Building o Excavations, 1979 and 1981-1982

First excavations at Building o began in 1979 when William Kelso opened a line of excavation units bordering its southeastern edge, in order to track the postholes of Jefferson’s 1809 garden paling (Figure 4). Between 1981 and 1982, Kelso extended the excavation to the northwest, opening forty-two units between the steep embankment above the vegetable garden and Mulberry Row, an area of 24 by 58 feet.

Figure 4: 1979 Excavations of the Paling Fence at the Building o Site

Figure 5: 1981-1982 Excavations of Building o (facing East)

Excavators initially laid out a grid of 10-by-10 foot units with 2-foot balks. As work proceeded, removal of the balks and extension of the site resulted in the excavation of quadrats varying in size from 2 by 2 feet to 8 by 8 feet (Figure 5).

Most of the site was relatively undisturbed, with exceptions along the north edge from Mulberry Row road work and extensive displacement of sediment caused by the root growth of a Kentucky coffee tree. In undisturbed areas, excavators identified three main stratigraphic layers (Figure 6). The uppermost consisted of a topsoil layer. The middle layer was an orange clay with decomposing stone fragments called greenstone, which is the parent material for Monticello Mountain. The orange clay with greenstone was found in every excavation unit except those that had modern intrusions, and it sealed an artifact-rich reddish brown clay loam identified by excavators as an “Occupation Zone” that was deposited while people were living at Building o. Within the Occupation Zone, excavators exposed the stone foundation for Building o. The Occupation Zone also extended outside of the foundations and also covered most of the stones.

Figure 6: Excavations of Building o, showing the profile and three predominant stratigraphic layers.

This layer also covered multiple features within the foundation, including a brick pad in the northwest corner and two subfloor pits – one large pit with unmortared rock walls and a smaller pit with unmortared brick floor and walls (Figure 7). Excavators were unable to distinguish between the sediment filling either subfloor pit. However, after lumping the first four inches of the large subfloor pit fill with the surrounding Occupation Layer, the decision was made to dig the rest of the stone-lined pit in four-inch increments.

Figure 7: Aerial view of Building o with features noted

Within the large subfloor pit, most pieces of a British brown stoneware jug were recovered (Figure 8). Another notable find was a cache of porcelain plate fragments (Figure 9). Found within the Occupation Zone just above a section of stone foundation, the eight partially complete plates ranged from 20 to 80 percent complete and most had the same handpainted pattern.

Figure 8: Nearly complete British Brown Stoneware storage vessel

Figure 9: Cache of nearly complete handpainted blue Chinese-export porcelain plates. Left image shows the cache after recovery and mending, placed back in the original location.

Excavations were completed in July of 1982 and subsequent analysis by Kelso determined that only one building event occurred at the Building o site, captured by Jefferson’s 1796 description (Hill 2003; Kelso 1982). Kelso calculated a mean ceramic date (or an averaged date weighted by the frequency of recovered wares and their manufacturing midpoints) of the Occupation Zone and estimated that enslaved house servants occupied the site between 1770 and 1800. He estimated that the orange clay layer found above the foundation dated to around 1810.

Scott Shumate, who joined the Monticello field crew in the mid-1980s after the excavation of Building o, touched on the site in his master’s thesis in 1992 (Shumate 1992). He observed that the configuration of wall fragments suggested that more than one building episode could be represented. He urged that additional stratigraphic and artifactual analysis be undertaken; reliance on a single document-the Mutual Assurance Declaration-tended to prejudice interpretation and obscure the rich depositional history.

