You are on page 1of 38

Statistical Methods and Reasoning in Archaeological Research: A Review of Praxis and Promise

DWIGHT W. READ
DepMment ofAnthroporogy

UCLA
Los Angeles, CA Pa024 U.S.A.

INTRODUCTION

In 1959 the symposium, The Application o f Quantitative Methods in Archaeology, was held at Burg Wartenstein. Despite its title, only a single paper, by Albert Spauldiig (1960), discussed statistical methods. Spaulding's paper, 'Statistical Description and Comparison of Artifact Assemblages,' set forth the thesis that archaeological analysis can be extended effectively through reasoning based on statistical methods. What statistics offered, Spaulding suggested, was not merely a collection of methods applicable to archaeological data, but a way to represent and express more exactly the ideas of archaeology: hence to make these ideas amenable to more precise reasoning - much as he had accomplished with his oft repeated definition of a type as the non-random association of attributes (Spaulding 1953). Statistics as a means to reason about quantitatively expressed information will be the organizing theme of this review. Statistics in this sense contrasts with statistics viewed as methods to be applied to data. The former, as Spaulding indicated, expands upon, and can become part of, archaeological thinking; the latter primarily provides additional information to be used in archaeological arguments. But statistics taken as providing a means to represent archaeological ideas has not, as will be seen, been the dominant perception of what constitutes statistical applications in archaeology. Instead, statistics as a means to represent and reason about ideas and concepts has played a role whose potential is yet to be realized fully. Though Spauldii's was the only confetence paper that made use of statistical concepts, substantive application of statistical methods to archaeology had begun much earlier. In 1950 the book, Some Applications of Stalistics to ArchneoIogy, by Oliver Myers was published in Cairo. Myers' work was evidently unknown to Spaulding for he commented (1960: 82) that he was unaware of any application of regression methods i n archaeology, yet Myers made extensive use of regression and correla-

DWIGHT W. READ

METHODS IN ARCHAEOLOGICAL RESEARCH

tion analysis in an analysis of the surface materials from a series of Egyptian sites. Whereas Spaulding emphasized the idea of developing archaeological ideas with statistical concepts, Myer's work was a forerunner (though apparently without descendant) of what today constitutes the bulk of quantitative and statistical applications in archaeology. Myers was concerned with using statistical methods to confirm or disconfirm several hypotheses about the covariance of pottery sherds and flint in surface collected sites as a way to determine the chronological or cultural affinity of the sites. Myers commented that "if statistics be fully applied to archaeology, it will be possible not only to draw new knowledge from the clues we now use, but to obtain information from sources which have never been considered as evidence" @. v). Though not a statistician - he admits to his lack of mathematical background - he hoped that "this admission will encourage other unmathematically minded archaeologists to make use of such [statistical] methods" @. v). The methods he used comprise most of the techniques introduced only much later into American archaeological research through the emphasis on quantitative methods which became associated with the "new archaeology." These included computation of correlation coefficients (correlation between flint and sherd frequencies), curviIinear regression analysis (relationship between sherd thickness and maximum dimension), analysis of variance and covariance (significance of the regression model for thickness and maximum dimension; independence of the regression model f the sherd measured on the Mohs' scale as a proxy with hardness o measure for likelihood of destruction through time), density contours (analysis of the spatial distribution for each type of sherd and flint), a probability model (to determine whether or not two types of sherds could have had essentially identical spatial distributions by chance through two separate occupations), simulation (to determine the optimal grid square size for recording flint and sherd frequencies per grid square), and deductive reasoning (expected pattern of correlation amongst the several types of sherds and flints according to the hypothesized sequence of settlements for the area). The book is remarkable not only for its early insights into the manner in which statistical methods can address archaeological problems, but also because it had no impact on the field despite careful work that currently could serve as a textbook example of statistical applications. Curiously, a paper that did have the effect of galvanizing archaeologists ' into awareness of the enormous potential of statistical methods is also a paper that, if anything could serve as an example of incorrectly applied statistical methods. The paper, "A Preliminary Analysis of Functional Variability in the Mousterian of Levallois Facies", by S. Binford and L. B i o r d (1966), used factor analysis in an attempt to determine an

underlying, hypothesized, process accounting for the observed frequency distribution of artifacts in sites. On the one hand, the authors argued from the archaeological side about how assemblages might be structured through the way in which tools are organized as sets of tools (the "toolkit") in the performance of tasks. On the other hand, they suggested that one could, apparently, use statistical methods to recover the posited organization from frequency counts of artifacts on sites. With this paper, possibility seemed to become reality; no longer was the archaeologist limited to making subjective inference about past processes, but could, it seemed, objectively determine that reality from the patterning found in data by the archaeologist. Spaulding's paper had given American archaeologists the intellectual justification for using quantitative methods as a basic part of archaeological reasoning; Binford and Binford's paper seemed to show that the potentid of what could be achieved with quantitative methods was limited only by the creativeness of the archaeologist. A similar theme was also established in England through the publication in 1968 of David Clarke's magnum opus, Ana&ical Archaeology. Clarke went beyond the arguments of Spaulding, and Binford and Biiford and laid out an entire program for what he called "analytical archaeology." It had the goal of "the continuous elucidation of the relationshius which Dermeate archaeoloeical data bv means of disciolined directedatowardsthe precipi&tion of a gody of general &wryn (pp. 662-433. Unlike the "new archaeology," which e m p h a s i i archaeG16gy as a science, Clarke argued that " h y t i c a l archieology is not a science but it is a discipline, its primary machinery is mathematical rather than scientific" @. 663). By this Clarke meant that archaeology is based on probabilistic, not certain, regularities, and only the latter are the hallmark of a science. T h i s is not a drawback, but a difference in focus: 'The relationships which analytical archaeology may hope to elucidate q e . . . [the] domains of archaeological syntactics, pragmatics and semantics" @. 663). These relate to "cultural systematics, cultural ethnology, and cultural ecology," respectively, and "the relationships within each domain form independent grammars .. . [that] may eventually be expressed in a symbolic and axiomatic calculus" @. 663) - hence the linkage of archaeology with mathematics. Finally, Clarke observes: "it looks as though the 'observer language' or meta-language of archaeological syntactics will be the first grammar to uncover a calculus of symbolic expression over the next few decades" @. 663). The beginn@ of such a calculus "are very largely the result of the developing contributions of quantification, statistical procedure, and the impact of the computer" @. 651). Clarke's argument canies the idea underlying Spaulding's use of statisf tical ideas as part of archaeological reasoning to its logical conclusion I the proper role of statistics, and by extension, mathematics, is one of

DWIGHT W. READ

METHODS IN ARCHAEOLOGICAL RESEARCH

reasoning and not just of providing techniques, then it follows that there should be formaVmathematicalrepresentation of the structuring ideas and relationships upon which archaeological reasoning and theory is based. The form of such a representation would be a calculus through which the logic of those ideas and relationships could be developed. Though this is a theme that has been taken up by others as well - e.g., Gibbon (1984: 352-83) discusses the need for formalized (axiomatic) theory, and Read and Leblanc (1978) give an example in a restricted domain of what an axiomatically expressed theory might look like - Clarke alone has set forth a grandiose, programmatic statement. In these four works by Spauldimg, Myers, the B i o r d s and Clarke, we find the two main threads that will be used here to categorize topics that have been taken up extensively in the last several decades by researchers trying to derive more refined information from archaeological data through quantitative methods. The two threads are: (1) reasoning about, and representation of, archawlogical concepts using the conceptual frameword provided by statistics and mathematics (eg., Clarke 1968, Spaulding 1960) and (2) statistics as a series of methods to be used in the analysis of archaeological data (e.g., Myers 1950, Binford and Binford 1966) for specific goals. This review will also be divided into two parts. The first part will consider publications from four international conferences and one national conference on quantitative methods in archaeology. The focus will be on developing ideas about the role of quantitative methods as a reasoning system about archawlogical concepts. The second part will review publicati0,ns that make substantive use of quantitative methods and q u t i tatively based reasoning. Here the discussion will focus on the way statistical concepts translate into archaeological methods as a means to address archaeological issues. The distinction is not exact in that the former ultimately has to be expressed in tenns of methods and the later, implicitly if not explicitly, depends on the researcher's persuasion about the proper role of quantitative reasoning in archaeological reasoning. Selection of material to be reviewed is generally restricted to publications appearing after 1975. The date is arbitrary; it is used to keep the review reasonably bounded. To thb degree it was feasible, non-American ~ublicationswere included. but coverwe is uneven This iournal will hopefully promote increased communication among all archaeologists interested in quantification and mathematical m e l i n g of archaeological , problems. Papers that simply used numbers, or applied methods mainly as an adjunct to other arguments were not included. The latter were excluded because they would require a review focused more on the efficacy of the application of quantitative methods in archaeological research, such as the ones provided by Thomas (1978), Clark (1982) and Scheps (1982), and

less on quantitative methods as a means to reason about archaeological ideas. On the o t h e ~ side of the continuum from applied to theoretical, papers whose emphasis was mathematical rather than statistical were also not included as these are the subject of a forthcoming review (Read 1987a)PART I. CONFERENCES AND SYMPOSIA ON STATISTICAL AND QUANTITATIVE METHODS Subsequent to the Burg Wartenstein Conference a number of international and national conferences and symposia devoted to the application of quantitative, statistical and mathematical methods in archawlogy have been held. Publications from five of these conferences will be reviewed to illustrate changing perceptions of the role of quantitative and statistical methods in archaeology. The conferences, identified by location and the title of the conference proceedings, are: (1) Marhemafia in the Archeological and Historical Sciemes, Mamaia (Hodson, Kendall, and Fautu, eds. 1971); (2) Archiologie et Calculotews: Probkms Semilogiques, Marseille ( G a r d i ed. 1970): (3) R a k o m m n t et Me'thodes Mathimatiques en Arche'ologie, Paris (Borillo, Femandez de la Vega, and Guenoche, eds. 1977); (4) To Pattern the Past, Amsterdam (Voorrips and Loving, eds. 1985); and (5) Quantitative Research in Archaeology: Progress and Prospects, Portland (Aldenderfer, ed. 1987).
MATHEMATICS IN THE ARCHAEOLOGICAL AND HISTORICAL SCIENCES

Whereas the conference held at Burg Wartenstein had but a single paper that dealt with statistical methods, the Mamaia conference was structured to bring together archaeologistsihistorians and statisticians/mathernaticians viewed as representing two separate disciplines. Participants were instructed to address their papers m the practitioners of the other discipline (Moberg 1971). In effect, the conference took the view that statistical techniques should be introduced into archaeology through problem identification by the archaeologist linked to methods provided by the statistician(see also Thomas 1975). Spaulding, in his "Introductory Address" to the conference, made explicit his more implici* stated theme from the Burg Wartenstein symposium when he obser;ed that "these [stadstid] techniques merely -- the - im~licitmathematical reasoning that aU make explicit and extend archawl6gists usen (Spaulding i971: 15; emphasis in the- original). Spaulding's claim recalls Clarke's (1968) view that analytical archeology will need to develop a symbolic and axiomatic calculus to express the fundamental ideas and reasoning of the discipline. 'The validity of the first

DWIGHT W. R E A D

METHODS IN ARCHAEOLOGICAL RESEARCH

part of Spaulding's assertion, namely that statistical techniques can serve to explicitly express otherwise implicit arguments, is exemplified by the conference papers that addressed artifact grouping and seriation. Thus, Hodson (1971) argued that the archaeological interest in grouping handaxes for analytical purposes could be realized through the methods of numerical taxonomy. Similarly, Rowlett and Pollnac (1971) suggested that plotting factor loadings is a way to spatially group cemeteries from the Marne Culture in order to better understand those data. Both were trying to connect the archaeologist's analytical goals with what were thought to be appropriate quantitative techniques. Conversely, the statistician Rao (1971) discussed his work on decamping a vector X of p standardized measurements, with correlation matrix R, into a size, s, and shape, h, component v i a .
s

<'R -'X and h = <'R-'X,

so that one can form a two-dimensional plot of the data (using s and h) w i t h i n which one can search for "natural" groupings. Rao suggested that the method should have direct application to the handaxes discussed by Hodson. The pattern is repeated with seriation: the archaeological problem was cast in a manner that pennits interpretation in a mathematicdstatistical idio, and work on resoiving the more abstract formulation of the problem was brought to bear on the archaeological problem. The problem as formulated by Robinson (1951) was to find a permutation matrix, P, for a a t r i x with 0's similarity matrix, S (alternatively, an incidence matrix [a m and 1's representing absence and presence, respectively], or an abundance S matrix [a matrix whose entries are frequency counts]), so that R -- P is a Robiion matrix, i.e., a matrix with similarity values monotonically decreasing as one moves away from the principal diagonal. In an earlier formulation of the seriation ~roblem. Pehie (1899). whose extensive work graves' insp&ed Kendall's work on on chronological ordering seriation (e.g., 1963,1971), had suggested a solution based on minimizing sequence-dateranges (Kendall1971: 217). Analytical solutions are given: by Gelfand (1971) who discussed algorithms for rapidly &ding a permutation (since brute force trial and error is impossible when there are more than a few objects to be seriated), if one exists, that converts a similarity m a y to a Robinson matrix. Kendall (1971) discussed a method based on multi-dimensional scaling ' (the so-called 'horseshoe' method) for replicating a Petrie seriation, while W i n discussed the mathematical theory behind the latter method. Both Gelfand and WiUrinson gave necessary and sufficient conditions (Gelfand 1971: 194; W i n 1971: 277) for the structure of the similarity matrix S that ensures being able to find the desired permutation matrix, P.

