Artifacts

by Scott G. Ortman

Chapter Contents

Introduction

1
This chapter synthesizes information on portable artifacts collected during the Crow Canyon Archaeological Center's test excavations at Yellow Jacket Pueblo (Site 5MT5). It also presents several intrasite analyses of Yellow Jacket Pueblo artifacts and compares these artifacts with those from other Pueblo sites in the Mesa Verde region of southwest Colorado. The tables and figures presented in this chapter were produced using the artifact databases as they existed in November 2001. I am not aware of any provenience changes that have been made since that time, but slight discrepancies between the data discussed in this chapter and those contained in the database may develop over time as errors in the database are discovered and corrected. However, it is likely that any such changes will be minor and will not affect the conclusions presented here.

Processing and Disposition of Materials from Yellow Jacket Pueblo

2
All objects collected during Crow Canyon's excavations at Yellow Jacket Pueblo were processed and classified into various stone, bone, pottery, vegetal, or other categories according to the standard procedures described in Crow Canyon's laboratory manual.

3
As of this writing, all artifacts, ecofacts, other samples, and original documentation, with the exception of wood samples submitted for tree-ring dating, are housed at the Anasazi Heritage Center, 27501 Hwy. 184, Dolores, Colorado, USA, for permanent curation. The collections are indexed to artifact databases, which are curated at both Crow Canyon and the Heritage Center and are accessible through The Yellow Jacket Pueblo Database and the research database on Crow Canyon's Web site. Tree-ring samples that produced dates, along with samples that might be datable in the future, are curated at the Laboratory of Tree-Ring Research, University of Arizona, Tucson, Arizona, USA. All human remains uncovered during excavations were dealt with in accordance with Crow Canyon's policy on the treatment of human remains (see the field manual). The analysis of these remains is described in detail by Bradley in the chapter titled "Human Skeletal Remains."

4
A few artifacts from Yellow Jacket Pueblo have undergone destructive analysis. We removed small portions from a sample of rim sherds of white ware bowls and corrugated gray jars for use in temper identification. A small number of rim sherds from white ware bowls were also subjected to instrumental neutron activation analysis to determine their chemical composition. Finally, samples submitted for tree-ring dating that possessed little dating potential were discarded by the Laboratory of Tree-Ring Research.

Additional Studies of Yellow Jacket Pueblo Artifacts

5
In addition to the analyses reported here, several additional studies using artifact data from Yellow Jacket Pueblo have been conducted. Ortman et al. (2000*1) present proportions of decorated pottery types by architectural block in their preliminary assessment of the occupational history of Yellow Jacket Pueblo. Ortman (2000*1) examines painted designs on white ware sherds from the great tower complex (Block 1200) at Yellow Jacket and from numerous other Pueblo II and Pueblo III sites in the Mesa Verde region. Temper, rim-arc, and rim-eversion data collected from the pottery assemblage from the great tower complex are presented for comparative purposes in the Castle Rock Pueblo and Woods Canyon Pueblo reports (Ortman 2000*2, 2002*1). Finally, Arakawa (2000*1) studied the chipped-stone tools and manufacturing debris from Yellow Jacket and presents several summaries of these materials.

Organization and Use of This Chapter

6
This chapter is organized into sections and subsections, a list of which can be viewed by selecting the expanded table of contents at the beginning of the chapter. Selecting a heading in the table of contents will take you directly to the section of interest. When you link to a table, figure, or reference in the text, a new browser window will open and display the selected information. You can move back and forth between the chapter text and the data window by keeping both windows open, overlapping them (that is, not viewing them full screen), and selecting one window at a time. The data window will be updated each time a new link for a table, figure, or reference is selected in the narrative text window; the text window will maintain your place in the longer document. In many subsections, contextual information taken from the field provenience database is provided in addition to analysis information for selected artifacts. Explanations of field provenience categories can be found in the field manual.

Unmodified Pottery Sherds

7
More than 66,000 unmodified pottery sherds, weighing more than 300 kg in total, were collected during excavations at Yellow Jacket Pueblo. Each sherd was analyzed according to Crow Canyon's standard procedures, which are described in the Crow Canyon laboratory manual. Most were identified as locally made, Mesa Verde–tradition white and gray wares. Several summaries of the basic sherd data are presented below.

Sherds by Ware and Type

8
The sherds collected from Yellow Jacket are tabulated in Table 1 according to pottery type (for type definitions, see the Crow Canyon laboratory manual). The list of pottery types is arranged according to the more general ware category to which each belongs. Unknown gray, white, and red ware sherds are listed separately because such sherds may or may not represent local wares. For each pottery type, the count, weight, and percentages by count and weight are presented. Pierce and Varien (1999*1) discuss the relative merits of counts vs. weights as measures of abundance.

9
Comparison of the counts and weights in Table 1 shows that the relative abundance of a specific pottery type in the assemblage varies depending on the measure of abundance used. This is especially clear for the formal white ware types—for example, Mesa Verde, McElmo, and Mancos black-on-white—which are much more abundant by weight than by count. In contrast, the relative abundance of Pueblo III White Painted, a more general type, is approximately equal by count and weight. Consistency in the relative frequency of a type for both count and weight probably indicates that sherds assigned to that type tend to be of average size for the collection overall. Greater relative frequency by count indicates that the mean weight of sherds assigned to that type is less than the mean weight of all sherds, whereas greater frequency by weight indicates that the mean weight of sherds assigned to that type is greater than average. It is expected that many sherds assigned to formal types will be larger than average because the classification of local white ware sherds relies heavily on identifying specific painted designs that are often difficult to identify on small sherds.

10
A few sherds were assigned to early (Basketmaker III and Pueblo I period) types—that is, Chapin Gray, Chapin Black-on-white, Moccasin Gray, Mancos Gray, Indeterminate Neckbanded Gray, Early White Painted, Early White Unpainted, Abajo Red-on-orange, and Bluff Black-on-red. The presence of these sherds suggests some human activity in the site area between A.D. 600 and 900. However, these sherds are too infrequent at the site to suggest significant early occupation in areas investigated by Crow Canyon. Nevertheless, Basketmaker III houses have been identified from excavations near Yellow Jacket Pueblo (Lange et al. 1986*1), so it is possible that habitations dating from the Basketmaker III or Pueblo I periods, or both, remain undiscovered in areas of the site that were not investigated during this project.

11
The vast majority of sherds recovered from test excavations at Yellow Jacket are of types manufactured during the late Pueblo II and Pueblo III periods, between A.D. 1000 and 1300. The large number and weight of sherds datable to these time periods suggest that the primary occupation of the site spanned more than two centuries. A detailed analysis of the occupational history of Yellow Jacket Pueblo is presented in the discussion of pottery dating (paragraphs 23–70) and in "Chronology."

Sherds by Ware and Form

12
As indicated above, all sherds collected from Yellow Jacket were assigned to one of six basic ware categories: plain gray ware, corrugated gray ware, white ware, red ware, nonlocal wares, and unknown. Sherds were also assigned to one of four basic form categories: bowl, jar, other, and unknown. Table 2 presents total counts, total weights, and percentages by count and weight for these various ware-form combinations. The percentages by count and weight are fairly comparable for most ware-form combinations, but some differences are apparent: corrugated jars and "unknown" white ware forms are slightly more abundant by count, indicating that these sherds tend to be smaller than average; white ware bowls and jars and "other" white ware forms are slightly more abundant by weight and thus are larger than average.

13
Sherds of these ware-form combinations have been found in roughly the same proportions in other Pueblo III sites in southwest Colorado that have been interpreted as permanent, year-round habitations (Pierce and Varien 1999*1). For example, at both Castle Rock (Ortman 2000*2:Table 2) and Woods Canyon pueblos (Ortman 2002*1), sherds from corrugated jars are most common, followed by sherds from white ware bowls, and then by sherds from white ware jars. This suggests that the Yellow Jacket pottery assemblage resulted from a set of domestic activities that produced sherds of various wares and forms at a relatively consistent rate across habitation sites. This inference is supported by the fact that nonhabitation sites possess strikingly different proportions of these ware-form categories in their sherd assemblages (e.g., Wilshusen et al. 1997*1).

White Ware Sherds by Type and Finish

14
"Finish" refers to specific surface treatments on decorated (red and white ware) pottery. For red ware vessels and sherds, "finish" refers to the presence or absence of slip; for white ware vessels and sherds, it refers to the presence or absence of paint and, if present, to the type of paint. Two kinds of paint are identifiable on sherds from decorated white ware vessels of the Mesa Verde region. Mineral paint derives from ground iron, manganese, or copper-rich rock that is held in liquid suspension. Carbon paint is believed to derive from the condensed extract of certain plants, such as Rocky Mountain beeweed (Cleome serrulata) and tansymustard (Descurainia richardsonii). Table 3 presents counts and weights of painted white ware sherds assigned to various type and finish categories. Table 4 presents the percentage of sherds of each type assigned to the various finish categories, as well as the percentage of each type among all white wares, regardless of presence, absence, or type of paint.