2003 DAACS Analysis of Building o

In the early 2000s, DAACS – the Digital Archaeological Archive of Comparative Slavery – reanalyzed the Building o excavation and re-catalogued the artifacts. The analysis agreed with Kelso’s belief that the orange clay layer on top of the foundations originated off-site, arguing it was deposited after the Building o inhabitants left, with the purpose to bury and level the area (Hill 2002, 2003). The analysis also agreed with Shumate that the remaining foundation walls represented two structures, one overlaying the other, with the later building destroying most evidence of the first. The site plan gives some insight into this interpretation (Figure 10). Looking at the southern alignment of rocks: if the entire length of foundation wall is part of the building described by Jefferson, we see that it is not parallel with the grid – the western end is about two feet further south than the eastern end. Usually this would not elicit much comment. However, the grid was purposefully chosen so that it paralleled Mulberry Row, the thousand-foot-long street that ran just north of the site. Of the dozen sites dug along Mulberry Row that had above-ground structural remains or archaeological evidence of buildings, Building o would be the only structure that was not aligned with the road.

Figure 10: Site Map of Building o. https://www.daacs.org/sites/building-o/#home

We see just how far out of alignment the building would be when the excavators strung out the 20.5 x 12-foot dimensions of the building stated by Jefferson and overlying the archaeological features (Figure 11). The photograph is framed so that the image is parallel to the grid and Mulberry Row. The excavators understood this discrepancy and all remaining final site photos do not include the building outlined in string.

Part of the DAACS reanalysis of the Building o artifacts included producing a Harris Matrix for the site as well as phasing all excavated contexts or groups of related contexts, such as layers within a feature (Figure 12). We refer to these related contexts as stratigraphic groups. The phasing used datable ceramics for the basis of the analysis, and we see that when mapping the phases on the Harris Matrix we get a fairly good result. Contexts have been color-coded by phase, with 1 being the earliest and 5 being the latest. We see that the contexts or stratigraphic groups for the earlier phases are found at the bottom of the Harris Matrix and the later phases towards the top of the graph.

Figure 11: Aerial view of Building o with the at-the-time perceived layout of the original structure marked in string

Figure 12: Harris Matrix showing the stratigraphic relationships of the Building o contexts and stratigraphic groups. The portion highlighted in red reflects the 1979 fence line units.

The main exception is on the left side of the Harris Matrix. In this section, we see that the phases are mixed up. These contexts all belong to the 1979 fence line excavation. These excavation units were only four feet wide and while good at efficiently locating the post holes and molds for the fence line, they were inadequate in width to allow for easy identification of deposits on the complex Building o archaeological site. Therefore, it appears that excavators, unable to sufficiently delineate discrete deposits, inadvertently mixed early and later deposits to get an assemblage of artifacts that were time-averaged and therefore dated to the middle Phase for the site. Also, in some instances, excavators did not see later intrusions and were excavated after removing earlier deposits which pushed them down towards the bottom of the Harris Matrix.

With an understanding of the excavation and interpretative history of the Building o site, why are we re-examining it again 40 years after it was dug?

The answer revolves around the fact that many interpretations for the site relied on single lines of evidence. The argument for two structures relies only on the surviving foundation walls. The belief that the orange clay layer overlying the foundation and occupation layer originated off-site is based solely on the color of the sediment. The remainder of this paper will use a second line of evidence – the recovered artifacts – to see if they help bolster or weaken these site interpretations. We therefore approach this reanalysis with three preliminary questions:

1.        The chronological phasing done in 2003 used a combination of ceramic ware type, decorative technique, and decorative color. Other studies at Monticello has shown decorative genre to not always be a product of time but rather stylistic preference. Therefore, recent projects have only used ware type to build chronologies at sites – and we apply that same method here to see how that impacts our understanding of the Building o phases.

2.        Additionally, we test the role of the orange clay layer as a kind of “cap” on the stratigraphic groups exclusively associated with Building o and whether we can determine its origin.

3.        Finally, we examine whether there is any evidence of distinct households within the ceramic assemblage, including a possible structure pre-dating the Building o structure noted in the 1796 Mutual Assurance plat.

What do new phasing methods reveal about Building o’s timeline?