Perhaps paradoxically, mathematical success for the seriation problem generated new archaeological problems: Do the data in fact satisfy the needed for a solution? Gelfand (1971) discussed this matter briefly by noting that "in addition to time, there may be geographic, cultural, or other dimensions . . . and a resultant estimated order may not actually represent the desired time dimension" @. 199, emphasis added). Here it is only a warning; as will be discussed below, this is a major problem facing application of statistical methods for constructing chronologiesfrom seriations. While these papers make explicit the link between archaeological and statistical ideas, they do not extend, in Spaulding's sense, the archaeological ideas. One of the few papers in the conference that does use mathematidstatistical language in this sense is the paper by Orton (1971) titled "Statistical Sorting and Reconstruction of Pottery." Orton states that the paper's "purpose is to restate an archaeological problem in more mathematical terms" @. 458) with the problem b e i i to "construct a statistical description of the production of the [Romano-British] kilns" @. 451) as seen through modeling the relationship between rim and base categories for the vessels produced in the kilns. Orton relates the number of rim forms of a given type from the vessels to the number of base forms from the same vessels as fo1lows. Let the number x,, of rim forms of type i in lot j (a lot is a level or feature in a trench) be related to the number of base forms of type h in the same lot j via:

ah

where p, is the proportion of vessels with rim type i that have base type hinlotj. The full set of equations for relating all rim types with all sherds in all lots can be given the matrix form as where: Y is the matrix of all terms, X is the matrix of all x, terms, P is the matrix of all p, terms. The matrix P is the quantity to be estimated from the data and the matrix equation may be solved to yield P (XIX)-'X' Y, assuming that X is non-singular (Orton 1971: 454). Orton notes the similarity between this model and the model for regression analysis and uses the latter to make the model more realistic by allowing for separate linear relationships between number of vessels and number of rim or base forms, as there may be differential breakage of rims and bases. Here, unlike the suggested application of Rao's technique to Hodson's handaxes or the use o f n u m e n d taxonomy procedures for constructing classifications. the mathematical construct has been built to match the archaeologist's perception of what the relationship is between the variables

Egypt&

12

DWIGHT W. R E A D

METHODS IN ARCHAEOLOGICAL RESEARCH

13

in question. This contrasts with the more common, implict argument by analogy that mathematical rigor will entail archaeological rigor (Read 1985; see also BoriUo 1977: 220).

Another conference on the role of mathematics in archaeology was held about the same time in Marseille under the auspices of the Colloque International du Centre National de la Recherche Scientifique. The conference was narrower in scope as it primalily addressed classification, but broader in approach as it focused on conceptualization of the problem of classification and not iust on mathematical/statistical techniaues. The papers in this conference highlight the philosophical split between realists who advocate methods aimed at the empirical level of patterns e.g., pattern recognition (Bordaz and Bordaz 1970), displayed in data cluster analysis and proximity analysis (Doran 1970), scalograms (Elisseeff 1970) and multidimensional scaling (Liigoes 1970) - versus rationalists who attempt to derive a theory of data structures - e.g., papers on automatic classification by Regnier (1970) Lerman (1970) and de la Vega (1979). Automatic classi6cation is not viewed in the more limited sense of numerical taxonomy, but as theory guided by the formal properties of a classification:

le nom de hslassificabilit6 nous dCsignerons I'aptitude d'une population E d'objets 6 h e organis& en u n e hiCrarchie de classifications (Under the name of h-classi6cation we designate the reasonableness for a population E of objects to be organized into a hierarchicaldassification .] (Leman 1970:319, translation added).
Sous

...

..

Turned around, it also becomes a theory explaining when it is not possible to construct a classification. In Lerman's paper, arguments about limitations are theoretical. More often, discussion about limitations reverts to empirical observation. Thus, Sparck-Jones (1970) noted that the lack of "global classification thwry" means it cannot be assumed that "a grouping algorithm which works well in one case will work well in another" - a theme echoed many times by other authors: e.g., BlashJield and Aldenderfer (1977), Read and Christenson (1978); Aldenderfer and Blashfield (1984), among others. Sparck-Jones concludes that it will be necessary to "construct alternative classifications, and to apply any plausible means of selecting the best of these, post hoc .. (pp. 258, 259). H i s logic is spund, but the conclusion ,is disturbing for those who would use a strictly empiricist approach as it implies a kind of quantitative nihilism: try anything and everything until results are obtained that can be rationalized as informative. Discussion of another kind of limitation was provided by Cowgill

."

(1970), who drew attention to the discrepancy between what archawlogists can sample and what they would like ti5 sample (1970: 162), hence to the problem of linking the interpretation of analytical results based on what is sampled to the underlying domain of interest. Cowgill distinguished three situations: (1) a population made up of events "involving behavior of members of a specific comunity . . ." @. 162), (2) a population of physical ansequences made up of the "physical consequences of the events constituting elements of the first kind of population . . @. 162) and (3) a physical finds population made up of "all those physical consequences of human behavior which are still present and detectable . . ." @. 163). Cowgill made these distinctions to highlight the need for "clearer thinking about the intervening stages between statistical results and behavioral implications* @. 164) - a topic later taken up and popularized by the work of Schiffer (1976) on site formation processes. Whereas Lerman was concerned with a topic that is properly in the province of the mathematical statistician, Cowgill raised questions that are properly in the province of the archaeologist, yet require integration of thinking from both sides for resolution. Cowgill correctly warned that statistical applications run the risk of being sterile exercises if the steps leading from data to analysis to interpretation cannot be adequately justified. Cowgill mildly chided those who have used factor analysis on small data sets (eg., Binford and Binford 1966) for this reason, noting that "the source of error I have just illustrated is, by itself, enough to make us doubt whether such results can be trusted to tell us very much about any larger populations which the 16 entities may represent" @. 165). However, the problem of linking technique to data to interpretation is only hinted at by Cowgill's comments about sampling. At base, there is a much deeper issue of what constitutes the theoreticaVsymbolic terms used in a theory, on the one hand, and what constitutes the empirical/descriptive terms used to discuss the data to which the thwry will be applied, on the other hand. The need for this distinction can be seen in the confounding of grouping and classification as if these are synonymous terms (eg., Sparck-Jones 1970, Dorm and Hodson 1975, among others); a confounding that underlies attempts to provide automatic cIassi6cation - as opposed to automatic grouping - techniques and theory. The rational for the grouping/classification distinction as it applies to archaeology has been given in detail by Dunnell (1971). Dunnell uses a distinction between grouping as having reference to the phenomenological domain and classification as having reference to the ideational domain. That is, grouping involves forming sets of concrete objects; classification deals with symbolic representation of those sets. Automatic procedures do not produce symbols, only sets of objects; the archawlogist may infer regularity among the grouped objects and impose symbolic identification

."

DWIGHT W. R E A D

METHODS IN ARCHAEOLOGICAL RESEARCH

15

of that regularity, but this cannot be automatic. Grouping is at the level of empirical phenomena; symbolic representation is an abstraction based on empirical phenomena. The symbolic/empirical dichotomy was discussed in some length by Borillo (1977) in a paper given in the Paris workshop on mathematical methods in archaeology.

In a workshop that took place in 1977 in Paris, Borillo (1977) presented an extensive discussion of the logic and rational of using formal, mathematical methods in archaeology, and by extension in anthropology. Borillo noted that mathematics operates with symbols, hence with representations of empirical phenomena, not with the phenomena directly. He observed that there are two translations involved when mathematical fonnatism is used: one going from objects to their "reprkentation r6gulibren, i.e. an unambiguous, weU-defined representation linking object and symbol @. 2); and the other a translation that gives the symbolic system interpretation through terms relevant to the archaeological domain, with the latter based on a research strategy and a coherent, theoretical framework @. 3). Mathematics, he argued, can be used to examine hypotheses about how a set E of data and a set R of measures for those data are structured through a formal representation and a calculus which bears upon the hypothesized structure for the sets E and R. An example would be Binford and Binford's (1966) using factor analysis as a way to recover groupings that allegedlycould be interpreted as tool kits. The linkage between formal methods and analytical results of archaeological interest can, he suggested, best be seen in the problem of automatic classification (the formal calculus) @. 15). Formal results (e.g., separation of data into d i i r e n t groups by means of an algorithm) must be compared with independent data @. 16) for their validation (eg., use of discriminant analysis to show the distinctiveness of the groups). But circularity could enter in as the validation techniques are themselves part of the same calculus. Borillo escapes the circularity through also requiring congruence with the opinions of archaeological experts. The analysis must be reexamined if there is disagreement with experts; however, if there can be no agreement, then the choice of a set E of objects, a priori classes, etc. are thrown into question (p. 21). Fially, there must be corroboration in the form of agreement with the logic of the'framework in which the ' problem is posed, and agreement between the interpretation made of the distinctive traits found through the calculus and other, relevant information @. 23). But if the expertise of archaeologists who use non-mathematical, nonformal methods is to be the arbiter of results arrived at through formal

calculi, then one might well ask what is to be gained through using a formal calculus. Borillo gives several answers, two of which will be mentioned here. First, it becomes possible to formulate more powerful theories. When or "laws" seen at the empirical level are recast as relations in a symbolic calculus, one goes from the more particular to the more general, from the more empirical to the more abstract, and in so doing to a level where these "laws" "s'organise selon des structures plus gknerals, a des niveaux #abstraction plus blev& w m e organized through more general structures, at higher levels of abstractiony (p. 26, translation added). Second, phenomena that appear empirically distinct, and often seen as pertaining to different disciplines, may become grouped in a revealing and significantway through formal representation @p.26-27). For Borillo, then, formal methods are a means to anive at a higher, more abstract level wherein the properties giving structure to phenomena at the more concrete level can be expressed and examined. Archaeological data can be embedded into this symbolic level through an appropriate representation, and if there is embedding into the symbolic level, there must also be the reverse. Unlike Clarke, Borillo does not Suggest that archaeologists should develop an appropriate symbolic calculus to express and develop the theory appropriate to archaeology. Rather, Borillo seems to suggest that there must be simultaneous development of two systems: the theoretical system of the archaeologist through which structures for the data are hypothesized and the formal calculus through which these are analytically examined. This leaves Borillo placing a curious emphasis on the archaeological expert, an emphasis which the 'new archaeologists' have rejected out of hand. Expertise, in the framework of the 'new archaeology', hash0 special episteomological role; rather, it is even to be abhorred in that validation is to come out of confirmation and verification separated from expertise, per se. It may take an expert to come up with significant ideas; but the 'new archaeologists' have argued that ideas have validation only through confirmation via appropriate testing against empirical observation For hew archaeologists' heavily steeped in an empirical, logical positivist approach, Borillo's argument is likely to seem almost irrelevant. Yet profound insights are often made by persons whose methods may seem cavalier and antithetical to "sound" scientific methodology. One has only to consider the insights and impact of persons such as Levi-Strauss and P i e t . Borillo correctly draws attention to validation as being more extensive than merely confinnation through empirical observation. Good theories also provide a coherent logic for the interpretation of phenomena. Doran (1977) took up a similar theme when he emphasized the need to explore the knowledge of the archaeologist: "more powerful techniques will be