15
The data in Table 4 clearly show that the use of mineral paint declined over time at Yellow Jacket Pueblo. This pattern has been documented in numerous studies of pottery from the Mesa Verde region (e.g., Breternitz et al. 1974*1; Rohn 1977*1). The proportion of Pueblo III–type sherds (McElmo Black-on-white, Mesa Verde Black-on-white, and Pueblo III White Painted) with mineral-painted designs is also consistent with results of previous studies, which have identified a southeast-to-northwest trend in the use of mineral paint during the Pueblo III period. These studies have found that mineral-painted designs are more common on sherds from Pueblo III sites in the northwestern portion of the central Mesa Verde region, toward the Abajo Mountains in southeast Utah (Ortman 2002*1:Table 7; Wilson 1988*2:Table A.19), than they are on sherds from contemporaneous sites to the southeast, toward Ute Mountain and Mesa Verde proper (Ortman 2000*2:Table 3; Varien et al. 1992*1:Table 5.3). The proportion of decorated sherds exhibiting mineral-painted designs at Yellow Jacket Pueblo fits neatly into this spatial pattern, and thus supports results of previous studies.

Rim Sherds by Ware and Type

16
Rim sherds usually provide better estimates of the proportions of vessels of various traditional types used during an occupation than do body sherds, because rim sherds usually preserve more diagnostic attributes and therefore tend to be classified more precisely. Table 5 presents counts, weights, and percentages (by count and weight) of rim sherds by pottery ware and type in the Yellow Jacket assemblage. Comparison of these data with the data in Table 1 clearly shows that specific, named types are more prevalent among rim sherds alone than they are in the entire sherd assemblage.

17
As was the case for the overall sherd assemblage, notable differences in the relative frequencies of different types by count and weight probably relate to the average sizes of rim sherds assigned to each type. As an example, Mesa Verde Black-on-white is much more abundant by weight than by count, whereas Pueblo III White Painted and Indeterminate Local Corrugated Gray are more abundant by count than by weight. These patterns indicate that rim sherds assigned to specific traditional types tend to be larger than average, whereas rim sherds assigned to generic types tend to be smaller than average. The higher frequencies of specific types among rim sherds indicate that rims were assigned to these specific types more often than were body sherds.

18
The distribution of formal types among rim sherds generally supports the site chronology suggested on the basis of all sherds. In both cases, differences in the representation of formal white ware types are more apparent by weight than by count. However, by weight, the rank order in the relative frequency of specific, named pottery types is different for rim sherds alone than for all sherds. Although Mesa Verde Black-on-white is most common in both datasets, McElmo Black-on-white is more common than Mancos Black-on-white among rim sherds only, whereas the rank order is reversed for all sherds. Thus, rim sherds emphasize the Pueblo III occupation of Yellow Jacket more so than do all sherds.

Rim Sherds by Ware and Form

19
Rim sherds often can be assigned to more-specific form classes than can body sherds, and when it was apparent during analysis that a rim sherd came from a ladle, canteen, mug, or kiva/seed jar, this was recorded in a "comments" field in the database. Ladle rims inscribe tighter arcs than do bowl rims and possess either distinctive use wear on the outside edge of the rim or evidence of a handle attachment. Small jar rims with very tight curvature are from canteens. Mug rims are flat and almost always upright (rarely everted); they usually possess intricate painted decorations on their exteriors, and they may also preserve evidence of a handle attachment near the rim. Finally, kiva and seed jars are slightly larger than canteens, do not have necks, and, in the case of kiva jars, have a distinctive lip that is designed to hold a lid in place.

20
Table 6 summarizes the wares and forms of rim sherds in the Yellow Jacket assemblage by count and by weight. The more-specific vessel forms of kiva jar, seed jar, ladle, and mug are subdivided in this table on the basis of information recorded in the comments field of the pottery analysis database. For this table, white ware rims with no additional comments recorded in the file are assumed to be from large storage jars, or ollas. As is the case for the overall assemblage, rim sherds show relatively little variation in relative abundance by count vs. weight when classified in terms of ware-form combinations. This suggests that sherd size does not significantly affect the ability of analysts to assign rim sherds to wares and forms. Also, as was the case for the entire sherd assemblage, the three most common vessel forms represented among the rim sherds are corrugated jars, white ware jars, and white ware bowls. The relative frequencies of these three forms, however, are strikingly different when rim sherds alone are considered. White ware bowls are by far the most common ware-form combination among rim sherds only, whereas corrugated jars are by far the most common among all sherds.

21
These differences relate to the typical shapes of the original vessels and to differences in the relative numbers of rim and body sherds produced by vessels of different forms. White ware bowls are open forms with large rim circumferences; when they break, they produce numerous rim sherds and a relatively high ratio of rim to body sherds. Corrugated and white ware jars are taller, closed forms, usually with smaller rim circumferences, that produce far fewer rim sherds per vessel than do white ware bowls. As a result, the best way to estimate the relative number of vessels of different ware-form classes in a pottery assemblage is to compare the total degrees of arc encompassed by the rim sherds of various ware-form classes.

22
Such data were considered by Pierce and Varien (1999*1) in their study of assemblages from the Sand Canyon Archaeological Project Site Testing Program. They found that raw counts of rim sherds, though less precise than degree-of-arc measurements, nevertheless give a closer approximation of the relative numbers of vessel ware-form classes than do raw counts of all sherds. Thus, using raw counts of rim sherds as a guide, it appears that white ware bowls were the most common vessel form at Yellow Jacket, followed by corrugated jars and then white ware jars and white ware ladles. Canteens, mugs, and kiva/seed jars were all relatively rare.

Pottery Dating

23
A primary research goal of archaeological testing at Yellow Jacket Pueblo was to reconstruct the occupational history of this large and complex site using pottery assemblages. Because of the limited testing strategy, the widely varying occupation spans of architectural blocks, and the stratigraphic mixing that resulted from ancient and recent disturbances, achieving this goal presents a significant methodological challenge. In the following sections, I (1) discuss the pottery assemblages collected from each architectural block tested at Yellow Jacket Pueblo, (2) develop methods appropriate for estimating occupation spans and periods of occupation from these assemblages, and (3) apply these methods to block-level pottery assemblages to reconstruct the occupational history of the site.

Pottery Assemblages from Architectural Blocks at Yellow Jacket Pueblo

24
The testing strategy used at Yellow Jacket Pueblo was designed to be consistent with the goals of conservation archaeology (Lipe 1974*1). The number and location of test units were tailored to disturb subsurface deposits as little as possible while allowing us to recover sufficient pottery samples for dating purposes. The resultant strategy, used everywhere except in Architectural Block 1200, was to identify midden (trash) deposits adjacent to each architectural block and to excavate up to seven 1-x-1-m test units in each of these areas. These excavations were limited to the portions of the site that we had permission to test. In most test units, excavators found midden deposits that are spatially associated with specific architectural blocks.

25
In addition to these artifact-focused units, a 1-x-2-m unit was placed along the exterior face of the north wall of each architectural block to examine stratigraphic relationships between the latest architectural features identified on the modern ground surface and those from earlier occupations. In many of these units, bedrock or undisturbed native sediment was discovered directly underneath the architecture visible on the modern ground surface, which suggests little or no prior habitation of these areas (see "Chronology"). However, several test units located along the north walls of architectural blocks in the central area of the site did expose structures, features, and other cultural deposits underneath the latest architectural features. This evidence indicates that the initial Pueblo occupation of certain areas predates construction of architectural features visible on the modern ground surface. Evidence of earlier occupations may also exist in other areas of the site but was not detected by our limited testing.

26
The sampling strategy followed at the great tower complex (Block 1200) was different from that used in other architectural blocks because many of the rooms and kivas of this complex had been excavated previously (see paragraphs 31–32 in "Architecture"). During these previous excavations, the excavated fill from one structure was used to backfill a previously excavated structure, and some complete artifacts were collected; however, many sherds, flakes, ground-stone tools, and other artifacts were left behind. Thus, the excavation and screening of this backdirt had the potential to yield an essentially unbiased sample of sherds and chipped-stone artifacts from the great tower complex and to expose architectural features without disturbing intact stratigraphy. We therefore decided to excavate test units inside several previously excavated structures in Block 1200. Due to the larger number of units excavated in this block, the artifact assemblage from the great tower complex is much larger than the assemblages from other architectural blocks. Also, because this complex was built directly on bedrock, its entire artifact assemblage can be confidently associated with structures visible on the modern ground surface.