The chronological phases of Monticello sites are determined through a series of statistical methods developed by the DAACS and Monticello teams. You can see the detailed workflow in the 2020 chapter published by Bates, Galle, and Neiman (Bates et al. 2020). In short, building these site chronologies relies on a frequency seriation model of ware types, adjusted mean ceramic dates, and correspondence analyses to identify context clusters. We will briefly take you through the steps for this – again, only using ware types with known manufacturing dates – and what it tells us about Building o’s chronology.

For this study, we included all contexts except those that were unprovenienced from Building o. We excluded samples if they contained a count of 5 or less of ware types with known manufacturing dates. Additionally, we removed one outlier context that had an unusual small cluster of Bone China – it did not change the overall positions of contexts in the correspondence analyses, but rather made it challenging for us see those patterns. So, we removed it from this study.

After filtering and organizing our data, we calculated the values necessary for running correspondence analyses. The scree plot shows the proportion of variation (or inertia) for each dimension of change that impacts our data (Figure 13). The dashed line represents a calculation of our expectation of the proportion of inertia under each dimension – in other words, any points above this dashed line indicate dimensions that significantly impact variation in our data. In our scree plot here, the first dimension is our predominant and significant cause of change. In archaeological data, Dimension 1 usually represents time. We will also look at the second dimension to see how that may impact our data.

Figure 13: Scree Plot of Building o contexts and ceramic ware types

A correspondence analysis calculates dimension scores for both our rows and columns – in other words, our contexts and ware types (Figure 14). These two plots show the Dimension 1 scores plotted against the Dimension 2 scores for contexts (on the left) and then ware types (on the right). The red dashed cross-hairs positioned at the 0,0 coordinates visually represent the average for the assemblage. We can examine which ware types significantly impact contexts based on their spatial association. For example, on the left, we circled two contexts that contain that cache of nearly whole Chinese-export porcelain plates. You will notice that the location of those two contexts in the left-hand plot corresponds to the location of the Chinese-export porcelain ware type (abbreviated as CEP) on the right-hand plot. This initial comparison gives us an idea of the average assemblages as well as identifying those unusual assemblages (in this case, they appear to be impacted by later ware types like Whiteware and Ironstone). Additionally, looking at the ordering of the ware types, Dimension 1 is clearly showing change in time – with older ware types on the right and younger ware types on the left. Dimension 2 does not appear to be time, given that Delft (with a manufacturing range of circa 1600 – 1800) is vertically between both Ironstone (post-1840) and Yellow Ware (post-1830).

Figure 14: Correspondence analyses plotting the first two dimensions of variability for (left) contexts and (right) ware types.

To confirm that our Dimension 1 in fact represents time, we next calculate mean ceramic dates and terminus post quem values for each context. We additionally calculate a BLUE MCD, which essentially adjusts the average by more heavily weighting ware types that have narrower date ranges. The original chronological assessment in the early 2000s did not use BLUE MCDs, only a standard MCD. Finally, TPQ values, based on the beginning manufacturing date of the latest occurring ceramic, provide the earliest possible date of deposit for an entire assemblage. Because accidental inclusion of later ceramics (whether they fell from the profile wall or excavation errors) will significantly skew a TPQ, we also calculate the 95^th and 90^th percentiles to hopefully remove those outliers.

Figure 15 now shows the Dimension 1 scores against the Blue MCD calculated for each context. This relationship is particularly useful in that it shows a correlation between two independent variables that both represent time. Remember, time moves from right to left. We labeled the outlier points with both their context IDs and their assigned stratigraphic groups, which help us look at contexts across the site with similar characteristics. We will address these in greater detail below.

Figure 15: Plot showing Dimension 1 scores by BLUE MCD values for each context. Only the outlier contexts have been labeled.

Figure 16 also shows what the distribution of these contexts looks like when we plot Dimension 2 by the Blue MCD. You’ll notice that we lose that clear linear regression, which makes sense given Dimension 2 did not appear to be time. Of note here are two of those contexts associated with the Chinese porcelain plate cache mentioned earlier.

Figure 16: Plot showing Dimension 2 scores by BLUE MCD values for each context. Only the outlier contexts have been labeled.