DWIGHT W. R E A D

METHODS IN ARCHAEOLOGICAL RESEARCH

obtained not by improving the quality of the mathematics, but by finding ways of using knowledge rigorously" @. 180). But whereas Borillo used the archaeologist's knowledge for validation, Doran suggested, following the work done on the expert system, DENDRAL (see Doran 1977 for references), that archaeologists should be concerned with understanding the origin (in the sense of what reasoning led to their formulation) of hypotheses and not just their validation. Thus, what constitutes the relationship between archaeology and mathematicdstatistical methods is not simple and different persons have had varying views. Moberg (1977) gives a different viewpoint - and one which seems to characterize many archaeologist's conceptuWtion of this relationship wherein mathematics (read: statistics) provides techniques that can be applied by archaeologists.Mathematics is needed, he suggested, not because one is going from concrete observation to symbolic representation but because the empirical context becomes too complex when the amount of data to be analyzed passes a certain threshold @. 194). Mathematics, in his view, provides a series of techniques and the problem the archaeologist faces is which technique should be used: "Les arch&logues ont besoin d'aide pour choisir les mkthodes adkquates [The archaeologists need help in choosing the right methodsy @. 194, translation added). However, if there are "mkthodes adiquates," then there would be no need for validation of the results as Borillo argues. Yet as Borillo (1977: 14) noted, all methods assume a certain structure and without adequate and accepted arguments linking empirical phenomena to formal models.(which do not yet exist), there cannot be a "right" method. Whallon (1977) examined this linkage by arguing that however much the polythetic model for grouping data may be useful for statistically designed automatic "classi6cation" procedures, it may not be valid for archaeological typologies @. 216). Whallon (1977) discussed his earlier paper (1972) aimed at replicating an "intuitive" typology for Oswego pottery using a monothetic, hierarchical procedure. While no statistical methods more complex than a chi-square test were used, his analysis is in keeping with Spauldii's comment that statistical analysis of archawlogical data is an extension of archaeological reasoning. Whallon's paper bridges the two modes of reasoning - embedding archaeological concepts into an appropriate calculus to examine their logical consequences versus viewing mathemapcs/statistics as providing , methods to be applied by the archaeologist - represented by Borillo's and Moberg's discussions. Whallon clarifies the issue from the archaeologist's viewpoint, hence lays the foundation for how one might begin to build a symbolic calculus (in Clarke's sense) appropriate to archaeological reasoning. The outline of a formalized, symbolic system for expressing the logic of grouping procedures as they relate to a classification was independently discussed in a paper by Read (1974).

Other authors in the workshop reviewed the "symbolic calculus" side of ~ o d l o ' sdiscussion; that is, they focused on the statistical analysis of smcture: Ihm (1977a) reviewed uni- and multivariate statistics; Sibson (1977) discussed the theory behind multidimensional scaling; Lerman (1977) developed a new rank-order correlation coefficient which avoids bias in Kendall's t; and Guenoche and Ihm (1977a) developed estimators that may be used in principal component and discriminant analysis for situations where data is missing. The remaining papers for the workshop considered several aspects of seriation. Ihm's paper (1977b) is particularly interesting for it formally defines what constitutes a solution for seriation. Given an incidence or abundance matrix X whose columns represent units (sites, graves, etc.) and rows the type of object being counted @resence/absence for an incidence matrix, frequency for an abundance matrix), then one needs to find a mapping T such that for the kth column vector, x,, of measurements for a unit (i.e., the vector whose components are the values of the measures for the kth unit) T associateswith xk a real number: T :x,

tk E Reals,

with t, proportional to the date of origin for unit k. Ihm shows how such a mappmg T may be defined if there is a distance measure, 4 such that for the ith and jth units, with column vector of measurements xi and xi respectively, Hence the problem is to find such a distance measure. Ihm correctly points out that there is no general solution, only a solution if one has specified a "modUe mathimatique de la variation de frkquence des diffkrents types avec le temps [mathematical model for the change in frequency of types through time]" @. 140, translation added). For an incidence matrix where types appear and disappear in a regular manner through time, Ihm noted that it suffices to use the L, (city block) Ix, - x,l, Z lx, - x,[. For abundance matrices norm; i.e., d(x,, x,) where the frequency of a type follows a Guassian distribution through time (the classical "battleship" curve), Ihm showed that the typical "horseshoe" shape that results from using multidimensional scaling techniques as discussed by Kendall (1971) can be linearized by taking a logarithmic transformation of (he vector dot product of the ith and jth column vectors (which represent the i th and jth unit) in the m a t r i x X. Ihm's (1977b) paper represents well Borillo's discussion of the double translation involved when using formal methods. First, there is a translation to a symbolic representation (the incidence or abundance matrix as the representation of an archaeological unit). Analysis takes place within the context of a symbolic calculi (definition of a distance measure, d;

DWIGHT W. R E A D

METHODS IN ARCHAEOLOGICAL RESEARCH

deriving a mapping T from d that maps the symbolic representation, xi, of units to an ordinal time scale, ti, based on a model relating time to the presence/ absence or frequency of types within units). Second, there is a translation back from the symbolic domain to the domain of archaeology (interpretation of the ordinal scale as a relative time scale for the archaeological units through verification of the model for change in presence/ absence or abundance through time). Whereas the other authors (e.g., Regnier (1977), de la Vega (1977)) dealing with seriation emphasized more the algorithmic or procedural side of a seriation solution, Ihrn effectively drew attention to the fact that no solution is possible in the symbolic domain without an a priori model; such a model can only be justified through observation and theory about the empirical domain.
TO PATTERN THE PAST

In 1981, at the Tenth International Union International des Sciences Preet Proto-histoire (UIPPS) meeting in Mexico City, what had been the Data Banks Commission expanded into Commission 4 on "Data Management and Mathematical Methods in Archaeology." Under its new aegis, Commission 4 sponsored an international symposium in Amsterdam in 1984, leading to the volume, To Pattern the Pu.sf. The volume contains 23 papers divided into four sections: (1) Documentation and Presentation, (2) Classification and Seriation, (3) Pattern Searching and Pattern Verification, and (4) Modelling and Simulation. While the sections generally represent the topics considered in the previous conferences, there are certain notable differences. First, extensive work had now been done on the mathematical basis for seriation methods, particularly by Ihm (e.g., Ihm and van Groenewoud 1984), based on correspondence analysis. Seriation using correspondence analysis figures in three of the papers: those by Djindjian (1985), Slachmuylder (1985) and Scollar et ul. (1985). In the first of these three papers, Djindjian (1985) pointed out what had heretofore been discussed as different techniques (the Brained-Robinson seriationmethod (Robinson 1951), the Petrie matrix analysis method (Petrie 1899) and the multidimensional scaling method (Kendall 1971)) were in fact identical and could all be subsumed under correspondence analysis. Djindjian noted I advancement in that it seriation based on correspondence analysis is @ can display the interfering effects of structures other than the supposed i n g analyzed, hence the effects of these other serial relation of the units W structures can be removed. While other seriation methods mainly examine units separated from their spatial context, Djindjian suggested using toperiation which combines topochronology (Werner 1953) - that is, "the chronological development of a site as expressed in space" @. 126) with seriation.

In the second paper Slachmuylder (1985) applied correspondence analysis to seriate a series of Belgium sites using an extensive typology (121 types) developed by Rozoy (1978) for the mesolithic of France and Belgium. The results agree reasonably well with14Cdates for the sites. The third paper, by ScoUar, Weidner, and Herrog (1985), discussed a series of programs based on Ihm's algorithm for seriation using correspondence ------ analvsis and written in Pascal for a mainframe computer. Important here is th; the suite of programs have been announced-as available for world-wide distribution. Since the date of this conference, the entire series of programs have been ported to PC's and are now avilable on a cost only basis, thus making a major addition to archaeological methods widely available. In effect, the development of this suite of programs signals that basic, underlying mathematics for seriation has been satisfactorily worked out. W e the work of Ihm is a contribution to archaeological methodology by a statistician interested in archaeological research, the Symposium also included a contribution to statistical methods by an archaeologist. Voomps, Loring, and Strackee (1985) developed what they call a gamma mir density function as a way to model a baseline condition of random behavior in the selection of animal species hunted during the Middle Paleolithic. Their model is motivated by a suggestion due to Johnson (1981) that settlement system integration may be a multiplicative function of various conditional probabilities. Using the assumption that species are procured at random from the environment, Voomps et al. (1985) assigned to each species a "popularity" index representing the proportion of sites in which the species is found. These indices are multiplied together to form a multiplicative popularity index, v, for a site, following Johnson's suggestion of using a multiplicative function to represent system integration. Under the assumption that the popularity indices are uniformly distributed over the interval [0, 11, and that the number of species in a site is uniform over the interval [I, N], Vonips et al. (1985) showed that the -In o will, for a fixed value of N, have a density function f (x) for x gamma distribution:

If the number of terms, n, in the products making up o has a uniform distribution over [I, N], then.thedensity function for x becomes

i.e., a mix of gamma functions. The model was applied to data on the coastal and inland Middle and Upper Paleolithic from West Central Italy. The authors found statistical fit to the gamma mix for the coastal data sets using the Kolmogorov-Smimov one sample test and concluded, assuming the validity of the model, that

20

DWIGHT W. READ

METHODS I N ARCHAEOLOGICAL RESEARCH

21

only the inland sites show evidence of specialization, with no detectable difference between the Middle and Upper Paleolithic sites. However, none of the observed distributions is well-modeled by the gamma mix distribution since all but one of the cumulative, observed distributions lies above the cumulative distribution for the gamma mix distribution, whereas the null hypothesis of no difference implies that the observed distribution should be randomly distributed around the theoretical distribution. Stochastic models such as the one developed by Vomps et al. (1985) and relevant to processes underlying how archaeological data are structured play a more prominent role in this Symposium than in the previous conferences. Modeling a situation as the outcome of a stochastic process is also used by Chippindale (1985) who presented a simulation of the shapes of cow horns found in rock-art on Mont =go. The simulation used a directed random walk on a computer screen. From the current pixel location, the next pixel was selected according to a probability scheme for the eight surrounding pixels. By varying the probabiities from one compute; run to the next, chippindale was able to ~roduce a varietv of simulated horn sha~es which resemble the rock-art ; o m , hence he &as able to demonstrateathat local "pattern" which might be used in a typology for the horn shapes in the rock-art is producible from a stochastic process, hence may have no relation to "intention" on the part of the artisans. These two papers share commonality with several other papers in the Symposium that all ask a similar question: What is the basis, or lack thereof, for structure seen in archaeological data? Whereas Vomps et al. and Chippindale examined baseline models using stochastic processes, Simek, Ammerman, and Kintigh (1985) saw the spatial distribution of data in archaeological sites as more complex than is generally supposed in arguments advanced about activity areas. They suggested the need for heuristic approaches that allow the analyst to decompose the spatial distribution into spatial clusters that can be examined in terms of their heterogeneity @. 230) while simultaneously taking into account the effect of small sample sizes on the number of artifact classes recovered in the site (see Kintigh 1984 for an extensive discussion of this problem). Bietti, Burani, and Eanello (1985) u s 4 much the same philosophy and suggested that pattern recognition does not so much need automatic methods as interactive ones that allow the archaeologist "to perform several experimenfs on the data, to play with the datan i @. 224, emphasis in the original). A paper by Carr (1985a) suggested that Fourier filtering methods may be a solution to the double b i d discussed in Christenson and Read (1977), namely that data need to be prescreened prior to applying grouping techniques precisely with the information that one is attempting to gain from the analysis. O'Shea (1985: 107) reached a similar conclusion

in his discussion of the difficulties with using cluster procedures to recover simulated groupings in mortuary data. And, as Christenson and Read had shown earlier, he found that the ability of cluster procedures to recover groupings diminishes with inclusion of measurements for nonessential, or irrelevant attributes.
QUANTITATIVE RESEARCH M ARCHAEOLOGY

I n 1985, a national symposium on quantitative methods was held jointly with the 50th Annual Meeting of the SAA in Portland, and Commission 4 of the UISPP. Whereas the charge in the Mamaia Conference was for the archaeologistrhistorian to address the statisticiadmathematician and vice versa, here the archaeologist participants were charged to "think about . how they would communicate their thoughts to archaeologists who are nonspecialists in the field of statistics or quantitative methods" (Aldenderfer 1987a: 7). Also different about this symposium was the many papers that questioned the fit between statistical methods and the goals of archaeological research. Rather than viewing statistical methods as the standard towards which archaeologists should strive (see Aldenderfer 1987: lo), these papers saw discordance between goals and methods. The discordance was seen as stemming in part from the non-holistic nature of an assemblage (Brown 1987: 295-5), and in part from lack of concordance among the variables selected, data to be analyzed and models inherent in methods being used (Carr 1985, 1987a; Clark 1987; Read 1985, 1987b; Brown 1987). Yet another sense of discordance was discussed by Aldenderfer (1987a) who pointed out that while there may have been conceptual acu a n t i t a t i v e / m a t h e m a t i c a lapproach to classificaceptance of a shift to a q tion in American archaeology, the means for carrying out the shift, as exemplified by Read's papers (1974a, 1974b; see also Read 1982) setting forth a mathematical basis for typologies, did not get equally accepted. This happened because (1) the formal language of set theory was a banier and (2) non-quantitatively formulated typologies allegedly "work," hence there was no perceived need to change past methods @. 26). Aldenderfer attributed the reaction in part to a social and political dimension @. 27) which continues to affect how quantitative methods are viewed in the discipline, separate from their intrinsic value. Although the various authors gave different suggestions, common is the sense that something is amiss, that the presumed promise of quantitative methods as a means to provide objectivity and insight is defective. Proffered solutions varied: Q-mode analysis (Aldenderfer 198%; Brown 1987); artiiicial-intelligence (Doran 1987); "middle range theory" (Voomps 1987); simple statistics (Wballon 1987); and causal modeling

..