27
In other architectural blocks, however, the possibility of buried cultural features means that pottery assemblages cannot be associated with visible structures so directly. Test pits were placed in middens to obtain pottery samples for dating adjacent buildings visible on the modern ground surface, and it is likely that much of the pottery collected from these excavations was indeed generated by activities in these buildings. However, if a given area was occupied before the buildings visible on the modern ground surface were constructed, a portion of the pottery assemblage from this area could have been associated with activities that predated the construction of those buildings. Midden deposits at Yellow Jacket were likely disturbed during the Pueblo occupation through trampling and the excavation of graves, and many have been disturbed in historic times by nonprofessional diggers looking for burials with complete vessels. As a result, midden stratigraphy cannot be used to confidently distinguish sherds that postdate the construction of buildings visible on the modern ground surface from sherds that predate these buildings. Therefore, the most conservative and consistent interpretation of the pottery sample from each architectural block is that it represents the entire occupational history of that area. These samples are also biased toward activities associated with buildings visible on the modern ground surface because architecture visible on the modern ground surface guided the placement of test pits.

28
With these assumptions and limitations in mind, in the following sections I use pottery assemblage data from architectural blocks at Yellow Jacket Pueblo to (1) estimate the total occupation span of each architectural block area and (2) define the time periods during which each of these blocks was occupied.

Estimating the Occupation Spans of Architectural Blocks

29
It is clear from our excavations that the occupation spans of the architectural blocks tested at Yellow Jacket Pueblo were highly variable in length. For example, evidence suggests that Block 1200 was occupied for fewer than 35 years: it was built directly on bedrock sometime after A.D. 1254, and it is believed that all ancestral Pueblo people had left southwest Colorado by about A.D. 1290 (Lipe 1995*1). In contrast, Block 200 appears to have had a complex and lengthy use history: the test unit in the roomblock revealed a full height, pecked-block masonry wall underneath the pecked-block masonry wall visible on the modern ground surface.

30
In this section I estimate the occupation spans of architectural blocks at Yellow Jacket using (1) middle-range theory that relates artifact accumulation to site population and occupation span and (2) data from the Site Testing Program (Varien 1999*1, 1999*2) that relate occupation-span estimates to the mean weight of corrugated pottery in midden test units. There is a long tradition of research that attempts to relate artifact accumulations to the person-years of occupation at archaeological sites (for a recent review, see Varien and Mills [1997*1]). The underlying logic of such research is that artifacts wear out and are discarded as they are used and that the number of times an artifact can be used before wearing out is predictable; that is, a histogram summarizing the number of times a particular kind of tool or utensil can be used before wearing out will be unimodal, with a mean number of uses and a standard deviation. As a result, artifacts used in routine domestic activities should accumulate on a site in proportion to the use life of the artifact type, the number of people who used this artifact type on a routine basis, and the time span over which the activities that required this artifact type occurred.

31
All 1-x-1-m test units (sampling columns) excavated in midden areas at Yellow Jacket have the same surface area. Because of this, the total number of artifacts recovered in a midden sampling column should vary in accordance with the depth and density of the trash deposit in that column. If a consistent proportion of the artifacts used in routine domestic activities was discarded in middens associated with each architectural block, the total accumulation of such artifacts in a sampling column should be proportional to the number of households contributing material to that column and the time span over which the trash was deposited. The mean accumulation of artifacts across sampling columns in a midden area, then, should provide a relative measure of the household-years of deposition represented by each midden.

32
Figure 1 presents the mean accumulation of chipped-stone manufacturing debris and corrugated cooking pottery across midden sampling columns for each tested architectural block. These measures have been found to correlate highly with household-years of deposition in other ancestral Pueblo sites (Nelson et al. 1994*1:128–130). The substantial variation in these two measures across architectural blocks and the strong, positive correlation r = .88) between them support the theory that these measures reflect variation in household-years of deposition rather than functional differences among architectural blocks. The one clear exception to this pattern is the assemblage from Block 2000, which lies farthest from the regression line along which most block-level assemblages fall when characterized using these two measures. This block does not contain surface rooms and kivas, but instead is a large earthen depression that might have been a reservoir or plaza. It is thus likely that the deposits in this area did not result from typical domestic activities (also see paragraph 165).

33
To move from artifact accumulations to occupation spans for architectural blocks, we need some way to translate measures of household-years of deposition to measures of years of occupation only. In other words, we need to control for the number of households that contributed artifacts to each excavated sampling column. We cannot estimate the population history of Yellow Jacket very precisely from existing data, because we have no way to determine which architectural features were in use at any given time, nor can we estimate the number of buried features. However, the architectural remains of a typical household in the central Mesa Verde region encompass an area of at least 50 m² (estimate based on data in Lipe [1989*1:Table 1]), so it is unlikely that more than one household discarded trash on any particular square meter of midden at any given point in time. It is therefore reasonable to assume that, on average, artifact accumulation in a midden sampling column is the result of trash deposition by one household at a time over the total occupation span of that area. This assumption enables us to consider the mean accumulation of artifacts across midden sampling columns as a relative indicator of the mean number of years that households in each architectural block deposited trash in associated middens.

34
The suitability of mean artifact deposition across midden sampling columns as a relative measure of occupation span can be tested using excavation data from the Site Testing Program (Varien 1999*1, 1999*2). In this project, 13 small ancestral Pueblo sites in southwest Colorado, each with architectural evidence of between one and 13 households, were tested using stratified random sampling methods. The random samples were used to estimate the total discard of corrugated pottery at each site. This figure was divided by the number of households indicated by the architecture present and a discard rate for cooking pot sherds derived from archaeological and ethnographic data to estimate the occupation span of each site. One of the six sampling strata defined for each tested site was midden areas visible on the modern ground surface. Thus, we can calculate the mean weight of corrugated pottery collected from 1-x-1-m test pits within middens at each tested site and compare this figure to its estimated occupation span.

35
Table 7 and Figure 2 examine the relationship between these two measures for the 13 sites tested during the Site Testing Program. The corrugated gray sherd weights from all 1-x-1-m units excavated within the midden sampling stratum at each site were used to calculate the mean accumulation of corrugated pottery in midden test units, with the exception of one unit (94N 133E) at Castle Rock Pueblo that exposed an anomalous deep crevice filled with artifacts. The relationship between these measures is close, positive, and linear for 10 of the 13 tested sites, but three sites deviate from the pattern suggested by the other 10. Two of these three sites, Roy's Ruin and Shorlene's Site, contain less corrugated pottery per 1-x-1-m unit than their occupation-span estimates suggest they should. A third site, Catherine's Site, contains more corrugated pottery per test unit than one would expect from its estimated occupation span. These deviations from linearity can be accounted for by postdepositional processes and architectural interpretations.

36
The lower-than-expected mean weights of corrugated sherds in midden test units at Roy's Ruin and Shorlene's Site are likely the result of recent plowing associated with modern agricultural activities. Plowing tends to spread pottery sherds across a larger area, thus reducing the amount of material in any given sampling column. The higher-than-expected mean weight in midden test units at Catherine's Site, on the other hand, may be due to an inaccurate estimate of the occupation span for this site. Varien's (1999*1:Table 5.3) estimate assumes that the two kivas at this site were occupied contemporaneously, based on similar abandonment stratigraphy in these kivas. Although the roofs of both structures were dismantled at abandonment, one kiva had been heavily remodeled before the roof was removed, whereas the other exhibited no evidence of remodeling. If the use of these kivas was sequential or partly overlapping instead of contemporaneous, the occupation span of this site would have been longer and therefore closer to the value predicted by the regression line in Figure 2. In other words, the total accumulation of corrugated pottery at Catherine's Site indicates 67 household-years of deposition, which could reflect occupation by two households for 33 years (Varien's estimate), one household for 67 years, or some combination of these variables. The mean weight of corrugated sherds in midden test units at the site suggests that a longer occupation by one household is more likely.

37
If the two tested sites with plowed middens are excluded, and the longer occupation span for Catherine's Site is used, then the mean weight of corrugated gray sherds in midden test pits becomes a very good estimator of occupation span for the tested sites, as is shown in Figure 3. The regression equation in Figure 3 enables one to estimate the occupation span of these sites based on the mean weight of corrugated sherds in midden test pits, with a standard error of +/- 5 years. This regression equation can also be used to estimate the occupation spans of architectural blocks at Yellow Jacket Pueblo.

38
Table 8 presents accumulation data for corrugated sherds and the resultant estimates of years of occupation for each tested architectural block at Yellow Jacket Pueblo. These data suggest that the occupation spans of these areas were highly variable, ranging from 7 to 230 years. It is also apparent from these data that the occupation spans of Blocks 100–700 and of Block 2000 were longer than those of other architectural blocks. These eight blocks are all located along the central, north-south "spine," or ridge, on which the pueblo was constructed. These data thus suggest that areas along the central spine were occupied for longer periods than areas to the east of the spine and on the talus slopes below the canyon rim. Most of the buried cultural features that predate buildings on the modern ground surface were also found along this central spine (see the last column of Table 8). This spatial association between extensive midden deposits and buried features thus supports the theory that accumulations of corrugated sherds in midden sampling columns can be used to estimate the occupation spans of architectural blocks.

Estimating the Periods of Occupation of Architectural Blocks

39
In this section, I present the pottery data used to estimate periods of occupation for each architectural block tested at Yellow Jacket Pueblo, and I illustrate why traditional seriation approaches are not suitable for these data. I then develop an alternative method that combines a calibration dataset with Yellow Jacket pottery data to calculate the probability of occupation in each tested architectural block during each of seven temporal phases.

Type and Attribute Data from Yellow Jacket Pueblo

40
An extensive literature addresses the relative value of traditional pottery types and design attributes for pottery dating in the Southwest (Duff 1996*1; Hegmon 1991*1; LeBlanc 1975*1; Ortman 1995*1; Plog and Hantman 1986*1). For this study, I used both design-attribute data and traditional typological data derived from the analysis of rim sherds from bowls. The decision to focus on bowl rims was based on several factors: (1) a calibration dataset for design attributes had been developed previously, using assemblages of bowl rim sherds from tree-ring-dated sites with short occupation spans (Ortman 2000*1); (2) certain of these design attributes have more-restricted temporal ranges than do traditional pottery types; (3) most of the decorated pottery sherds that could be assigned to specific types in the Yellow Jacket assemblage were from bowl rims; (4) I believe there is more consistency in the classification of rim sherds than in the classification of body sherds; and (5) the recording of attributes as well as types maximizes the chronological information obtained from small pottery samples such as those recovered during our testing of each architectural block.

41
Table 9 presents counts of these pottery types and design attributes for decorated bowl rim sherds from each tested architectural block at Yellow Jacket Pueblo. The following "type" categories were tabulated: slipped and painted San Juan Red Ware (Deadmans Black-on-red and Indeterminate Local Red Painted with slip), Cortez Black-on-white and Pueblo II White Painted, Mancos Black-on-white, McElmo Black-on-white, Mesa Verde Black-on-white, Pueblo III White Painted, and Late White Painted. The attributes selected for recording are relatively easy to identify and record consistently and are known from previous studies (e.g., Hegmon 1991*1; Ortman 2000*1) to be chronologically sensitive. The presence, absence, or indeterminate presence or absence of the following attributes was recorded for each decorated rim sherd from a white ware bowl in the Yellow Jacket assemblage:

• Line painted on rim: a solid line around the vessel circumference at the rim
• Ticked rim: discontinuous, repeating dots, dashes, or slashes on the rim
• Xs and zig-zags painted on rim: continuous, repeating Xs or zig-zags on the rim
• Undecorated rim: no painted decoration noted on the rim surface or edge
• Corrugated exterior: unobliterated coils with patterned indentations on the exterior surface
• Exterior band design: continuous painted design around the circumference of the vessel on its exterior surface
• Thick and thin framing lines: parallel lines of varying thickness around the circumference of the vessel on its interior surface just below the rim
• Dots between framing lines: dots, ticks, or "musical notes" painted in the spaces between framing lines on the interior of the vessel just below the rim
• Mineral paint: use of mineral-based paint in any of the recorded designs

42
It is important to note that in design-attribute analysis, in contrast to traditional typological analysis, not every attribute can be evaluated on every sherd: on one sherd the exterior may be eroded but the rim present, on a second sherd the interior will be encrusted but the exterior clean, and so on. Thus, the effective sample size—the number of times the presence or absence of an attribute can be determined—will vary from attribute to attribute using the same sample of sherds. To produce accurate estimates of attribute proportions in an assemblage, then, one must consider the number of sherds on which the presence or absence of each attribute could be determined, rather than the total number of sherds in the assemblage. To this end, Table 10 presents sample size data, or the number of determinate presence/absence observations made for each attribute, for each architectural block.

43
Initial efforts to reconstruct the occupational history of Yellow Jacket, conducted by Glowacki (1999*1), used typological data like those presented in Table 9 in multivariate analyses to produce seriations. For a seriation to represent a relative chronology, the ordered assemblages must have accumulated over comparable time spans (Dunnell 1970*1). Glowacki attempted to create assemblages representing comparable time spans by separating out sherds known, from stratigraphic evidence, to have been deposited before roomblock construction. Unfortunately, the absence of clear stratigraphy in most midden deposits made it impossible for Glowacki to define midden assemblages representing comparable spans of sherd deposition, despite evidence that these middens accumulated over widely varying lengths of time (see the discussion of occupation-span estimates, paragraphs 29–38). Thus, there is no means by which to consistently subdivide block-level pottery assemblages into groups that represent comparable spans of sherd deposition. As a result, the relative frequencies of types and attributes in block-level pottery assemblages can be expected to vary in accordance with the overall occupation span of each architectural block as well as with the mean occupation date of each area.

44
Variation in deposition of corrugated sherds across architectural blocks can be used to illustrate why it is difficult to interpret multivariate analyses of block-level assemblages in chronological terms. Figure 4 examines the relationship between multivariate seriation results and corrugated sherd deposition across architectural blocks. The x-axis gives the first axis score from a correspondence analysis of type and attribute counts for the rim sherds from decorated bowls from each architectural block. This axis accounts for approximately one-half of the total inertia, or variation, in this dataset. It is standard practice to assume that most of this variation is due to the passage of time, and therefore, that the axis that accounts for the greatest proportion of this variation captures the chronological sequence for these assemblages. A noncutting tree-ring date of A.D. 1254 from Block 1200 anchors the late end of this putative seriation and suggests that assemblages with negative scores date from the late years of the site occupation, whereas assemblages with positive scores date from the early years of the occupation.

45
The y-axis plots the deposition of corrugated sherds in midden test units as an estimate of the relative occupation span of each architectural block. The resultant graphic shows that a chronological interpretation is misleading in this case. All assemblages with more than 1,500 g of corrugated pottery per midden test unit fall in the middle range of the seriation suggested by the first axis of the correspondence analysis. In contrast, all assemblages that fall at either end of the seriation produced less than 1,500 g of corrugated pottery per midden sampling column. That assemblages deriving from long deposition spans should fall in the middle of a seriation makes some sense, since such assemblages are more likely to contain pottery exhibiting characteristics of both the early and late years of deposition. It also makes sense that assemblages exhibiting predominantly early or late characteristics will tend to have been deposited over shorter time spans. However, this analysis does not provide enough information to determine the relative chronology of architectural blocks inhabited for longer periods.

46
What we need to overcome this problem is a method of pottery dating that (1) takes the relative span of sherd deposition in each roomblock area into account and (2) allows for mixing of pottery from multiple periods of occupation. Kohler and Blinman (1987*1) used multiple regression analysis in combination with a calibration dataset to apportion pottery from mixed assemblages to time periods. An alternative approach developed here combines features of mean ceramic dating (Christenson 1994*1), assemblage-based dating (Wilson and Blinman 1999*1), and composite ceramic–distribution dating (Steponaitis and Kintigh 1993*1) to estimate the occupational histories of architectural blocks using pottery samples recovered in Crow Canyon's testing at Yellow Jacket. The following sections present the calibration data, assumptions, and calculations of this method, as well as the results generated from its application to block-level pottery assemblages from Yellow Jacket Pueblo.

Calibration Dataset

47
A calibration dataset summarizes quantitatively the characteristics of pottery assemblages from sites of known age. Using this information as a gauge and employing a variety of statistical techniques, one can apportion pottery assemblages like those from Yellow Jacket to various time periods. The calibration dataset for this study consists of type and attribute data for rim sherds from decorated bowls at 35 tree-ring-dated sites with relatively short occupation spans. These assemblages have been analyzed previously as part of a study of pottery painting in the Mesa Verde region (Ortman 2000*1). The same analysts examined both the Yellow Jacket Pueblo assemblage and the calibration dataset assemblages, and the same types and attributes were recorded for all sites. The information from these well-dated sites can be used to develop a calibration data set that summarizes pottery design change over time.

48
Twenty-nine of the sites in the calibration dataset contain a single occupational component, and six contain two occupational components. In two-component sites, sherds were assigned to one of the two components using provenience information. At sites with two components separated by an occupational hiatus, the type to which individual sherds were assigned was also used when it was obvious that a sherd was not associated with the component suggested by its provenience. At two-component sites with continuous occupations, however, sherds were assigned to components using provenience information only. Table 11 and Table 12 give counts and sample sizes, respectively, for types and attributes recorded for each of the 41 components in the calibration dataset.

49
Because these components do derive from comparable spans of sherd deposition, traditional seriation approaches can be used to test the suitability of the recorded data categories for chronological study. This can be accomplished by comparing a relative chronology suggested by multivariate analysis of the pottery data with the absolute chronology of the tree-ring-dated components. Figure 5 makes this comparison. The x-axis plots these components according to the latest tree-ring date with which each is associated, regardless of whether the date is a cutting or noncutting date. A certain amount of error is unavoidable when placing these components in an absolute chronology because of sampling and preservation issues, variable relationships between tree-harvesting and construction and occupation, and differences in the treatment of structure roofs at abandonment. As long as the chronological relationship between the latest tree-ring date and sherd deposition is basically similar across components, however, the tree-ring data should provide a simple and straightforward way of estimating the absolute chronological relationships among these components.

50
The y-axis gives the first axis score from a correspondence analysis of type and attribute counts for tree-ring-dated assemblages of rim sherds from decorated bowls in the calibration dataset. This axis accounts for approximately three-quarters of the total inertia, or variation, in this dataset. It is clear from the high correlation (r = .9616) between the correspondence analysis results and the tree-ring data that the type and attribute data recorded for the components in the calibration dataset closely reflect the absolute chronology of these components. Thus, the type and attribute categories tabulated in this dataset and in block-level assemblages from Yellow Jacket Pueblo should be adequate for chronological study.

51
On the basis of the latest associated tree-ring date, each of the 41 components in this dataset was assigned to one of seven temporal phases. The seven phases are A.D. 1020–1060, A.D. 1060–1100, A.D. 1100–1140, A.D. 1140–1180, A.D. 1180–1225, A.D. 1225–1260, and A.D. 1260–1280. In a few multiple-component sites, only one of the two components is associated with tree-ring dates. In such cases, the undated component was assigned to a phase on the basis of architectural details and pottery assemblage data.

52
For each of the seven temporal phases, Table 13 estimates the proportion of sherds that are of a given type or that exhibit a given attribute. These estimates were derived using empirical Bayesian statistical methods for proportions as described in Iversen (1984*1:18–33) and as applied to archaeological data by Robertson (1999*1). Bayesian methods, in which prior information on a given population parameter is combined with the sample data to produce a refined estimate of that parameter, are an alternative to classical statistical techniques. In this case, the proportion of sherds of a given type or exhibiting a given attribute was estimated for each phase, using the calibration dataset and Robertson's (1999*1:140) techniques. The mean of the posterior probability-density function (µ'') for each parameter—an estimate of the proportion of sherds deposited during a certain phase that exhibit a particular type or attribute—is given by equation 1:

where x = the number of presences of a given type or attribute in a given phase, n = the number of determinate observations for that parameter in the given phase, and a and b are derived as follows (equations 2 and 3, respectively):

In equations 2 and 3, µ' and s'² are parameters of the prior distribution derived from the sample data: µ' is the mean proportion of a given type or attribute across calibration dataset components dating from a given phase, and s'² is the variance in these proportions. The resultant µ'' values are given in each cell of Table 13. Because a small amount of mixing is unavoidable among the components included in the calibration dataset, and also because it is likely that some analytical errors exist in the database, the estimates in Table 13 assume that (1) a given type or attribute did not appear until the phase during which it occurs on more than 2 percent of sherds and (2) the attribute or type disappeared by the phase during which it occurred on less than 1 percent of sherds in the calibration data.

53
Using the data in Table 13, we can divide the value in each cell—the estimated proportion of sherds showing a given type or attribute in each period—by the sum of proportions for this type or attribute across phases to calculate the probability that a sherd of a particular type or exhibiting a given attribute dates to each of the seven phases. These probability distributions for types and attributes across phases allow us to assess the chronological significance of each sherd in assemblages that contain sherds deposited during more than one phase. In like fashion, the chronological significance of an entire assemblage can be assessed by adding together one associated probability distribution (an array of seven numbers representing the probability that a sherd of a given type or exhibiting a given attribute was deposited during each of the seven phases) for each occurrence of a type or attribute in the assemblage. This will produce a composite probability-density distribution that summarizes the probability that sherds contributing to the assemblage were deposited during each of the seven phases.

54
Phase probabilities for each type and attribute are given in Table 14. These values indicate, for example, that a rim sherd classified as Mancos Black-on-white was almost certainly deposited sometime between A.D. 1020 and 1180, and that its most probable single phase of deposition was between A.D. 1100 and 1140, although deposition in earlier phases was also reasonably likely. To make these phase probabilities more suitable for composite probability-density analysis, they have been scaled to account for biases that can be incorporated into this kind of analysis when several attributes with correlated chronological distributions are recorded for each sherd. For example, many rim sherds deposited during Phase 7, between A.D. 1260 and 1280, will be classified as Mesa Verde Black-on-white and will exhibit rim ticks, thick and thin framing lines, and exterior band designs. Thus, one sherd exhibiting these characteristics will contribute four probability distributions to a composite distribution. In contrast, many rim sherds deposited during Phase 4, between A.D. 1140 and 1180, will be classified as McElmo Black-on-white and will exhibit rim ticks but no additional attributes. A sherd exhibiting these characteristics will thus contribute only two probability distributions to the composite distribution.

55
This example suggests that the estimates of type and attribute proportions in Table 13 may be biased in favor of certain phases and against others. One can gauge the extent of this bias by summing the type or attribute probability for each phase. The second to last column in Table 13 presents these phase probability totals, which illustrate that these data are slightly biased against the middle phases of the time span encompassed by the calibration dataset. This bias probably derives from the analysis system used rather than from historical trends in occupation. To take this bias into account, the phase probabilities in Table 14 have been weighted so that there is equal prior probability for each time period across types and attributes. This was accomplished by multiplying the raw phase probability for each type or attribute by a ratio calculated for each phase; the ratio itself was calculated by dividing the sum of probabilities for all types and attributes within that phase by the mean of these values across all seven phases. As a result, certain types and attributes contribute more total "probability" per occurrence to the composite-probability distribution than others. The bottom row of Table 14 gives the relative weight given to each type or attribute in this analysis. If these phase probability distributions had been left unweighted, each would have summed to 1.000.

56
In the absence of a fairly complete calibration dataset like the one used here, previous studies using composite probability distributions (e.g., Carlson 1983*1; Christenson 1994*1; Hurt 2001*1; Steponaitis and Kintigh 1993*1) have assumed that the chronological distributions of pottery types take the form of normal, bell-shaped curves. The scaled phase probability distributions for pottery types and attributes derived from the calibration dataset are plotted in Figure 6 and Figure 7; these graphs show that this assumption is not realistic for these data. These distributions are highly variable, and many more closely approximate an S-shaped logistic curve than resemble a bell-shaped normal curve. In fact, most studies on the diffusion of culture traits in living human populations have found that logistic curves best capture the adoption dynamics of such traits over time (Henrich 2001*1). Thus, analyses using these empirically derived probability-density curves are likely to produce more realistic results than approaches relying on theoretical probability-density curves that do not in fact model the typical dynamics of cultural transmission seen in empirical studies.

Probability-Density Analysis

57
The next step in the analysis is to use phase probability distributions for types and attributes in conjunction with sample data to compute a composite probability-density distribution for that sample. This is accomplished in three steps: (1) the array of phase probabilities for each type or attribute given in Table 14 is multiplied by the number of sherds of the corresponding type or with the corresponding attribute in an assemblage (in this case, each assemblage is a block-level pottery assemblage from Yellow Jacket given in Table 9); (2) the results are summed across all types and attributes; and (3) the total probability for each phase is divided by the number of observations in the assemblage to obtain the proportion of total probability assigned to each phase. The resulting proportions are initial estimates of the probability of occupation of each block during each of the seven phases. These initial proportions are given for the block-level assemblages from Yellow Jacket Pueblo in Table 15(A).

58
To produce Bayesian estimates of these proportions, we also need to calculate the probability of actually obtaining the sample proportions of types and attributes from each assemblage, on the assumption that each sample was deposited during a particular phase with associated population proportions given in the calibration dataset. The question asked in this step is, What is the probability of obtaining the observed sample proportion of a type or attribute if the sample were drawn from a given phase, given the population proportion estimate for that type or attribute in that phase in the calibration dataset? This figure can be obtained for each period for each type or attribute in each assemblage using the assemblage data, the calibration dataset, and the binomial distribution. For each of the seven phases, the mean of these values across types and attributes is taken as an estimate of the conditional probability that the sample was drawn from that phase. The resulting array of conditional probabilities for the block-level pottery assemblages from Yellow Jacket Pueblo is given in Table 15(B). These data indicate, for example, that the mean probability of obtaining the observed sample of type and attribute proportions from Block 100, if it in fact dated to Phase 1 (A.D. 1020–1060), is .08.

59
Finally, these conditional probabilities are combined with the initial proportion estimates using Bayes' theorem to produce final estimates of the probability of sherd deposition during each of the seven phases under consideration. The version of Bayes' theorem used to calculate, for example, the probability of sherd deposition during Phase 1 in Block 100 is shown below in equation 4 (after Iversen 1984*1):

This equation reads, "the posterior probability of sherd deposition in Block 100 during Phase 1 equals the prior probability of sherd deposition in Block 100 during Phase 1, times the conditional probability of obtaining the Block 100 sample if it were deposited during Phase 1, divided by the sum, across Phases, of the prior probability for each Phase for Block 100, times the conditional probability for the Block 100 sample associated with each of the seven phases." The results of these calculations are given in Table 15(C). These final estimates are used below to assess the occupational histories of architectural blocks at Yellow Jacket.

A Test of the Method Using Sites from the Calibration Dataset

60
To my knowledge, the method of pottery dating presented in the above paragraphs has not been attempted previously. Thus, it may prove useful to check how well this method can predict periods of occupation at sites of known age. Sites in the calibration dataset (Table 11 and Table 12) can be used for this purpose. The analyzed assemblage from each site includes one or two occupational components. Each component dates from only one of seven time periods between A.D. 1020 and 1280, and most are associated with tree-ring dates. At least one of the two components at every two-component site is associated with tree-ring dates, and the relative chronological relationships between components at two-component sites are clear in every case.

61
Table 16 presents Bayesian posterior probability distributions for the sites in the calibration dataset. These distributions were calculated using the same procedures applied to the block assemblages from Yellow Jacket Pueblo (see the discussion of the probability-density analysis, paragraphs 57–59). In this analysis, assemblages associated with each component at multiple-component sites were aggregated to produce "mixed" assemblages. These whole-site assemblages were used to determine whether the methods developed in this report can (1) assign assemblages from sites of known age to their proper temporal phases and (2) distinguish and date components from mixed assemblages. The table entries in boldface type represent phases during which occupation is known to have occurred for each of these sites (inferences based on tree-ring dates). Underlined table entries indicate phases of occupation suggested by the application of the following decision rules to the data in the cells of the table: (1) if the site is known to be a single-component site, the single most probable phase (or the mode of the posterior probability-density distribution) was chosen as the phase of occupation; and (2) if the site is known to have included two occupational components, the shape of the probability distribution determined which two phases were chosen. If this distribution was bimodal, the two modes were chosen; if it was unimodal, the two most probable phases were chosen.

62
An important point raised by these decision rules is that, by themselves, posterior probability distributions are insufficient for determining the occupational histories of sites. One must also know something about the occupation span of a site to interpret these distributions. In the absence of this sort of ancillary information, other researchers working with cumulative probability distributions have defined the occupation span of a site by one of two standards: either as all phases above a certain probability level (for example, .2 in Hurt [2001*1]) or as the interval that encompasses a specified proportion of the total area under the distribution, excluding the tails (for example, 75 percent in Steponaitis and Kintigh [1993*1]). Application of such rules to the Bayesian posterior probability distributions for the sites in the calibration dataset would produce imprecise results and would overlook the bimodal probability distributions that characterize sites with two discontinuous occupations separated by a period of abandonment.

63
For example, using the posterior probability distribution for Saddlehorn Hamlet (Table 16), we can say that the likelihood that the artifacts at this site were deposited between A.D. 1180 and 1280 is .92. Indeed, the probability of sherd deposition during each phase in this interval is greater than .2. But this does not necessarily mean that the actual occupation of Saddlehorn Hamlet was a century in duration. In fact, accumulations of corrugated sherds (Table 7) suggest that the occupation span of this site was only about 23 years and therefore occurred during only one phase. In addition, the latest tree-ring date from the site, a noncutting date of A.D. 1256, suggests that this occupation occurred in the mid–A.D. 1200s. So in this case, picking only the most probable phase in the site's probability-density distribution, A.D. 1225–1260, yields an estimate of its occupational history that is consistent with pottery accumulations and tree-ring data. But we need to know how many phases to pick from the probability distribution to arrive at this best estimate.

64
The components in the calibration dataset were defined in such a way that each component would date from one, and only one, phase. If we interpret posterior probability-density distributions for these sites using decision rules that take this prior knowledge into account, the similarity between phases suggested by the probability-density analysis (the underlined entries in Table 16 ) and those suggested by tree-ring data (the Table 16 entries in boldface type) shows that probability-density analysis can accurately date sites with known occupation spans. Thirty of the 41 components in the calibration dataset are assigned to the same phase by both tree-ring data and probability-density analysis. Furthermore, the 11 components that do not coincide are, in every case, assigned to adjacent phases by the two methods. Figure 8 shows that discrepancies in these phase assignments are minor and random. However, there is a slight tendency for assemblages tree-ring-dated to Phase 2 or 3 to be assigned to an earlier phase by the probability-density analysis. The assemblages from several of these sites are quite small (Table 16), so it is possible that sampling error is responsible for at least some of these discrepancies.

65
On one level these results are not surprising, because the components in the calibration dataset and the site assemblages assigned to phases in this test use sherds from the same sites. At a minimum, this test shows that the method's calculations do not obscure the occupational histories of these sites. The strongest possible test would be to use probability-density analysis of "mixed" or total-site pottery assemblages to date the occupations at multiple-component sites of known age that are not included in the calibration dataset. It would be somewhat misguided, however, to conduct such a test without using the best possible estimates of type and attribute proportions for each phase, and to do so one would need to incorporate information from as many appropriate contexts as is feasible in the calibration dataset. It is also important to recognize that the calibration data values are not simple proportions calculated by aggregating all components dating from each phase into a single assemblage. Rather, they are estimates of population parameters derived from several independent samples drawn from the population of sherds deposited during a given phase. It is not at all guaranteed that individual samples drawn from such a population will be assigned to the proper phase through comparison with population parameter estimates derived from numerous samples. That Bayesian probability-density analysis can do so with an acceptable degree of accuracy suggests that errors in assigning the sites in the calibration dataset to phases probably derive more from shortcomings in the available data than from the mathematical characteristics of the method.

Pottery Dating Synthesis, Yellow Jacket Pueblo

66
We are now in a position to combine occupation-span estimates with the results of probability-density analysis to estimate the occupational history of each investigated architectural block at Yellow Jacket Pueblo. Table 17(A) presents occupation-span estimates for each architectural block derived using the regression equation developed from the Sand Canyon locality tested sites (see paragraphs 29–38). Table 17(B) presents the Bayesian posterior probability distribution for each architectural block (see paragraphs 57–59). The phases during which sherd deposition (in most cases resulting from habitation) is inferred for each architectural block are indicated by the table entries in boldface type. The following decision rules were used to allocate years of occupation in Table 17(A) to specific phases in Table 17(B): (1) the single most probable phase was chosen, and the number of years in this phase was subtracted from the occupation-span estimate; (2) if the probability distribution was bimodal and the remaining years to be allocated were more than half the number of years in the phase corresponding to the secondary mode, the phase in which the secondary mode occurs was chosen; (3) additional phases of occupation were chosen in decreasing order of their relative probability, and the number of years in each additional phase was subtracted from the remaining years of occupation for that block, until the difference between the occupation-span estimate and the number of years in the chosen phases of occupation was closest to zero.

67
We can make several interpretations regarding the occupational history of Yellow Jacket Pueblo on the basis of the table entries in boldface type in Table 17(B). First, we can say that the occupation of the site was continuous over approximately 220 years, because sherd deposition is likely to have occurred in multiple architectural blocks during each period between A.D. 1060 and 1280. These data also suggest that relatively few architectural blocks were occupied between A.D. 1140 and 1180. This suggestion is consistent with evidence of declining agricultural productivity (Van West and Dean 2000*1), increased social conflict and violence (Billman et al. 2000*1; Kuckelman et al. 2000*1; Turner and Turner 1999*1), and declining construction activity (Varien 1999*1:188–192) across the central Mesa Verde region during this period. However, the calibration data are also weakest for this interval, so it remains possible that this apparent decline in population is illusory. Finally, these data suggest that the site population was greatest between A.D. 1180 and 1225, that the population remained high between A.D. 1225 and 1260, and that the resident population declined slightly during the final decades of occupation, between A.D. 1260 and 1280.

68
A more detailed model for the occupational history of Yellow Jacket Pueblo can be developed by examining the spatial and temporal distribution of architectural blocks occupied during various time periods. Figure 9, Figure 10, and Figure 11 summarize the spatial distribution and relative intensity of occupation in tested architectural blocks during three stages of development: A.D. 1060–1140, A.D. 1180–1260, and A.D. 1260–1280. Blocks that were not tested by Crow Canyon, including structures on the southeast talus slope and those on the uplands outside the Archaeological Conservancy boundary, are outlined but not labeled on these maps. The relative intensity of occupation in each architectural block during these three stages was determined by averaging the posterior-probability values for each block over the phases in each stage, and ranking these values in decreasing order of magnitude. Darker shading indicates a higher-ranked intensity of occupation. It is likely that, for the most part, the architecture of each block reached the form and extent recorded on the site map during the stage in which it was occupied most intensively. However, no attempt has been made to estimate the number and arrangement of architectural features occupied during other stages. Thus, the current model simply estimates times and places where sherd deposition resulting from occupational activities occurred.

69
Although the available data offer only a broad-brush view, they suggest several trends in the historical development of Yellow Jacket Pueblo. First, the longest-occupied architectural blocks (Blocks 100–700) cluster along the central, north-south topographic "spine" on which the village developed. Second, early occupation at the site, between A.D. 1060 and 1140, appears to have consisted of several habitations spaced along this spine, including areas on the talus slope near the spring at the confluence of the two drainages that define the "point" on which the site is located. The Chaco-era great house and great kiva at the northern end of the site were also probably constructed during this period. But, unfortunately, these features are located in an area of the site that we were not granted permission to test, so we do not have comparable pottery data with which to support this hypothesis.

70
Third, the occupation of Yellow Jacket Pueblo was most extensive and intensive between A.D. 1180 and 1260. During this period, the area along the central spine filled in, and additional roomblocks were constructed east of the spine and on the talus slope below the southwest canyon rim. For the most part, the occupations of these more-peripheral roomblocks were shorter than those of the central roomblocks. Finally, during the last 20 years of occupation, roughly A.D. 1260 to 1280, the village population either declined or coalesced into fewer architectural blocks (Figure 11). Several peripheral roomblocks were abandoned, and the settlement contracted back toward the central spine. In addition, a major new construction, the great tower complex (Block 1200), was undertaken along the canyon rim at the northeast edge of the village. Kuckelman and Ortman incorporate these pottery dating results with architectural and stratigraphic information in a more synthetic discussion of the chronology of Yellow Jacket Pueblo in "Chronology."

Pottery Rim-Arc Analysis

71
Rim-arc measurements were made on a sample of rim sherds from white ware bowls and corrugated gray jars from Yellow Jacket Pueblo. The radius of the parent vessel from which each rim sherd derived was estimated by placing the rim face down on simplified radial graph paper and comparing its curvature to a series of nested circles drawn with radii at 3-cm intervals. Radius Interval 9 encompasses radii that lie somewhere between 6 and 9 cm, Interval 12 encompasses radii that lie somewhere between 9 and 12 cm, and so on. The degrees of arc encompassed by the sherd was also estimated to the nearest 5 degrees, using the upper boundary of the interval as a guide. In the analysis of bowl rims, an attempt was made to identify sherds from the same vessel, and in such cases, only the largest sherd from each vessel was included. No similar attempt was made for rims from corrugated jars. However, because of the much smaller number of rim sherds generated from a single corrugated jar (compared with the number of rim sherds generated from a single bowl), it is reasonable to assume that few of the jar rims analyzed derive from the same original corrugated vessel. Thus, it is reasonable to assume that these rim-arc data reflect rim-radius distributions for samples of vessels rather than for samples of sherds per se.

72
Analyses of reconstructible vessels from Sand Canyon Pueblo, a late Pueblo III village located approximately 15 km southwest of Yellow Jacket Pueblo, indicate that, in general, rim diameters and volumes are correlated (Ortman 2000*2:par. 44–46). This holds true for both white ware bowls and corrugated gray jars. These analyses also suggest that the rim radius, or the radius measured at the rim, of broken vessels can be estimated with an acceptable degree of accuracy through rim-arc analysis (Ortman 2000*2:par. 49). Thus, distributions of rim-radius estimates should reflect the size distributions of white ware bowls and corrugated gray jars used and discarded at a site.

73
Rim-arc data were collected for white ware bowl rims from contexts that could be assigned to one of two time groups on the basis of architectural, stratigraphic, tree-ring, and pottery evidence. Sherds from the great tower complex (Block 1200) can be securely dated to the final decades of occupation, between A.D. 1260 and 1280; sherds from Blocks 700, 2200, 2600, and 3200, and from Structure 903, date from earlier periods, primarily between A.D. 1100 and 1225. These two groups of sherds are labeled "Late" and "Early," respectively.

74
In addition, every corrugated jar rim that was large enough to be measured was analyzed. Pottery dating results were used to place these sherds into four context groups. Rims from architectural blocks that were occupied for 100 years or more, including the final decades of occupation (Blocks 100–600, and 2000), were placed in the "Central" group. These blocks are all located along the central, north-south "spine" of the village. Rims from architectural blocks with occupations dating before A.D. 1180 ( Blocks 700, 2400, 2600, 3200, and 3300) were placed in the "Peripheral Early" group; these blocks are scattered throughout the site. Rims from architectural blocks with occupations that date from A.D. 1180 and later (Blocks 800, 900, 1000, 1100, 2100, 2200, 2300, 2500, and 3400) were placed in the "Peripheral Late" group; these blocks are located in various areas of the site, but most are not along the central spine. Finally, rims from Block 1200, the one block occupied between A.D. 1260 and 1280 only, were assigned to the "Great Tower" group. Comparing the rim-arc data for white ware bowls and corrugated gray jars in these time and context groups allows us to examine changes in food preparation and serving activities over the life of the village.

75
Figure 12 and Figure 13 present the results of rim-arc analysis for these various time and context groups at Yellow Jacket Pueblo. Rims that encompass fewer than 20 degrees of arc are excluded from these summaries because the accuracy of rim-radius estimates derived from such sherds is questionable (Ortman 2000*2:par. 49). The total degrees of arc assigned to each radius interval, rather than the count or weight of sherds assigned to each radius interval, is used as the measure of abundance. This compensates for the tendency of smaller-diameter vessels to break into fewer rim sherds that encompass more degrees of arc than do larger-diameter vessels (Pierce and Varien 1999*1).

76
The white ware bowl data in Figure 12 show that the distribution of radii in the Early group has a single mode, whereas that in the Late group has two modes, at the 9- and 15-cm intervals. The Late group also contains more large vessels than does the Early group. These differences suggest that more large bowls, and bowls of two distinct sizes, were used and discarded during the later years of occupation at Yellow Jacket. Such changes in bowl size over time probably relate to changes in the nature of meals served in the village. Rim-arc data from sites in the Sand Canyon locality (Ortman 2000*2:par. 53–54; Ortman and Bradley 2002*1) and from Woods Canyon Pueblo (Ortman 2002*1) exhibit similar patterns.

77
Cowgill (1990*2:68) argues that a bimodal size distribution for a group of artifacts likely reflects the existence of a conceptual distinction between large and small versions of that artifact type among their makers. Following this argument, it is likely that bowls discarded at the great tower complex during the final years of occupation at Yellow Jacket were conceived of as having two distinct sizes. Additional analysis will be necessary to determine whether this distinction was prevalent throughout the village during the final decades of occupation or was restricted to bowls used and discarded at the great tower complex only. Because bimodal size patterns are apparent in late Pueblo III assemblages from Sand Canyon, Castle Rock, and Woods Canyon pueblos, however, it is likely that similar food-serving and food-consumption practices characterized several late Pueblo III villages in the central Mesa Verde region.

78
Previous analyses of corrugated gray jar rims have also identified changing patterns in vessel size associated with the development of late Pueblo III villages. Data from sites in the Sand Canyon locality suggest that more large-volume cooking pots were used and discarded at Sand Canyon Pueblo than at smaller villages and earlier hamlets in the locality (Ortman 2000*2: par. 57). Because the sizes of households do not appear to have changed over time, these data suggest that more communal meals were prepared and consumed at Sand Canyon Pueblo than at smaller Sand Canyon locality sites (Ortman 2000*2; Ortman and Bradley 2002*1). Rim-arc data for corrugated jar rims from Yellow Jacket Pueblo (Figure 13) also duplicate these results. The rim-radius distributions for corrugated jars in the Peripheral Late group and for jars in the Great Tower group clearly show that more large-volume corrugated jars were used and discarded in these areas than in roomblocks in the Central and Peripheral Early groups. Assemblages in the Central group contain a mixture of vessels deposited early and late in the village's history, and thus they are not suitable for illustrating time trends. Nevertheless, comparison of rim-radius distributions for the remaining three context groups suggests that more large meals were prepared and consumed at Yellow Jacket during the final century of Pueblo occupation, when the site was a large village, than in earlier times.

Analysis of Pottery Rim Form

79
Classifications of corrugated gray ware pottery devised for chronological purposes derive from the observation that rim eversion in corrugated jars increased gradually over time. Wilson and Blinman's (1999*1) classification includes three types, as follows:

• Mancos Corrugated, for rims with eversion of less than 30 degrees (most common between A.D. 1025 and 1100)
• Dolores Corrugated, for rims with eversion between 30 and 55 degrees (most common between A.D. 1100 and 1180)
• Mesa Verde Corrugated, for rims with eversion greater than 55 degrees (most common between A.D. 1225 and 1280)

The classification system used by Crow Canyon, in contrast, recognizes only two types: Mancos Corrugated, for rims with eversion less than or equal to 30 degrees, and Mesa Verde Corrugated, for rims with eversion greater than 30 degrees. Dolores Corrugated is, in effect, subsumed by Mesa Verde Corrugated. In both systems, the degree of eversion is estimated visually by holding the sherd as it would have been oriented in the original vessel and examining its profile. Crow Canyon uses the simpler, two-type system to minimize interobserver variation in typing. An unfortunate result of this system is that most corrugated rim sherds deposited at Pueblo III sites are classified as a single type, Mesa Verde Corrugated, and thus are of little value for relative dating arguments.

80
Because variation in rim eversion in corrugated jars is continuous, an alternative to typological classification for assessing the chronological value of sherds from such vessels is to measure the eversion angle directly. This has been attempted previously using samples of corrugated rim sherds from Woods Canyon Pueblo and the Yellow Jacket great tower complex (Ortman 2002*1). Results of this previous study suggest that it is difficult to calculate eversion angles directly from rim measurements, but that both the horizontal width and the diagonal length of rims on corrugated jars increased over time. This suggests that time-sensitive variation in corrugated jar form might be captured more simply by measuring the curvilinear distance from the inflection point marking the minimum orifice diameter of the vessel out to the edge of the rim. A flexible ruler, graduated in fiftieths of an inch, was used to record this measurement for all 257 corrugated jar rims in the Yellow Jacket assemblage on which both the rim edge and the inflection point perpendicular to and below this edge (marking the minimum orifice diameter) could be identified.

81
Figure 14 uses box plots to summarize the distribution of these flare measurements across four vessel-size categories and across the four context groups defined in paragraph 74. The four size categories are as follows: small (radius intervals 3 and 6), medium (radius interval 9), large (radius interval 12), and extra-large (radius intervals 15 and 18). For each distribution, the shaded box represents the midspread (middle 50 percent of cases); the thick, horizontal line inside the box represents the median; and the tails illustrate the range of values, up to 1.5 box lengths from the edges of the box. Outliers are excluded from these charts.

82
Figure 14 shows that the flare of corrugated rims is associated to some degree with both vessel size, as estimated by rim-arc analysis, and time, insofar as it is captured by the four context groups. The midspread of flare measurements increases gradually with vessel size, and the median flare measurement increases gradually across the four context groups. In contrast, evidence that variation in rim-flare measurements is associated with occupation span is equivocal. The midspread of measurements for the Central group assemblage, deposited over a two-century period, is actually less than that for the Great Tower group assemblage, which was deposited over the span of a few decades. And although the range of flare measurements is greatest for the Central group, this group also has the largest sample size.

83
It is therefore clear that some of the variation in flare measurements results from the fact that larger vessels tend to have larger rims that will produce larger measurements somewhat independent of the shape of the rim. Because rim-arc data suggest that late assemblages from Yellow Jacket Pueblo contain more large corrugated vessels than do early assemblages, one might expect sherds from these late assemblages to possess larger rim measurements overall. However, the pattern of increasing "flare" through time appears to hold even when vessel size is taken into account. Figure 15 shows that, for the most part, the flare of corrugated jar rims increased through time within individual size categories. Thus, the increase over time in the degree of rim flare in corrugated jars observed in this dataset appears to reflect both an inherent increase in the flare of rims and an increase in the number of large vessels used and deposited at the site. Additional studies using unmixed, single-component assemblages may be necessary to identify chronological patterning in corrugated vessel form more clearly.

Modified and Shaped Sherds

84
A number of modified sherds and shaped sherds were collected during excavations at Yellow Jacket Pueblo. Table 18 summarizes the pottery types of these sherds by count and weight and also presents relative frequencies of different types by count and weight. Modified sherds possess at least one abraded edge. In some cases this modification may have resulted from scraping wet clay during pottery making. However, no attempt was made to identify pottery scrapers among the modified sherds recovered from Yellow Jacket Pueblo. Shaped sherds have edges that were flaked, ground, or both, to make a specific shape. Some larger shaped sherds may be pottery fragments that were used as containers (called "sherd containers" in Crow Canyon's analysis system) or as pottery-molding trays called pukis. Perforated sherds with shaped edges were classified as sherd pendants and are discussed in paragraphs 155–156. Sherds with shaped edges but lacking a perforation, such as disks, triangles, and rectangles, were classified as shaped sherds and are included here. These shaped sherds may have been pendant blanks, gaming pieces, or other nonutilitarian items.

85
Table 19 summarizes modified and shaped sherds according to the ware and form of the parent vessel for each piece. A comparison of the percentages by count and weight shows that relatively fewer modified or shaped sherds came from corrugated vessels, but that these sherds tend to be larger than the modified or shaped sherds from vessels of other wares. Corrugated sherds are not well suited for use as pottery scrapers, because they have uneven surfaces, coarse paste, and large temper inclusions, all of which make it difficult to create a smooth scraping edge. However, several complete corrugated sherd containers have been found in excavations at other sites, such as Sand Canyon Pueblo. Most modified and shaped sherds at Yellow Jacket Pueblo are from white ware vessels. White ware jar sherds were often of an appropriate size, shape, and curvature for use as pottery scrapers. In contrast, sherds from white ware bowls were typically more intricately painted and finely finished and, therefore, were more suitable for shaping into gaming pieces or pendant blanks.

Pottery Vessels

86
Seventeen whole, partial, or reconstructible vessels were collected from various contexts at Yellow Jacket Pueblo. Ten of these are white ware bowls, four are corrugated jars, one is a lid from a kiva jar, one is a mug, and one is a sherd container made from the base of a corrugated jar. No nonlocal vessels were found.

87
Vessel type, form, and condition are recorded during analysis. In addition to these data, specific vessel measurements and information about archaeological context are presented in Table 20 and Table 21, respectively. If a vessel has been reconstructed, you can click on its vessel number in Table 20 or Table 21 to see a photograph of it. Vessel numbers 11–18 were assigned to pottery vessels from the excavations of "Square Mug House" (that is, the great tower complex) by the Museum of Western State College in 1931 (Hurst and Lotrich 1932*1) and are not considered here.

Pottery Production and Exchange

88
This section summarizes the direct and indirect evidence of pottery production at Yellow Jacket Pueblo and examines the intraregional networks of pottery exchange in which Yellow Jacket participated. Evidence of long-distance pottery exchange is presented in the discussion of objects of nonlocal materials (paragraphs 157–159).

Direct Evidence of Pottery Production

89
Direct evidence of pottery production in the Yellow Jacket Pueblo assemblage is summarized in Table 22. Such evidence includes manufacturing tools such as polishing stones, raw materials such as pottery clay and temper, and unfinished vessels such as unfired sherds. A fourth potential type of direct evidence is pottery scrapers made from sherds. Although pottery scrapers have been collected from other sites in southwestern Colorado (e.g., Wilson 1988*2:Table A.6), no attempt was made to identify them in the Yellow Jacket assemblage.

90
Prepared clays suitable for use in pottery making were found in several locations at Yellow Jacket Pueblo. Igneous rock samples are also considered to be raw material for pottery manufacture because this material was not often made into stone tools, but was often ground for use as temper in white ware and corrugated gray ware vessels. Because the closest major source of igneous rock is the Dolores River valley, approximately 10 km to the east, this material must have been brought to the site by humans.

91
Polishing stones are small, very smooth, and very hard stones or pebbles that exhibit evidence of abrasive wear. Most polishing stones from Yellow Jacket are of high-quality, fine-grained stone, including cherts, quartzites, and agate/chalcedony. Although some of these stones might have been found locally, many others were of rare materials that required some effort to procure. Traces of clay found adhering to such stones from other sites indicate that at least some polishing stones were used for polishing the surfaces of unfired vessels during manufacture. Because surfaces of corrugated gray ware vessels were not polished, polishing stones constitute direct evidence of white ware manufacture only. It is unknown whether polishing stones had other uses.

92
A number of clay objects that were clearly not pottery sherds were found at Yellow Jacket. A few may have been appliques attached to pottery vessels, but several appear to be remnants of manipulated pottery clay, leftover from pottery making. Some of these objects had been fired. Because it is unlikely that such items would have been traded, many probably represent by-products of pottery manufacture.

93
Previous studies (e.g., Ortman 2002*1; Pierce et al. 2002*1) have suggested that production of corrugated gray ware pottery might have been more specialized than production of white ware pottery in the central Mesa Verde region. This inference is based on (1) the extreme rarity of unfired corrugated sherds; (2) the absence of coarsely ground clay samples with large, chunky temper; and (3) the more widespread distribution of igneous-tempered corrugated vessels than igneous-tempered white ware vessels. No attempt was made to determine whether the pottery clay samples and "other ceramic artifacts" from Yellow Jacket are of white ware or corrugated gray ware pastes. However, the presence of igneous rock samples in the assemblage does suggest that potters living in the village had access to this temper material. Thus, direct evidence of pottery making at Yellow Jacket does not indicate whether corrugated gray ware vessels were produced in the village. Indirect evidence, in the form of temper materials incorporated into white ware and corrugated gray ware vessels deposited at the site, is discussed in paragraphs 98–112.

94
Although the available data are insufficient to determine the kinds of vessels produced in the village, the amount and distribution of direct evidence for pottery making across architectural blocks can be used to examine the overall organization of pottery production at Yellow Jacket. If pottery making took place throughout the village, direct evidence of such activity should be present in each architectural block. On the other hand, if pottery production was centralized in certain areas, one would expect the direct evidence of pottery making to be more abundant in these areas.

95
Direct evidence of pottery making at Yellow Jacket Pueblo is tabulated by architectural block in Table 23. The total weight of corrugated gray sherds recovered from each block is also presented in the right-hand column as a measure of sample size. This figure can be used to standardize pottery making against cooking, a routine daily activity. Because Yellow Jacket Pueblo received only limited testing, it is possible that there are concentrations of direct evidence in contexts that were not excavated. Nevertheless, direct evidence of pottery production was found in 15 of 22 sampled architect