Given our interested in determining chronological phases across these contexts, we focus on Dimension 1. The histogram in Figure 17 plots the number of occurrences of each Dimension 1 score across Building o – in short, this and particularly the trend line theoretically shows us the peaks of different chronological phases. You can see here five phases that we identified – and it is worth saying that we went through a series of splitting and lumping to determine the most reasonable groupings.

Figure 17: Histogram showing density of Dimension 1 scores for each context. Phases are distinguished by the red dashed line.

From here, we assign contexts to specific phases based on where their Dimension 1 scores fall. Figure 18b shows you the MCD and TPQ calculations for each phase.  Figure 19 again plots Dimension 1 scores by the BLUE MCD for each context, but now color-coded by these chronological phases – the dashed lines also represent those phase boundaries.

Because of our interest in what would happen if we removed the possible bias of decorative preference from this method (and perhaps any differences in using adjusted MCDs and TPQs), Figure 18 shows a brief comparison in the new versus old results. We essentially discover that they are relatively comparable. One thing to note is the smaller sample size in the newer results – this likely results from us implementing more filters than the analysts in the early 2000s. And these differences really only seem to have adjusted the overall span of time from Phase 1 to Phase 5.

Figure 18: Phase dates from a) the original DAACS analysis and b) the new analysis with just ware type

Figure 19: Plot showing Dimension 1 scores by BLUE MCD values for each context, color-coded by phase assignments. 

Although not the most dramatic outcome, this has generally answered our initial question of whether decoration skewed our initial chronological phasing of Building o. It seems that it has not on this larger site-wide scale. Undertaking these steps has also allowed us to re-examine those stratigraphic groups mentioned previously. And we particularly want to highlight those that are associated with the Building o occupations (related to the 1796 Mutual Assurance plat and the earlier structure identified via the foundations).

What is the temporal role of the clay layer?

Undertaking these steps has also allowed us to re-examine those specific layers we mentioned earlier. Most of the contexts associated with Building o are in phases 1 through 3. The clay layer mostly straddles Phases 2 and 3. Again, we want to determine whether we can corroborate the temporal role of this layer and whether it was brought from elsewhere. Therefore, we focus on five stratigraphic groups (SGs):

SG 03: Shallow Feature west of Building o

SG 07: “Occupation Zone”

SG10: Large subfloor pit

SG11: Mixture of subfloor pit fill and the “Occupation Zone”

SG 13: Orange clay layer with greenstone

To assess the role of the orange clay, we ran the first half of that chronological phasing method again, but just on those stratigraphic groups we listed (Figure 20). Dimension 1 is still time. Notably, there is a cluster of the clay layer contexts that do not overlap with any other Building o-period contexts. They are distinct enough to be classified in a different phase than the other Building o contexts. In other words, perhaps the deposits in the clay layer are not all coming from the same place or at least the same time period.

Figure 20: Plot showing Dimension 1 scores by BLUE MCD values for each context in the five stratigraphic groups dating to the Building o period. The blue circle highlights the clay layer contexts that are not contemporaneous to Building o.

So, we removed just that subset of the clay layer and ran the analysis again (Figure 21). Dimension 1 is still time, and the group of clay layer contexts we left appears alongside several other contexts from the Occupation Zone and Subfloor Pit. So, these contexts are no longer distinct from Building o in terms of time. At the very least, we have answered the question and confirmed that the clay represents this transitional layer between Building o occupation and later stratigraphic groups.

Figure 21: Plot showing Dimension 1 scores by BLUE MCD values for only the Building-o period stratigraphic groups (with the later clay layer contexts removed).

Is there evidence that the clay layer was transported from elsewhere?

In order to assess that question, we focus on these same four stratigraphic groups (with the partial clay layer) and investigate whether other ceramic attributes signify different behaviors. The first of these attributes is Decorative Genre. In DAACS, this refers to commonly identified and temporally significant decorations – they typically combine decorative technique, color, and stylistic elements, and are usually affiliated with certain ware types. We were hoping to ascertain any differences in decorative preferences as well as time period.

Figure 22: Correspondence analyses plotting the first two dimensions of variability for (left) contexts color-coded by stratigraphic group and (right) decorative genres.

Figure 22 shows a correspondence analysis of the variability in contexts based on their makeup of different decorative genres. Note that nearly all of the clay layer contexts pulled out to the right, distinct by their association with Polychrome Warm and Transferprinted wares. It is possible that Dimension 1 could be time, although it is impacting the clay layer contexts more significantly than simply ware type did in the chronological phasing. In fact, Figure 23 shows that same correspondence analysis of contexts but color-coded by the original phasing – we noticed that the contexts are relatively evenly distributed by time. It is possible, therefore, these clay layer contexts on the right are contemporaneous but otherwise different from the Building o contexts, supporting the hypothesis that the clay layer (even contemporaneous deposits) originated elsewhere.

Figure 23: The same plot in Figure 22 showing the correspondence for contexts by decorative genre, now color-coded by their original phase assignments.

We examined several other ceramic attributes, including Form and Material, but they did not reveal interesting results. Therefore, we will continue to the final ceramic attribute we analyzed: Sherd Size. A maximum sherd measurement was taken for each sherd, grouped in 5mm increments. Because we batch undiagnostic sherds at 15mm, we filtered any examples that were less than this. With this analysis, we ask whether there are site formation differences that are evident within artifact size and therefore could have implications for distinct occupations.

Figure 24 plots the recovery of different sherd sizes across stratigraphic groups. The x-axis is showing the log of sherd size in order to better see differences in each group’s distributions – otherwise, it would have an extreme left skew because of course we recover more tiny fragments than large fragments. The clay layer is most distinct by its lower median sherd size – this could be more evidence of the clay layer deposits (including contemporaneous contexts) being brought from elsewhere. On the other end of the spectrum, the subfloor pit has notable occurrences of larger sherds. This makes sense, given the nature of how features like subfloor pit fills are formed – we might expect there to be some preservation of larger fragments.

Figure 24: Ridgeline plot showing the distribution of sherd size across the five stratigraphic groups of interest to the Building o period (excluding later contexts from the clay layer).

What evidence is there for a household pre-dating the structure noted in 1796?                                                  Regarding our aims to find evidence of a pre-1790 assemblage, the most likely contexts are from an early shallow feature (SG 03). Figure 20 and 21 consistently show the two contexts making up this feature as significantly earlier than the other Building o contexts (BLUE MCDs calculated as 1756 and 1788, respectively). Maybe we are seeing a small deposit of trash from those pre-1796 occupants, however neither context has a large frequency of artifacts. Given the small size of this assemblage and the general outlier nature of this feature, that is a hard conclusion to make at this time.

Summary         

1. We happily determined that the use of decoration in previous chronological phasing does not seem to have significantly skewed our overall understanding of Building o’s timeline.

2. Using the ceramic assemblage and this phasing method, we can see evidence for the clay layer to be a temporal mixture, with some contexts dating to the Building o-period and some post-dating it. The results of examining decorative genre and sherd size suggest that even the contemporaneous contexts are different than Building o in how they were formed – this supports initial hypotheses that the layer was brought from elsewhere.

3. Finally, the ceramic assemblage did not reveal any clear clustering of contexts that could represent a structure pre-dating the one mentioned in the Mutual Assurance plat in 1796, with the exception of a shallow feature with a relatively small ceramic assemblage.

Next steps

As all preliminary projects do, this leads us to next steps and future questions. For example, we want to further test these results and examine possible earlier households using other artifact types at Building o. Additionally, we plan to run similar analyses on other quarter sites along Mulberry Row to get a better understanding of the broader applicability of these methods to stratified sites. Although these initial steps focus on identifying nuances and shifts in site chronologies, this process is important for reaching the final questions regarding what these timelines can reveal about individual households, their role in local economies, their distinct foodways, and ultimately what life was like for those living across the Monticello plantation.