For a long time, social sciences have been criticized for not being “scientific” enough. Archaeology as a branch of social science has unavoidably suffered the same kind of accusation; however, being interdisciplinary in nature, archaeology is also considered to provide a way of integrating hard sciences into social sciences and humanities. The research questions archaeologists ask need to be answered based on hard evidence acquired from scientific excavations or surveys, and they apply principles borrowed from natural sciences such as stratigraphy (borrowed from geology) and sampling rule (from mathematics). Since the 1960s, proponents of New Archaeology have held that archaeologists need to go beyond past traditions of cultural-historical research by sampling and classifying archaeological materials in a manner subject to repeated tests. Even though the questions archaeologists ask might involve complex theoretical concepts, and even if the so-called evidence itself often leaves room for interpretation, the proposed solutions and inferences need to be examined in a scientific manner. Such an approach is necessary because it provides a basis for communication between the social and natural sciences. Furthermore, it has led to a revolution in the methodology of the discipline. That said, the new approach has not only changed the way we solve research questions, but even how we ask them in the first place.
Today, expressing concern about science and numbers has become general practice and applications of statistics are no longer limited to mere counting of artifacts. However, numerous jargons together with dazzling tables and plots have also created barriers for the audience. The introduction of new methods and techniques also leads to the re-emergence of an old problem in archaeology, the logic of classification (or grouping of things): Are the categories discovered with the help of statistical methods really the ones that would have mattered to the ancient humans, or are they a mere modern invention? A major worry is that statistical methods could hide some important patterns while highlighting irrelevant ones, thus doing the opposite of what we hope to gain from the application of statistical methods. This difficulty cannot simply be overcome by the enhancement of statistical complication; instead, we first have to reconsider the purposes of classification. After all, an important goal of searching for patterns in material remains is to discover the emic categories in the minds of the past people who left these remains. In this way, we hope to understand how past people perceived and responded to their surroundings.
Many Patterns Means No Patterns
Although it is expected that classification conducted by archaeologist should reflect the inner patterns of the cultural system under analysis, this goal is nearly impossible to reach, because the artifacts we encounter are themselves the products of crafters’ interpretation of the real world and not its unfiltered reflection. No one is omniscient or has the full picture of a culture, be it their own or a foreign one. What we see in the archaeological record are samples of samples instantiated by ancient artisans, who themselves rarely spelled out their thoughts in manufacturing, at least not in a direct fashion. It can be a pitfall that the patterns we find are indeed what we expected to discover, but they only represent one of the many possible patterns and interpretations of the material record at hand. Regardless of sample size, one can always find a pattern and explain it in many different ways. For instance, when we conduct cluster analysis, we need to decide in advance how many clusters we expect to obtain. This assumes that the analyzer already knows the number of clusters and is clear about what constitutes the criteria to divide the objects under investigation. However, such information per se is not self-evident and requires prior knowledge. When this kind of knowledge is lacking in the first place, we do not know which feature(s) to choose to suggest grouping, so classification cannot proceed and we cannot acquire the information required to go any further in classification. This can be a “double bind” dilemma (Dwight Read. “The Substance of Archaeological Analysis and the Mold of Statistical Method: Enlightenment out of Discordance? In Christopher Carr, ed. For concordance in archaeological analysis: bridging data structure, quantitative technique, and theory. Kansas City, MO; Fayetteville, AS: Westport Publishers; Institute for Quantitative Archaeology at the University of Arkansas, 1989).
Discovery or Invention? The Philosophy of Classification
The dilemma arises because patterns in material remains are meaningful only if they are suggested by culturally salient features. To ascertain if a feature is culturally salient, we need to investigate the social contexts of the objects in question. We therefore need to use social and cultural information to recursively discover artifact categories. Such information, including context of manufacturing and use of artifacts, is relevant because, as stated above, we want to avoid imposing our own categories on ancient contexts, such creating “empirical types” based on our own experiences and serving our own ideas and convenience. Even if such empirical types might be efficient in grouping, they do not conform to past categorization and are not culturally meaningful in this sense.
The dilemma can also be illustrated by differentiating classification or categorization from clustering. While the former refers to the situation in which analyzers have no idea about any category in advance, clustering means to assign objects to given groups that are already known. In some situations, we can create temporary categories by grouping artifacts at hand and modify these categories after we have more data or information. However, even if such categories are only temporary and allow changing thereafter, the criteria used to define them can still influence the classification because they create the framework for all following classification attempts. By now, most archaeologists are aware that there is always more than one way to classify things and create groups. To determine which methods to use, one relies on one’s experience, purpose, and philosophy of classification. Such methods do not need to be quantitative but can sometimes be very intuitive or be illustrated simply by the distributional modes of particular features of a group of objects.
For instance, when studying pottery types used in Bronze Age Sichuan in Southwest China, I found that archaeologists have used the popularity of different shapes of point-bottomed saucers in different periods as time markers. However, if we take metric measurements on their dimensions, such as rim diameter and height, and conduct basic statistical analyses on them, it becomes clear that the vessels’ height comprises more than one mode even during the same period. Potters living in certain loci of the site under study produced short saucers while the others made higher ones. To discover such modes does not require complex statistical techniques but it is enough simply to look at the “error bars,” which plot standard deviations of the measurements along with their means. When doing so, I found that there are at least two modes of height clearly distinct from each other, indicating craftsmen’s intention to make two different types of saucers. This finding of modes should trigger further investigation into why two types coexisted, which requires the supplement of other information such as population movement or replacement. Vessel types defined in this way contain and reflect only a piece of the cultural picture and need to be combined with other pieces.
Final Remarks: The Simpler, the Better
In the example cited above, some might argue that the factor time could still relate to the change in vessel shape; furthermore, time and location might be correlated to each other as well. It becomes even more complex when we add other factors (e.g., contexts of use), with some of them interacting with each other. The list of such factors possibly relating to change in vessel shape is indeed endless, as are the factors influencing certain human behavior. To learn how these factors or variables interact and what mixing effects they have can be exhausting even with powerful tools of multivariate analysis. Reducing the number of factors and ascertaining which of them are really important is as critical as suggesting new factors. But ironically, to meet the need of variable reduction, multivariate analyses themselves often contain complicated mathematical principles and may require prior fitting tests on the properties of the input data. Only when the researcher is clear about the properties of the data and the goal of the analysis can he or she begin to set each analytical and decision-making step, and then explain the consequent patterns properly. The subsequent explanation is just as important as initial analysis. Although it is a myth that the use of statistical techniques is a special skill that only people versed in mathematics can handle, it should still be recognized that when using increasingly complicated statistical methods, one may create more barriers for the reader. Only if such methods help to reduce the complexity of the data structure or when they can make certain patterns clearer, is it useful to apply statistical methods. Otherwise they could create more work and more confusion instead of helping to solve problems.
Institute of History and Philology
Academia Sinica, Taiwan
Image: created by the author.
The views, perspectives, and opinions expressed here and by those providing comments are those of the author(s) and commentator(s) alone, and do not reflect the opinions of Dissertation Reviews, its members, editors, or advisory board members.