DWIGHT W. READ

METHODS I N ARCHAEOLOGICAL RESEARCH

(Kimball 1987) were a l l suggested. Read (1985, 1987b) did not so much advocate a specific method as examine the discordance that arises from a statistical framework implicitly oriented towards description, on the one hand, and research goals aimed at explanation, on the other hand. Read (1985, 1987b) pointed out that if statistical analysis is to have explanatory content, then a population brought forward for study must be homogeneous with respect to an underlying process giving these data their structure. Also, the variables used in the analysis must measure relevant aspects of the outcomes of the process. Otherwise, the analysis only has descriptive value in the sense that estimates for parameters in posited models may bear only an uncertain relationship to the properties of the processes allegedly being modeled. Read (1985, 1987b) exemplified the argument with previous work on the spatial configuration of huts in !Kung Sun camps and a typology for projectile points from the Chumash Indian site, 4VEN39, in California. Read concluded that
we must come to grips with how the cultural context .. is strctured, and how the C U ~ Ncontext I ~ ~ relates to che phenomeoa that we 0 b s e ~ eand study. SmMcal reasoning is one means to express those relations s o that their comequencu can be dmwn out and cxpmed (1987: 103,emphasis added).

analytical techniques consistent with the mathematical model and data structure. Making the interconnections among data, structuring processes and analytical methods, is, he points out, a basic aspect of scientific methodology (1987a: 238). Related to the lack of fit problem between data structure and model structure is the concern Cowgill expressed in 1970 about data reliability. Data reliability is examined in the symposium paper by Nance (1987). Nance pointed out that archaeologistshave not been sufficiently concerned about the reliability and validity of the data they collect, yet the methods for measuring the reliability of measurements are well established and can, as he demonstrates,be fruitfullyapplied to archaeological research.
SUMMARY OF CONFERENCE PAPERS

'

Return has thus been made to Spauldiing's implication that application of statistical methods should be but extension of archaeo~ogicd~easonin~. The theme of discordance and its resolution was also taken UD bv Carr (1985b, 1987a). Carr sees "specification of potentially rele4t &ta or techniques using middle range theory, constrained exploratory data analysis [Carr 1985b1, entry models, and stepwise cyclical analysisn (1987a: 237) as ways to minimize the problem of discordance between heterogenity of data and presumed homogeneity of technique. Carr argued the "theories, models, hypotheses, test implications, mathematical techniques, data collection methods, andlor the data that are involved in an investigation be relevant to and logically consistent with each other, and ultimately to the empirical phenomenon of interest" (1987a: 187). Carr suggested that the "methodological double bind" discussed by Christenson and Read (1977) may be overcome by a multifaceted f data structure to gain approach to data analysis aimed at exploration o insight into how the phenomena of interest art structured, and through using initial insights in a reflexive, cyclical manner (1987a: 213). Included in this approach is the use of multiple entry models for approaching the analysis of complex data. Carr (1987a: 224-35) considers an entry model to consist of three parts: (1) a mathematical model specifying the relationship between the variables in question and the implied data structure, (2) an enumeration of possible processes that could give rise to data structured according to the mathematical model, and (3) specification of

While these conferences all place attention on what can be achieved with quantitative methods in archaeological research, what this means as praxis has been less unified. Three Werent approaches are exemplified by these papers: (1) a statistical-method/archaeological-problem dichotomy, (2) statistical reasoning as an extension of archeological reasoning and (3) formaVmathernatical representation of archaeologically defined relationships. The first approach is meant as identification of a specific archaeological problem along with a relevant statistical method. Many of the papers from the Mamaia conference are of this nature: the archaeologists identify a problem and want to know an appropriate method; the statisticians identify a method and want to know an appropriate problem. When done well, it can be effective; done poorly, it reintroduces subjectivity through contrived interpretations. Effective work using this framework depends upon having a well-argued archaeological problem that clearly identifies the specific relationship to be examined rather than relying on an implicit analogy between problem and method (see Read 1985) for justification. That is, the difference is between an assertion such as "we want to know if records of hardness of sherds according to the Mohs' scale are useful for distinguishing groups of sherds, knowing already that the hardness is an indication of the method of manufacture or the manner of conservationn (Myers 1950: 17, emphasis added) versus "it seems preferable where taxonomy is the main interest, to carry out a special purpose cluster analysis" (Doran and Hodson 1975: 185, emphasis added). In the second grouping are included statistical distinctions used to provide a framework for expressing concepts which are part of archaeological reasoning. Papers with this orientation are less frequent than those of the first group and are scattered through the conferences; e.g. of the papers mentioned above, those by Orton (1971), Whallon (1977), Chippindale (1985), Carr (1987a),and Read (1985,1987b) are examples.

24

DWIGHT W. READ

METHODS IN ARCHAEOLOGICAL RESEARCH SAMPLING

The third group includes papers that focus on developing properties in a formal framework once the basic relationships of interest have been identified. The mathematical work on seriation and on the mathematics of automatic classification, both of which figure heavily in the Marseille conference and the Paris workshop, are examples of this group. These differences in approach are partially the consequence of different trajectories that archaeological research has taken in different countries, hence there is no common, unidirnensional time trajectory. Nonetheless, there has been a significant trend towards questioniFg the framework used for the first grouping. That framework assumes a division of labor between archaeologist and statistician. But as the UISPP and SAA symposiums e i n g eroded as archaeologists increasingly make clear, this division is b regard statistical methods as a means to reason quantitatively; hence they increasingly use statistical methods as part of, and as a means to extend, archaeological reasoning. The third group has had limited development despite programmatic statements (e.g., Clarke 1968; Gibbon 1985) about the importance of developing formal representation of archaeological arguments (coma Salmon 1982: 178). Research fitting into this group requires shifting away from an empirical framework aimed at identification and interpretation of data structure to one centered on developing theory as an abstracted system of reasoning utilizing the principles and processes thought to structure the data of interest. The groundwork for such a shift has been h i s kind will become more prevalent. laid; undoubtedly research of t PART LI. QUANTITATIVE Pr/LETHODS:IMPLEMENTATION The first part of this review has focused on the changing perceptions of what constitutes the role of quantitative methods in archaeological research. In this second part, focus will be on the manner in which quantitative methods have been used in research and on conceptual problems that arise from their application. For this part of the review, articles have been arouwd in accordance with a basic analvtical scheme: sampling, measweieni data stiu&re, inference, and modeiing. Measurement will be examined for three topics that share the common problem of constructing a measure for a (or quality) that do& not have simple formulation. The three topics are: (I) shape of artifacts and features, (2) number of individuals represented by a faunal assemblage, and (3) chronology. Means for assessment of structure will be taken up (1) inference of structure as ex~ressed in a data matrix with two to~ics: and (2) skcture' & a spatial arrangement. ~taksticalmodeling will be discussed as an extension of archaeological reasoning.

Sampling in American Archaeology The Burg Wartenstein conference concluded, with regard to statistical methods, that "a central problem appears to be the establishment of the correct method of sampling in the individual casen and "archaeologists [should] familiarize themselves with the basic principles of sampling theory" (Heizer and Cook 1960: 357). Shortly after this conference Binford (1964) proposed probability sampling methods as the basis for this "correct method of sampling." Yet almost two decades later, in a review and critique of the application of statistical sampling methods in archaeology, Hole (1980) noted that even though texts on sampling theory are accessible, and even though sampling theory has great potential for archaeological research, "its contributions to archaeological knowledge have been disappointing" @. 217). Hole is critical of archaeologists who view sampling merely as a construct to be mapped onto archaeological data recovery fieldwork orocedures. Her criticism was foreshadowed in a commentarv Binford j1974) made on the Symposium "Sampling in Archeology" heldas part of the 1974 SAA Annual Meetings in which he observed that "a sampling strategy must be evaluated with regard to the character of the target population to be evaluated, not in absolute termsn @. 252). Similarly, Read (1974) had commented: 'There i s no best sampling procedure . . . The sampling procedure must take into account . . the information desired, the distribution of that information in space, cost of obtaining samples, and degree of precision needed, etc." @. 60) - a theme also echoed by several of the other participants of the symposium. If so, the recommendation of the Burg Wartenstein conference was in error, and also in error were all subsequent papers centered on delineation of a specific sampling scheme as the best for archaeological research. Hole described papers that limited their concern with sampling to the virtues (or lack thereof) of specific sampling schemes as "Algorithmic Archaeology" @. 220) and "Paper Surveysn@. 224). By these terms Hole refers to a shift from "papers exploring and advocating" probability sampling to "a veritable flood of prose regarding it as the only way" (p. 218). Hole discussed the Cache River project (Schiffer and House 1978) as an example of algorithmic archaeology and a paper by S. Plog (1978) as an example of paper surveys. Hole commented that in the Cache River project "the research design utilized no prior knowledge about the distnbution of archaeological sites, their nature, the prior likelihood of sites existing in different areas of the survey tract . . One could design the same research for this area . . without knowing much about archaeology

'

. .

DWIGHT W. READ

METHODS IN ARCHAEOLOGICAL RESEARCH

. . [and] could equally employ the design anywhere else on the face of the earth" @p. 222-223). In Plog's paper an attempt had been made to assess the relative value of different sampling designs through simulated surveys based on the distribution of sites in a single region. Hole noted that "Plog's critical misfortune is the failure to realiie that the results are more informative of the data than the statistics" @. 225). That is, simulations of this kind are neither about statistical properties of sampling procedures nor the relative efficacy of one or another sampling procedure for estimating a parameter such as the number of sites in the region; rather, they are about how spatial distribution properties of archaeological data differentially affect sampling designs. The latter is a mathematical issue; the former an archaeological one. The one depends upon mathematically framed arguments; the other upon archaeologicallyframed arguments. The mathematical argument has two foci: one is inferential statistics the relationship between sample statistics and population parameters as expressed through estimators of population parameters from sample data. The other has to do with extending the underlying statistical theory to sampling designs more complicated than simple random sampling. Read (1974) highlighted some of the mathematical results of sampling theory which bear on choices one might make in designing a sampling program for a region; for example, stratification is advantageous only if the resulting strata have greater internal homogeneity with respect to the variable of interest than does the region as a whole. Hence, the answer to the archaeologist's question: 'Should I stratify?" from a sampling theory perspective is, "That depends on the spatial variabiity of the data." The statistical argument for stratification is based upon a particular goalqelection of an estimator with minimum variance. An estimator that is constructed using a stratified sample based upon intra-stratum homogeneity and inter-stratum heterogeneity will have less variance associated with it than an estimator constructed using a simple random sample. Although these are important properties upon which to base research centered around inferential statistics, the rationale for designing an archaeological research p r o w upon goals derived from sampling theory is less clear. Archaeologists do use stratification in regional studies, but stratification is generally seen in terms of the importance of ecological andlor geological characteristics in a materialist interpretative framework, not because of distributional properties of populations whose parameters data are to be estimated throueh sam~le s Further, a property &h as spatial variability in site density need not be a good archaeological criterion for stratification. To illustrate, a matter which arises with Htratiiication is the size of the sub-sample allocated to a stratum. Sampling theory shows that if the size of the sub-sample for a stratum is proportional to the variance of the variable for that stratum (Neyman optimization), then the parameter value will be estimated with

equal efficiency across all strata. From a sampling theory viewpoint it is reasonable to use Neyman optimization (see also the discussion by voomps et al. 1987: 233). But the same viewpoint need not be true from an archaeological pe~~pe~tive. Suppose that in one stratum, sites are perfectly uniformly arranged over ,pace and in another stratum they are clustered. The former has low variance for the number of sites per sample unit; the latter has high variance. Under Neyman optimization most of the sample should be allocated to the latter and only a small sub-sample should used for the former. It is unlikely that many archaeologists would accept such a sampling program - not because of any failure to appreciate the mathematics of sampling theory, but because archaeological research is unlikely t o be focused solely on an estimate of the number of sites in the region (Plog, Weide, and Stewart 1977: 115-116). Instead, it requires many kinds of information, little of which is distributed in a parallel fashion with site density. As Hayes, Brugge, and Fudge (1981: 109) noted in their discussion of a transect sample survey of Chaco Canyon National a s . . . directed toward obtaining as Monument - "The transect survey w much information as possible . . [and] conducted specifically as an inductive searchn - the antithesis of the rationale for the Neyman optimization criterion for allocating samples to strata. The Chaco Canyon survey is also instructive for what it informs about sources of bias in obtaining archaeological data. The Chaco Canyon survey had two parts: (1) the transect survey,which used north-south running transects located randomly also an east-west line and covered 25% of the Monument, and (2) a 100% coverage of the 32 sq. mi. of the Monument. The latter was done to inventory all sites and so allowed for comparison of the "actual" and the estimated total number of sites in the region based on the transect survey. Hayes et al. treated the transect samples as observational values from a simple random sample and computed a 95% conflidence interval for the total number of sites to be 1547.8, 717.61 for the estimate of 644.1 sites in the region. However, in the complete inventory a total of 1,689 sites had been found - approximately a three-fold discrepancy between the estimated and the observed totals. The authors then proceeded to rest the accuracy of the transect design" @. 123) by using the site data from the complete survey corresponding to the transect sample to estimate the total number of sites. Not surprisingly, a 95% confidence interval, 11593.9, 2159.7 based on this datum includes the population total of 1,689. Indeed, the meaning of a 95% confidence interval is that the constructed interval has a 95% chance of including the true value so long as no systematic bias had been used in locating transects and the distribution is approximatelynormal. Ofmore importance than this elementary confirmation of a theoretical

DWIGHT W. R E A D

METHODS IN ARCHAEOLOGICAL RESEARCH

property from inferential statistics was their discovery that the discrepancy appears to be related to different number of person hours/transect used in the two surveys (1 1.7 versus 30.6 person days/sq. mi.). The total inventory had 2.62 times as many person-hours per transect and if the number of sites found is approximately linear with intensity as Hayes et al. argued, then the corrected estimated total would be 632.2 X 2.62 = 1,655.6 versus the actual value of 1,689 sites. Hence difference in sampling intensity seems to account for the discrepancy. Concern with 6nite population corrections, correction for sampling by spatial unit as a form of cluster sampling (e.g., Mueller 1975; Nance 1983) and other technical details which are part of sampling theory become h i m c a n t when compared to a bias of this magnitude. This and other studies (eg. Coombs 1978) suggest that precision and accuracy may need to be replaced by reliability and validity (see Nance 1987) when discussing research design and field methods. The background theory for so doing already exists as Nance (1987) has pointed out. Whether it will be used effectively depends upon archaeologists addressing questions such as: What are the archaeological r y i n g to get reliable and valid data?" not problems to be f a d when t "which test of reliability or validity is better for this data set?". Nance (1987: 290) correctly sees the need to "collect data in ways that permit --reliability and validity to be assessed;" even more, data needs to be collected in ways that are reliable and valid.
Sampling in British Archaeology

reconnaissance, the second stage is to use probability sampling of the site based on the information obtained from stage 1 and the last stage should include judgmental selection of units from stage 2 which require more extensiveexcavation. Nonetheless, discussion frequently reverts to the level of generalities which provide little information on what, substantively, needs to be considered when forming a sampling program. It does little good to say that "the specilic research aims, previous knowledge of the population characteristics, and the ease or difficulty of implementation given cost, time, storage and sample size limitations must be kept clearly in mind" (Fasham and Monk 1978: 379) without effectively translating the general problem into specific guidelines for making descisions when formulating a sampling program. Attempts are made to do this, but marred by misconceptions. For example, Fasham and Monk (1978: 388) are encouraged that a sampling fraction as small as 1% yielded on unbiased estimate of the mean even though bias is a property of estimators, not estimates, and (except for asymptotically unbiased estimators) unrelated to sample fraction.

Form Versus Substance in Sampling


Errors of this kind aside, both in this conference and in the earlier SAA symposium one senses that more often than not the arguments are about the form and not the substance of sampling theory. The idea of sampling as a way to reason about the relationship between what can be observed and what one wants to know gets translated into an exercise focusing on which method better answers questions as defined by sampling theory. Is the intent of archaeological sampling to estimate population parameters? Or is sampling the means to recover the data which constitute the domain of study for the archaeologist? The first question is answered by sampling theory; the second is not. Orton (1978), in a brief conhibution to the conference, recognizes the difference when he rejects the sampled-population/target-popuktion dichotomy as largely irrelevant to archaeology and DroDoses instead that one should set UD "mathematical models which will k i u d e both archaeological and sampkg ideas" and "it may even be that .sampling theory itself is irrelevant [except as] a starting point for discussion about why we quantify, how we quantify, and what sort of statements we can reasonably expect to prove useful" @. 400, 402). And using a mathematical idiom to reason about archaeological problems is, Groube (1978: 405) suggested in his closing remarks to the conference, but a shift from ad hoc to formal methods, hence is an extension of already existing archaeological reasoning. The sentiment is reminiscent of Spaulding's comments at the Mamaia conference. There is a place for testing sampling designs with archaeological

Symposia on sampling have not been limited to American archaeology. A conference titled 'The Role of Sampling in Contemporary British Archaeology' was held in 1977 at Southampton. Unlike the earlier SAA symposium, the British conference did not have a background of indigenous archaeological work to draw upon and only some of the American literature had been accessible to British archaeologists (Cherry, Gamble, and Shennan 1978: 1). In these papers there is a curious naivete about what statistical theory establishes - even a distrust of theory by British archaeologists (Cherry and Shennan 1978: 30) - that shows up in reliance upon experimentation 'to compare the efficiency of sampling designs against one another. . .using archaeologicalmaterials" (Haselgrove 1978: 164; see also Voorrips e t al. 1978: 241; Fasham and Monk 1978: 395). Archaeological data are seen as a different kind of data and sampling theory has to be tested to see if it "really worksn for these data. More positively, there is often an attempt to fit sampling to archaeological practice rather than the reverse (e.g. Peacock 1978; Jones 1978; Champion 1978). Cherry e t al. (1978: 154) see the need for a three stage research program (see also Redman 1973) in which the first stage is aimed at site

.. .

DWIGHT W. READ

METHODS IN ARCHAEOLOGICAL RESEARCH ARTIFACT FORM

data; that is, for testing how one can recover the data of concern to archaeologists in a manner consonant with the interpretive goals of archaeological research. Archaeological research has gone from inferring patterns on the basis of single sites and single extensive excavations to patterns seen through aggregated data, where the data are obtained from physically distinct locations, be they sample units in a site or spatial units in a region Interpretation that is based on pattern seen in the aggregate places new demands on the way data are obtained to ensure that pattern used to make interpretation is not pattern created through data recovery methods. Read (1986) has argued that spatial sampling is primarily a means to create a representative (in the sense of preserving proportions) sample of the full range of archaeological data and is only secondarily a means to estimate population parameters - except for predictive modeling, which explicitly has as its goal estimation of population parameters. Read suggests that sampling efficiency should be seen as relating to data recovery, not to estimator variances; it should implement current knowledge about the spatial distribution of sites both in relationship to ecologicaVgeological features and to other sites as part of a settlement system. Efficient sampling in this sense requires the application of archaeological reasoning, not just sampling theory reasoning. There may be overlap in the two modes of reasoning,but they are not identical.
MEASUREMENT

Artifacts are material objects, hence have form. Artifact form has both quantity (size) and quality (shape). Shape is comprised of the relationship of parts to one another; size is the magnitude of the parts. In addiion, form, as it relates to amfacts, has both a material aspect (function) and an ideational aspect (emic conceptualization of fonns). While both size and shape may involve the material and the ideational dimensions, shape would seem, in most cases, to contain the greater amount of information; hence the frequent emphasis on measuring shape separated from size. Though shape belongs to the domain of geometry, geometry does not adequately give us a vocabulary to discuss the separation of form into size and shape except in special cases; e.g, forms that have the same shape but different size (congruent shapes) or fonns which differ in shape but without specification of size (e.g., hiangles versus rectangles). In the absence of geometric speciiication, archaeologists have largely relied on linear measurements to measure forms, and ratios of measurements to infer shape.

One of the more extensive and systematic of these schemes was given by Benfer (1967) who used a suite of 23 measurements on a projectile point. Benfer's system is but the elaboration of a tradition of measurement whose simplest canonical form is: length, width and thickness of an object. Shape in this canonical form is taken to be a ratio such as shape length/width. This, for example, became the basis for Bordes' (1961: 6) distinction of a blade from a flake: if Uw < 2 the lithic is a flake; otherwise it is a blade. Bordes' system is arbitrary in the sense that it has no underlying geometric or cultural theory. The extremes one can be led into by an arbitrary system of measurement are demonstrated in an article by P i t t s (1978) who applied principal components, Wards cluster technique, Gower's Constellation analysis (described in Doran and Hodson 1975: 205-209) and spatial analysis to published data on flakes from England and Italy in which the length/width ratio was divided into six shape classes. Though Pitts noted the criticism (e.g. Hassan 1974, Chapter m)made of Bordes' blade/flake dichotomy as possibly being nothing more than an arbitraq subdivision of a continuum, Pitts only sees the problem as one of choice of the number of classes (the dichotomy used by the French, or the several fold subdivision used in Italy and England) and standardizationof classes. Pitts 6nds that something non-constant has been measured - the data have spatial and temporal patterning - and concludes that the several fold distinction is better than the dichotomous one. The implication is not seen: by the same logic an even h e r distinction is Likely to be more informative and camed to its extreme one would say that the best approach is to make as many classes as possible; i.e., use length and width measurements to make each object the sole member of its own arbitrarily defined class. Clearly the problem lies not in whether one should use 2 or several shape a s made classes, but in the premises upon which the original dichotomy w as an imposed and arbitrary system for the numerical representation of shape. Similar criticisms of an arbitrary approach to artifact measurement appeared in two independent sets of papers (Main 1978, 1979; Read 1982). Both authors noted that the arbitrary approach fails to provide information h m which the original outline can be reproduced. Call the ability to recreate the fonn from a numeric representation the archival property. In addition to the archival property, Read noted that all measures or coefficients making up a numeric representation should have meaningful interpretation within the theoretical framework used in the analysis. The two authors differed, however, on specific suggestions for a measurement system that would satisfy the archival property.

DWIGHT W. R E A D

METHODS IN ARCHAEOLOGICAL RESEARCH

Main (1978) represented a smoothed outline given in digitized form through two coefficients: arc length and slope of the chord (tangent angle) connecting consecutive points in the digitization. By plotting arc length against tangent angle, a profile for a shape is produced and profiles, or sections of profiles, can be compared and a similarity measure constructed. Then the similarity measure can be used in a cluster analysis for grouping. Main applied the technique to cross sections of train rails and "achieved a very reasonable classification of 48 fairly irregularly shaped objectsn @. 46). Read (1982) discussed the need for a non-redundant and sufficient set of coefficients for representing the outline of projectile points. By sufficient he meant that the original form could be reproduced from the coefficient values; i.e., the archival property would be satisfied, and by non-redundant he meant that no smaller number of coefficients would be sufficient. Read viewed the outline of a form as a series of connected arcs (curves with no inflection points). Read represented a smooth arc via three coefficients: (1) length of a chord connecting the endpoints of the arc, (2) maximum height of the arc above the chord and (3) distance along the chord to the maximum height. This, along with the angle of connection for two arcs, leads to a representation in the form of a vector: the first three components of the vector represent an arc, then the next component is the angle joining this and the next arc together, the next three components define the next arc, etc., until the outline has been completely traced. Read showed that this representation scheme was sufficient and non-redundant and applied it to Scottsbluff points so as to examine intersite diierenw in the assemblages from which these points came. Phagan (1985) has applied a similar scheme to projectile points from the Dolores Project. Read noted that the distinctions upon which the representation was based - arcs and angles - could be interpreted for cultural and functional saliency through the form of Frequency distributions for coefficient values. Read used this technique to show that the angle for the tip of the point is in accord with material constraints whereas the shape of the base section of the Scottsbluff points is more in accord with ideational constraints. A third representation for the outline of an artifact was discussed by Gero and Maxxullo (1984) who illustrated that if the outline is embedded in a polar coordinate system centered at the centroid of the form, then the outline (at least for most lithics) could be viewed as the graph of a . periodic'fun~tion, the Fourier function (defined below) with period in. Whereas Pitts carried the arbitrarv measurement svstem to an extreme. Tyldesley, Johnson, and S n a p (i985) did the &me with so-called objective measurement procedures. They suggested representation in the form of a profile computed by constructing the angle between a major axis through the centroid of the form and a minor axis through the centroid of

the two sub-shapes formed by the major axis (references for the method are given in the paper). The process starts by drawing a major axis vertically and measuring the angle between the major and the correspondi n g minor axis. The construction of the major and minor axis and the measurement of the angle between them is then repeated for each 1 degree increase in the angle between the major axis and a vertical line as one goes around the outline. The profile is formed from a plot of the interval angle (the angle between the major axis and a vertical line) versus the angle between the major and the minor axis. The authors commented that when the profiles are used to group paleolithic bifaces, groupings are formed "which may differ from those classified by more traditional methods, but which are worthy of consideration as an entirely objectively genemted system" @. 23, emphasis added). The profile based on centroids does not satisfy the archival property as congruent shapes will have identical profiles, though one could add a size coefficient. The Fourier function can satisfy the archival property, but does not satisfy the non-redundancy property as the number of coefficients used to d e h e the function relates to accuracy of fit to outlines with "corners" (points of discontinuity for the first derivative), not to shape properties per se. As Read (1982) discussed, values for coefficients which are part of the characterization of a shape can be treated as values of variables when different shapes are compared. However, these need to have interpretation at the level of process. The Fourier coefficients, except possibly for the first 2 or 3, cannot be so interpreted. The tangent angle used in Main's representation is too "fine" a measure for interpretation as it (potentially) varies at the scale of pixels. Read's vector representation is at the right scale, but whether arcs and angles between arcs have emic relevance was only partially resolved. From these papers we may distinguish two separate problems that need not have identical solutions (see also Main 1979). The first problem has several solutions for simpler cases such as planar curves; the other has not yet been resolved. The first problem is archiving a form in an efficient manner, or creating Main discussed what Main (1979) calls an internal representation p). properties an ideal IR should have, including one-to-one mapping between outlines and JRs, ease of performing Euclidean transformations (changes of scale, rotations and translations) and amenability to analytical treatment, such as constructing similarity coefficients. Digitized information, for example, is an IR but not ideal as it requires large numbers of coordinate points for accurate representation. In addition to his profile technique, Main suggests using local curve fitting techniques when representing the outline of a shape. Spline functions are of this nature and have seen extensive development

DWIGHT W. READ

METHODS IN ARCHAEOLOGICAL RESEARCH

in computer graphics for locally fitting curves and surfaces. They hold much promise as a general solution to archiving two and three dimensional shapes. Global techniques such as the fitting of Fourier functions are also possible, especially with elliptical Fourier functions that can represent complex planar curves (Kuhl and Giardina 1982). The second problem is to find a representation that also has interpretation, as Borillo (1977) has discussed, in the context of archaeological theory. Most of the archiving methods given in these papers (including spline functions) do not satisfy this criterion. Read's (1982) system comes closest, but is still a first approximation. Better understanding is needed of the process by which shapes are formed or constructed, e.g., Krause's (1984) work on pottery production sequences using formal grammars.
FEATURE SHAPE

Angell argued that Thom makes a circular argument through first using

an estimate of 6 to determine the integers m,, and then using the mi to justify the estimate for the "megalithic yard," 6. Angell used the model to
determine appropriate statistical tests of the various assumptions made by Thorn. The re-analysis by Angell showed that Thom's conclusion of having a "megalithic yard" is not supported by his data. In two of the papers (1976, 1978b) Angell addressed the problem of fittinga curve to the approximately circular shape of the stone rings. In the earlier of these two papers he developed a polygonal method and in the latter one a method that is a modification of the well-known technique for drawing an ellipse by using two fixed points and a cord attached to these two points. In effect, the construction technique uses a sequence of elliptical arcs to match the outline of the megalithic stone r i n g s . Both methods used a computer based, iterative least squares solution (based on a minimization algorithm due to Powell (1964)) for fit between the constructed curve and the cosrdimte location of the stones. In the fourth paper (1979) Angell developed an "explanationn for the megalithic rings by suggesting that they were devices for recording and retrieving certain special days during the year. As part of the argument, Angell developed yet another way to fit a curve to the stone rings, this time based on the changing orientation and length of a shadow for an upright pole during the year. The model Angell developed uses two coefficients to d e h e a curve: 6 - the angular altitude of the center of the sun, and d - a fixed offset from the pole from which shadow lengths are measured. Four other coefficients are used to orient a constructed curve (origin, reference point, angular difference of coordinate systems and i n g . Angell again used an iterative fitting scale) with the megalithic r procedure based on Powell's algorithm and was able to get good fit to rings which vary from "flattened circles" to "egg-shape." In this model the stone locations correspond to calendrical days, hence the stones serve as "reminders" of certain special days each year. The author commented that "AU existing theories reflect to some extent their author's particular interest be it engineerin& astronomy or mathematics. The present author therefore submits his theory as equally acceptable. Whether these hypotheses are fact or fancy will never be knownn @. 12) -caution that is exemplary but often ill-heeded. Leaving aside the interpretation to be made of the stone rings,this suite of papers highlights the problem noted above with measurement of artifact shapes: Is the goal an archival or an interpretive/explanatory one? These can overlap; eg., Angell's procedure based on shadow lengths is both archival and was constructed to provide a hypothesized interpretation. Angell's procedure also satisfies Read's criterion of non-redundancy; the shape construction procedure has associated with it a minimal number of coefficients needed to recreate the form; these coefficients then serve as

Whereas artifacts are single entities with form, features are composed of several smaller units which are combined to make a larger whole. These raise the question of what design principles were used to locate the smaller unit within the larger structure. More mundane features such as firepits, etc., have had less speculation about design principles than more spectacular features; many of the latter have also been given interpretation as devices for storing and retrieving astronomical and calendrical obsewations. Interpretation of features, structures and the like as early calculating or information storage devices depends upon l i n k i n g aspects of form with appropriate physical obse~ations,hence requires delineation of design principles.

Megalithic Stone Rings


In 1967, A. Thom published a treatise on megalithic sites in England and claimed to have discovered a number of design principles, including a unit of measurement called the "megalithic yard," or quantum, equal,to 2.72 It 0.003 ft. He said it was used as a standard for the construction of the more-or-less circular stone rings. Thorn's work on isolating this quantum was challenged by Angell(1976, 1978a, 1978b, 1979) on both statistical grounds and failure to consider alternative design principles that could be used to construct the circles. s AngeU. (1978a) recast Thom's argument in h e form of a statistical model: where yi is the diameter of the ith stone ring, 6 is the "Megalithic Yard", mi is a positive integer, fi is the error in locating the center of the circle, E , is the error term for the measurement of y,.

DWIGHT W. R E A D

METHODS IN ARCHAEOLOGICAL RESEARCH

its representation. It should be noted that this number of coefficients need be minimal only for the specific rule which relates the numeric values for the coefficients to a specific form. Further, the rule becomes an assertion about how the form can be constructed, hence is interpretable as having cultural meaning. The rule not only states how to construct the form from the coefficient values, but in so doing identifies a potential set of principles underlying production of that form. From here, it is a small step to the idea of a grammar of artifact shape, a topic taken up in a 1986 SAA symposium called "Form and design in archaeology: A grammatical approach" (C. Chippindale, organizer). Quantitative structures now become qualitative structures, reversing the earlier shift from quality to quantity as the means for representationof structure in data. Postholes: Rectangular and Round Structures Another group of papers (Cogbill 1980; Fletcher and Lock 1981,1984a, 1984b; Litton and Restorick 1984; Bradley and Small 1985) considered the analysis of postholes for regular pattern that may have corresponded to round or rectangular structures. The papers by Fletcher and Lock and by Bradley and Small examine the likelihood that a regular shape (rectangular for Fletcher and Lock, circular for Bradley and Small) could be a chance occurrence. In their paper, Fletcher and Lock (1984a) are concerned not only with methods for determining if the number of rectangles is greater than what l s o testing intuitive preconwould be expected in random data, but are a ceptions that structures in English hillforts are generally between 2 and 4 meters in size and are square. The authors establish (1984b) that the expected number of rectangles in N randomly distributed points over an area of A m is given by:

between postholes. From the Poisson distribution the expected number of instances can be determined in a collection of points where a subset of n points will fall on the circumference of a circle with tolerance E in its estimated radius:

E (number of circles) = g(n)pnR 2 n - 2 A ~-), n


where p = Poisson parameter, R radius of the circle, g(n) is determined by Monte Carlo simulation (tabulated in Bradley and Small (1985). The authors use the model to show that for the Aldennaston Wharf excavation (Bronze Age, England), two candidates for circular structures could just be chance occurrences, whereas for the South Lodge Camp excavation (Bronze Age, England), two candidates for circular structures are unlikely to have arisen by chance. The statistical conclusions are corroborated by other data. Shape of Corbelled Domes In a series of papers (Cavanagh and Laxton 1981, 1982,1985; Cavanagh et al. 1985), the shape of corbelled domes was examined by fitting a power h e dome. In the last paper of function f (x) axb to the interior curve of t this series, the question considered was whether the dome plus the supporting wall formed a single curve, or whether it was the join of two curves: one representing the dome and the other the supporting wall. To test the hypothesis, a two-phase regression model W e y 1971) was used:

E (number of rectangles) = Me-%/2(1-

E);

where E = p(1 - e-7, p = (N 2)2(b a) TA, t 2.41 T2(N 2YA, T tolerance for agreement between opposite sides, a, b length of the sides of the rectangle. Random data are generated corresponding to the parameters for the excavation and the authors (1984a) use the above model to demonstrate that the number of rectangles found in the excavated data significantly exceed what would be expected by chance. Their detailed analysis goes on to confirm the preconceptions about the hillfork based on visual inspection of posthole locations, yet it also shows that identification of square structures by eye is problematic. Whereas the data used by Fletcher and Lock is relatively dense (935 postholes in 2,664 mZ), the data used by Bradley and Small are quite thin. They based their analysis not on the random distribution of points over a region, but on a Poisson distribution for the nearest-neighbor distance

- -

where: x, < x, d x, + E, are N(0, 03 distributed error terms, 8, d l , d,, n,7, u2are parameters to be estimated. Maximum likelihood methods were used to estimate the parameters and it was shown for the corbelled domes that there is a change point in the curve for the corbelled tholoi, i.e., they are best described as the join of two curves, but not for the corbelled nmghi types of domes. Minimwn Number OflndivkhIs While counting the Minimum Number of Individuals (MNI) in a faunal collection is not problematic, it represents a recurrent problem of measurement what can be directly measured is not the phenomenon of interest. Faunal studies aimed at reconstructing subsistence strategies must estimate data on quantities of animals hunted and diversity of species procured by modeling the relationship of these measures to bone part frequencyin the faunalassemblage. The first published use of MNI for faunal material in archaeological

,,

DWIGHT W . R E A D

METHODS IN ARCHAEOLOGICAL RESEARCH

sites dates back to work by Inostrantsev (1882) in Russia (Casteel 1977: 125). S i then a number of papers have appeared, including the widely cited paper by White (1953), that discusses methods for estimating the MM (references in Casteel 1977). The comparison of methods for computing the MNI, the relationship of the MNI to the original population of animal kills, and the effect of sampling on estimating both the MNI and the frequency of species represented are three topics taken up in a series of works by Binford and Bertram (1977), Casteel (1977), Grayson (1978, 1979, 1981, 1984), Lie (1980), Fieller and Turner (1982) Wid and Nicho1(1983), and Horton (1984). Horton's paper and Grayson's book (1984) review the main issues, including Casteel's failure to distinguish MNI from the original number of kills. As pointed out by Fieller and Turner (1982), Casteel (1977) compared erroneously a measure proposed by Krantz (1968) for computing the original number of kills with a measure given by Chaplin (1971) for calculating the MNI. Krantz's measure was based on the proportion of pairs of bones to single bones. Lie (1980) used a different estimation procedure for the original number of kills. He assumed that the disparity between numbers of left and right bones is due to chance effects leading from the original population of bones to the excavated faunal assemblage. Fieller and Turner correctly pointed out that extreme disparity between left and right counts may be due to selective procurement and/or preservation; hence Lie's estimation ispefective. In turn, Fieller and Turner suggested using an estimate based on a capture/recapture model used in animal ecology, with the capture sample being, say, left bones and the recapture sample, right bones. However, this IS an argument by analogy that attempts to treat as identical in their consequences two very different processes: (1) an animal being captured, marked and recaptured in the ecological context and (2) bone material brought into a site, but not necessarily in paired form, then subjected to dest~ctive forces kadiig to a final number of unpaired left and right bones versus paired left and right bones. Wild and Nichol (1983) criticized Fieller and Turner's methods on methodological grounds as it requires the analyst to correctly match pairs of left and right bones from the sameindividual in the assemblage. While Horton kept clear the distinction between MNI and Krantz's estimate, Horton gave the latter the status of a factual, deterministic relationship (1984: 268) and used this to criticize Grayson's (1979, 1981) statistically well-argued conclusion that both the measure for MNI and the number of species can be highly affected by sample size (see also Kintigh 1984). Hence, Grayson argued, these cannot be assumed to have analytical usefulness. Grayson (1981) concluded from the statistical analysis that there is a "need for more detailed analyses of the quantitative structure of

rrrchaeological and pdaeontologid faunas, and of the causes of this .&cture';@86). Such an approach was used by Binford and Bertram (1977) when they attempted to retrodict o r i w bone distributions from the current bone distribution via modeling the unequal rate of bone destruction by bone part. Binford and ~ e r t k nmodeled rate of bone destruction via the differentidequation: where S surface area, V = volume, D density. Although it is supposed to be a model for the rate of bone destruction, the model asserts that bone density (D) varies with time (left side of equation), which makes no sense. On the right side of the quation, it is not clear why destruction should be proportional to the ratio between surface area and volume; rather it would seem that it should be proportional to S alone, for as they comment "bone with more surface area is destroyed more rapidly" @. 112). Next, the ratio SIV is incorrectly W = R, the radius) assumed to be constant (consider a sphere where S and with this assum tion the equation reduces to dD/dt = -AD, whose soIution is D (t) It would seem more reasonable, in terms of their assumptions, to model bone destruction via d m -aSD -- -aVu3/D (since surface area = Volumeu3, yielding V(t) (Vh'3 am)'. Graphs of the equation derived by Binford and Bertram were then used to argue that the percetage of bone surviving at time t will be a sigmoid curve with horizontal asymptotes at 0 and 1. The latter curve was modeled as follows:

- hi. --

where: F -. proportion surviving, X = density of the bone part, PI,P,,P3 are model parameters. This model was expanded to include terms representing different age classes for the animals and then fitted to data on MNI's (but see Grayson's (1984: 88-90) criticism of the MNI definition used by Binford) from a Navajo winter sheep site. The value of r2 for the data is high, but not unexpectedly so since there are only U data points and 8 parameters are being estimated. Further, the equation for F is predicted upon bone loss being due to attritional effects whose magnitude is proportional to bone density - a reasonable assumption for the data they used. Binford and Bertram used the data on sheep densities to re-analyze the purportedly hominid collected bone material found at Makapansgat in South Africa. The hypothesis they examined is wheher or not the differential frequency of bone parts at Makapansgat can be accounted for by an attritional process as modeled by them and acting on what were originally

DWIGHT W. READ

METHODS IN ARCHAEOLOGICAL RESEARCH

whole skeletons. Unfortunately the authors do not provide any test for goodness-of-fit; however, their "best fitn has a residual sum of squares = 14,672 compared to a value of 15 for the Navajo data, strongly suggesting lack of fit. Nonetheless, the authors argued that the attritional model accounts for the distribution and concluded that "there is absolutely no busis for the assumption that the horninids present played a behavioral role in the accumulation of the deposit" @. 148, emphasis added). The h i s conclusion turned out to be premature for later Binford assurance of t asserted: 'The interpretations Jack Bertram and I offered regarding the Makapansgat fauna (are) open to b e i i seriously wrongn (1981: 225). Much of the controversy in these articles on quantitatively analyzing faunal remains seems to result, first of all, from not accurately delineating the processes by which the empirical data are structured and second of all, from not clearly expressing the relationship between empirical data and mathematical formalism, on the one hand, and the interpretations to be drawn from the mathematical formalism, on the other hand. The formal de6nition of a measure (eg., the MNI) and classification of the material (e.g, left side, right side, matched pair) is a mapping from the empirical to the symbolic domain. Determination of precisely how a frequency distribution may vary through time when acted upon by a specified process (e.g, constant, possibly stochastic, rate of destruction independent of classification into left or right side) is an argument to be developed through the mathematical formalism. Description of how measurements are to be made (e.g., frequency counts based on excavation units) and assessment of biasing effects that may be introduced (eg., effects due to sample size and method of sampling) connect the empirical and the symbolic domains. Finally, conclusions obtained from deductions made in the symbolic domain must be accurately related back to the empirical domain for evaluation and validation. Of these authors, Binford and Bertram alone attempted to develop a complete argument; it is unfortunate that a number of conceptual errors mar their argument.

Measurement of Time: Seriation


No attempt will be made here to review the more mathematical side of seriation for two reasons. First, Marquardt (1978) has provided an extensive review and discussion of the various methods that have been developed to senate a collection of units. Secoyd, the mathematical side of the problem has essentially been resolved through an algorithm due to Ihm based on correspondence analysis (e.g., Ihm and van Groenewoud 1984) and available as a suite of computer programs (Scollar et al. 1985). Consequently, focus here will be on issues arising from interpretation of the ordering produced through the seriation analysis as a time scale. Not discussed will be the several papers (eg., Gelfand 1971; Landau and de la Vega 1971; Wilkinson 1971, 1974; Cowgill 1972; Graham, Galloway, and

scollar 1976; Laxton 1976; Laporte 1976) that have either considered the problem of efficiently finding a seriation from an incidence matrix, or from an abundance matrix using principal component analysis and/or multiple dimensional scaling (eg., LeBlanc 1975; Marquardt 1979). The mathematical definition of the seriation problem was presented in part I. Less formally, the problem is to determine an ordination for a series of objects (graves, sites, etc.) that can also be interpreted as a time scale. Construction of the ordinal scale is a mathematical problem; associating the scale with time links the mathematical argument to the processes structuring the data and remains the more problematical part. The linkage depends upon assumptions made about change in frequency of objects through time. Whether these assumptions are satisfied by the objects in question relates to arguments about the appearance, persistance and disappearance of artifact forms through time and across space. But even more fundamental is whether or not the problem is well defined from an archaeological perspective. For a seriation of units to be interpreted as a chronology there must be a temporal, ordinal scale for the objects; i.e., each unit must have associated with it a single point in time. Objects such as graves are often constructed (with notable exceptions!) at essentially a specific time, thus this is not an unreasonable assumption for graves. Contrariwise, a settlement has a history, hence occurred over a time period. One settlement does not occur "before" or "after" another settlement except in terms of some distinguished point in their respective life cycles, such as date of founding, date of abandonment, etc. Further, the material found on the site and taken as a single assemblage is a compression of time and, hence, has a complex relationship to any distinguished temporal point for the settlement. The total number of artifacts of a given type is affected not only by the "popuLarity" of a type (assuming the validity of that model), but by the settlement population size which may be varying through time as well. If we let p,(t) be the probability of selecting or producing artifact of type i when an artifact is acquired or manufactured at time t, n(t) the number of persons in the settlement at time t, and a(t) the acquisition rate of artifacts per person at time t, then fi(t), the number of artifacts of type i added to the settlement at time t, will be given by: h(t) a ~i(t)n(t)a(t). The total number, F,, of amfacts of type i accumulated from time & , to time t, in a settlement will be given by:

[/6(t) dt

-1

p,(t)n(t)a(t) dt.

Now suppose a value for I, call it t;, is associated with settlement j (e.g., ti

DWIGHT W. READ

METHODS IN ARCHAEOLOGICAL RESEARCH

might be the midpoint between when the settlement was formed, b, and when it was abandoned, t,). The seriation goal is to find a correspondence between 4 and ti;. But there are several factors which iduence the value of F, for a single settlement. These are: (1) the functions pi(f), n(t) and a(t) (where the later two functions are assumed not to be site specific and a(t) to be identical for all persons, for simplicity); (2) the value to be assigned to each site, j ; and (3) the bracketing values to and t, for site j. The assertion that type Frequencies first appear, then increase in popularity and finally disappear only relates to the function pi(t). Whether a ,similar claim holds for the accumulated quantity 8 of type i in site j depends upon the properties of the other factors. Three implicit simplifying assumptions that are usually made are: (1) n(t) and a([) are constant and do not vary with time, (2) two sites with overlapping times of occupation also had the same values for t, and f,, and (3) there are no time delays in the system; i.e, diffusion of popularity change is instantaneous throughout the system. The extent to which these are unrealistic assumptions depends upon the relative time scales within the system. For example, if the time scale for diffusion of concepts is short in comparison to the overall time range for all of the sites and to the time differences between sites, then the assumption of instantaneous diffusion is approximately valid. However, when one attempts to use seriation to chronologically order sites "separated" in time on a scale comparable to the scale for measuring delays in information d i i o n , then the simplifying assumption will not be valid. Similar comments apply to the other factors. Paradoxically, then, earlier successes in using seriation for chronologically ordering sites when the sites were well separated in time does not allow the procedure to be extended to fine chronological control. This has been demonstrated with simulated data by de Bmos (1982) who showed that overlapping site occupations, differing lengths of occupation, etc. drastically affect one's ability to recover via seriation the correct time sequence used in the simulation. Another confounding factor is the multi-dimensionality of factors affecting frequency of artifacts. Change in popularity through time is but one potential dimension along which the data are structured. As noted in Part I, correspondence analysis (see also Bertelsen 1985, Holm-Olsen 1985 for other applications of correspondenceanalysis) has been suggested as a nominal-level technique that can display these distorting effects even if it does not "comt" for the multi-processual, multi-structured character of the data. Seriation is also but one means to ordinally rank sites as a way to infer a chronology. Any measurement, or group of measurements, on artifacts which varies regularly through time has the same potential. If a measure-

ment, x, varies monotonically with time, t, then the relationship linking t and x, say x = f(t), can be inverted to yield t = f-l(x). If, however, the graph of f(t) has reversals, then the inverse function is not everywhere defined; that is, the same measurement, x, can correspond to more than one point in time. Braun (1985) used the idea of a measurement varying with time to date Middle Woodland on the basis of sherd thickness. In his analysis Braun took up the problem that radiocarbon dates are estimates with errors, yet it is desirable to use smoothed (filtered) data. Braun resolved the problem through application of probability-normal (PROBNORM) methods. Braun applied the method to 56 dated samples of pottery, constructed an estimated graph for the Function f(t) and determined confidence intends for the estimated dates. He observed that
the curve of X provides far greater accuracy than any previous temporal seriation for the region, with two Further advantages; We can estimate the precision (probable error) of any predicted date, and we know in a d m e to whai contexts d may be reliably applied @. 534, emphasis added).

Braun aptly draws attention to precisely the property that has generally been lacking in methods for constructing chronologies from incidence or abundance matrices.
ASSESSMENT OF DATA STRUCTURE

Structure in D a t a Sets: Clustering Methods and Log-Linear Models Delineation of structure in data sets has proceeded from two directions: (1) inference of groupings through measurement of similarity between pairs of objects and (2) inference of structure among already defined groupings through assessment of variable interaction. The former is based on clustering methods and the latter has been addressed through log-linear models. The former uses variables to delineate the relationship of objects to one another and the latter uses objects to delineate the relationship of variables to one another. As methods, the two approaches are distinct; obfuscation entered through each being touted as the means to achieve the same goal, namely, construction of a typology. Doran and Hodson (1975: 168) and Hodson (1982) suggested that Spaulding erred by placing the problem in the context of the analysis of contingency tables whereas it was actually a problem better handled by numerical taxonomy; Spaulding (1982) countered that when a contingency table is reduced to s h h i t y among objects, 'poverty-stricken results emerge" @. 13). Cow@ (1982) reviewed the arguments and concluded that when one is working with nominal variables, a variable association approach is more useful, whereas an

DWIGHT W. READ

METHODS IN ARCHAEOLOGICAL RESEARCH

object-clustering approach is more feasible for interval or ratio scale variables. His conclusions are reasonable but sidestep Spauldiig's premise

t h a t
definition of nominal variaMes and investigation of their i ~ t ~ r ~ e l a J i 0 ~is h ia p~ fundamentally important aspect .. . because of the d m connection between nominal variables and cultural patterning of human behavior (1982: 19)

~t the other extreme, the following is an easily solved, wen defined criterion: x, y E Pi if, ahd only if, d(x, y) d

4.

Read (1982) argued that the difference is one of kind: cluster methods lead to paradigmatic classifications,whereas Spaulding's notion of types as non-random association of attribute states is an assertion about structure within a paradigmatic classification. The ceUs of the contingency table defined by the amibute values are already types (Spaulding 1982: 13) and what the contingency table analysis tries to uncover, guided by methods such as log-linear models, is how these types, whether defined by nominal distinctions or inferred through object-clustering methods, are interconnected as a structured set of classes.
CLUSTERING METHODS

The goals of clustering methods are sound; problematic is the means for achieving the stated goals. Intuitively, the analytical goal is to sort objects into groups that exhibit internal cohesion and external isolation Pragmatically, internal cohesion is taken to mean roughly a set of points in n-dimensional space that are close to one another and separated from other sets of points. "Closeness" is based on a suitable norm, such as the L, norm used to compute Euclidean distance, and objects "close" to one another are taken to be "similar" to one another. More formally, (see Read 1974), the problem can be construed as follows: Given a set P of points, find a partition P = {P,, P,, . , P,} of P (that is, subdivide P into disjoint subsets P,) that satisfies some specified criterion, C. The problem is to define the criterion C. For example, the criterion might be: of all possible partitions of P with n subsets, select the partition P that minimizes

..

where: x, y E Pi for some i, 1 < i < n, d(, -) is a normed distance t measure. While the criterion is well defined, it assumes the value of n is known and does not lead to an exact solution due to the extraordinary rate at which the number of possible partitions grows as the size of the set P increases.

The latter is also known as single linkage clustering. The partition is easily constructed and has attractive mathematical properties (Jardine and Sibson 1971), but pragmatically it suffers from the well-known phenomenon of chaining. These two criteria express the dilemma of clustering methods: either one attempts to use a criterion that is satisfactory but impossible to implement except on very small sets, or one uses a criterion that does not have an effective statistical implementation. This leaves clustering procedures as a series of approximations that mitigate against using them in anything but an exploratory manner as the results of one procedure may be at odds w i t h the results of another (Aldenderfer and BlasMeld 1984; e.g., Ottaway 1974: 221). Interpretation of dendrograms becomes an art and dependent upon one's subjective impressions of what constitutes a "good" cluster. Further, statistical analysis of "signiscant" levels of association are of no help if the whole constructionis suspect. Despite the problems, applications abound. Clustering methods, however much they lack solid theoretical justifcation, are grappling with a widely shared problem: data are structured along the lines of similarity/ dissimilarity and partition along this dimension is fundamental to further analysis. Cluster techniques are seen as providing "objective" answers where data have reached Moberg's (1971) complexity threshold. Whereas one does not need to use k-means clustering or Ward's hierarchical clustering to separate lithics from pottery from bone, sorting graves by similarity as measured across dozens of grave goods seems to demand a level of "objectivity" that is not provided by "subjective" assessment. With the exception of Christenson and Read (1977), OShea (1985) and Read (1987b, 1988), little has been done in the archaeological literature to assess critically whether or not clustering algorithms accurately delineate structure or merely create partially correct solutions. Results from cluster analysis are sometimes compared against groupings formed by more intutive methods, but the latter are already self-evident and so agreement does not determine if the algorithm can group data at a finer level. More often than not the results are taken at face value (e.g., Simek and Larick 1983). Occasionally some discussion of the mathematics behind an algorithm is included to justify the choice of one algorithm versus another (e.g., Bieber, Brooks, Harbottle, and Sayre 1976; Carr 1985c), or corroboration with other data are used to show the relevance of the groups (e.g., Stafford and Stafford 1983; Carr 1985~). In excep-

DWIGHT W. R E A D

METHODS IN ARCHAEOLOGICAL RESEARCH

tional cases (e.g., WhaUon 1984) it is recognized that the algorithm is not as critical as one's formulation of the problem. Cluster analysis gives the sense that something has been done. Dendrograms are produced, they can be assessed for implied groupings, these groupings can be given interpretation, and so on (eg. Philip and Ottaway 1983). The danger lies in the failure of cluster analysis to provide information signalling that the groupings found may be only partially correct. Christenson and Read (1977) demonstrated that cluster procedures may simply fail to distinguish groups that would be recognized by any archaeologist (see also Read 1985, 1987b). Conversely, Read (1988) shows via simulated data that cluster procedures may find what appear to be distinct groups when in fact these crosscut distinct, well defined structure in the data. Read (1988) argues that the problem stems from inclusion of measures irrelevant to the data structure; the dilemma lies then in the lack of a priori information that informs the researcher as to the variables that are irrelevant Aldenderfer (1987) is critical of what he sees as the incorrect identification that has been made between numerical taxonomy and cluster analysis and comments that "Clustering methods have been tarred with the same brush that has blackened numerical taxonomy" and this, he suggests, has lead to "premature closure of thought about the value of these procedures in archaeological classification" (p. 25). The value of these procedures does need a sounder footing. Without algorithms that demonstrably recover structure in data, and without independent means of verification that are not circular, one runs the risk of creating analyses that falter on their built in distortions. Cluster procedures do not "work" in general. This is not to say there are no cases where the results are valid; rather, there are no statistical means to assess whether or not the results are correct, just a partial, or even incorrect delineation of structure for the data being explored (Read and Christenson 1978; Read 1988). Whallon (1987) has argued for the use of simple descriptive statistics and graphical display techniques as a prerequisite to any analysis. Histograms and scattergram plots can be more informative in exploratory work about data structure than complex statistical techniques (e.g., Geyh and de Maret 1982), if only for the simple reason that statistical techniques are based on assumptions about data structure, hence provide information structured according to that assumption. ~ f e r r and a Carlson's (198i) example of "structure" found in random data through factor analysis illustrates the problem well. Dendmgrams from cluster analyses in which groupings are not clearly evident should be viewed with healthy skepticism. If groupings are selfevident (e.g. Impey and Pollard 1985) then one has the basis for further

exploration. Lack of clarity as to what level to use as a cut-off point for defining groups may be more a signal that the methods for data selection and variable choice are inappropriate, rather than reflecting some important subtlety within the structure of the data. However, when groupings ate self-evident, it can be sobering to reexamine the analysis using simple statistics. For example, Impey and pollard (1985) found three groups in a set of pots made by three different potters (but do not indicate whether the groups recovered correspond exactly with the potters). They measured 13 variables and used Ward's clustering method, average linkage clustering and a plot of principal componentsin the process. Simple examination of the variables and data used by the authors shows that the same results could have been obtained by looking for anti-modes in a histogram of two of the variables (internal diameter at the lip and overall height): evidently one of the potters was making taller pots than the other two, and the latter two made their pots with different internal diameters. The cluster analysis seems to work here not so much because of the algorithm chosen and the multivariate approach, but because the data had structure that was well expressed and easily seen with the variables as measured.
CONTINGENCY TABLES

Since log-linear modeling was introduced into archaeology by Read (1974), several papers (Clark 1976, 1984; Butler, Tchemov, Hietala, and Davis 1977; Hietala and Stevens 1977; Spaulding 1977; Hietala and Close 1979; Hietala and Larson 1980; Hietala 1983, 1984; Leese and Needham 1986; Lewis 1987) have appeared that discuss both the underlying theory and applications. Briefly, the procedure is based on (1) specifying a model for variable interaction which determines the product form for computing the expected cell frequencies in a multi-way contingency table and (2) u s i n g logarithms to linearize the products. Thus, in a two way table with variables A and B, the cell values may be fit exactly by a product of the form: whereni. qj,n., qj,n. . -87 I n,,,ev - n . ./(ni.)(n.j). Then log n, -- logn. log(n,./n. .) log(n. ,In. .) log(n,/e,), which is of the form:

==?

=?

.+

where ai is a row effect, Bj is a column effkt, aB, is an interaction term for the (i, j) cell.

DWIGHT W. READ

METHODS IN ARCHAEOLOGICAL RESEARCH

Here the model is exact and is called a saturated model. If the interaction term, a&, is deleted, then one has the stochastic model: where: eU is the error term for the (i, j ) cell. The model for a two way table may be generalized to an n-way table; e.g., for a three way table the saturated model will be of the form: where a& ~ r , and ~k ,r ? ,!, are two-way interaction terms, U B ~ , is , ~a three-way interaction term. By deleting the three-way interaction term and possibly one or more of the two-way interaction terms, a stochastic model is formed, e.g., log(ntjk) =p a , 4 rk (ar,) E,,~would be a model which includes two two-way interaction terms. Fitting a model now reduces to a problem of parameter estimation in a linear model using maximum likelihood estimators. Models are tested for for a goodness-of-fit using the G2statistic: e.g., G2= 2 x 2 n,jlog(n,/e,,) two-way contingency table, which has a chi-square distribution if the model is correct, with (r - l)(c - 1) degrees of freedom under the typical hypothesis of row and column independence. Analysis generally proceeds by specifying a hierarchy of models and then testing for the simplest model in the hierarchy that fits the data. Fit to the data is determined by (1) the specified model having non-sigificant deviation of expected from observed values and (2) no simpler model in the hierarchy satisfying criterion (1). Criterion (2) is tested by showing that the posited model has significantly better fit than a simpler model as i l l have a chi-square measured by the difference in G2values (which w distribution if the models are hierarchically related) for the respective models. L~g-~near analysis is not sufficient by itself. As noted by Read (1974), Scheps (1982) and Leese and Needham (1986), one wants more than just a model of variable interaction One also wants to know which cells or combiiation of cells determine the interaction effects. Leese (1985) and Leese and Needharn (1986) discussed a simple procedure called Iterative Proportional Scaling in which the row and column totals are scaled to the same value. Under random effects, every cell should now have the same value, hence cells with extremely large or s m a l l values mdicate "deviantn cells. Though simple in implementatioh for two-way tables it becomes unwieldy for higher dimensional tables (Leese and Needharn 1986: 10). More problematic is that the deviations in the cells are not independent since a 2 X 2 table only has 1 degree of freedom. Hence, if the value Oo

in a two-way table is suspected to be deviant and replaced by a value, Oh, so that now the expected and observed values agree for the (i, j ) cell, the marginal totals will also change and all other expected values in the table will change. Hence one needs to use a step-wise procedure for identifying deviant cells. The only difficulty that arises is that OI,. is not ev, the expected value for the (i, j) cell based on the original marginal values. Read (1974) discussed an iterative procedure for finding the value Oil. Alternatively, one cau compute: Read (1974, 1982) illustrates the technique with Aurignacian endscrapers and Scottsbluff points, respectively. In both cases the deviant cells are used to define types in accordance with Spaulding's notion of a type as a non-random association of attribute values. CLuistenson (1987: 91-93) has applied the method to an analysis of ground stone from Black Mesa. