Correspondence Analysis and West Mexico Archaeology: Ceramics from the Long-Glassow Collection

4: Ceramic Analysis

145 Chapter Four Ceramic Analysis By C. Roger Nance, Jan de Leeuw, Kathleen Prado, and David S. Verity Las Cuevas vvvOur quest to determine the ceramic sequence for the Etzatlán region began with research on the relatively large collection from the site of Las Cuevas. Here, Long and Glassow had excavated sixteen pits scattered over a relatively flat promontory of land that extended into the lake on the eastern side of the island. We processed about 7,800 potsherds from these sixteen units, classifying and physically sorting the sherds into sixty-five different types. We also added a unique specimen number to each sherd. Next, we recorded the data in notebooks, listing for each sherd its type, unique ID number, and provenience (square and depth). Finally, we entered the data into a computer and then compared computer printouts to handwritten data, line by line, to eliminate data entry errors. When we attempted to study type distributions in a traditional way, looking for stratagraphic changes through time, level by level, the data appeared confusing—trends in some units did not appear in or were even contradicted by data from others. Part of the problem was that we were dealing with part of a large open site that had been intensively occupied. Deposits with cultural debris tended to be shallow with 94.9 percent of all potsherds from Las Cuevas excavated from deposits less than 1 m deep. Also, we found ample evidence of disturbance: field Chapter four / 146 notes for the Las Cuevas excavation record fragmentary burials, a rock oven (?), rock concentrations, intrusive pits, and a rock wall foundation . As a result, we took a statistical approach and concentrated at Las Cuevas on data from fifty-two lots of potsherds (samples from individual square levels) with more than fifty sherds each. We also concentrated on those types present in large frequencies in order to reduce the problem of sampling error. Our approach was to look for covariation of types among the samples with the idea being that types, which covaried closely, would tend to be contemporaneous. Initially, we used the correlation coefficient in an explorative way to discover this covariation among important or high frequency types. At this stage of the Las Cuevas analysis, we were able to discern two type complexes. Each group was defined by positive correlations among some of these types, while negative correlations separated these same two complexes on a typeto -type basis. Also, we had some reason to believe that one complex was earlier than the other, given slightly different average levels for the two groups of types in the deposits. The types in each of the suggested groups are listed in table 4-1. In order to increase sample sizes, in several cases we combined sherd data from related types for the analysis. Specifically, sherd data for Huistla Polychrome grater bowl types (7, 8, and 9) were grouped, as were data for Huistla Polychrome non–grater bowl types (19, 20, and 21) and White on Red types (16, 17, and 18). Morphological differences among these individual types are described in chapter 3. Unfortunately, we had been forced to use type (or combined type) percentages by lot in constructing the correlation matrix in order to obtain any patterning at all. This practice of employing the correlation coefficient with percentages instead of frequencies is considered Table 4-1: Two provisional ceramic complexes Early Complex Late Complex Incised Polychrome Huistla Polychrome Gray Slipped and Polished Huistla Polychrome Grater Bowl White on Red Brown Slipped and Polished Comal C. Roger Nance, Jan de Leeuw, Kathleen Prado, and David S. Verity / 147 White on Red Incised Polychome Grey Slipped, Polished Comal Huistla Polychrome Huistla Polychrome Grater Bowl Brown Slipped, Polished -2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 Figure 4-1. Type plot, original Las Cuevas CA. to be statistically unreliable. Our initial effort, however, did show some potential for a seriated sequence of ceramic types at Las Cuevas and indicated how those types might figure in such a sequence. We turned next to a different statistical treatment of these data, correspondence analysis (CA). This procedure uses the chi-square statistic to measure the degree of covariation among both types and samples and maps the results on a two-dimensional grid. CA is described and its archaeological...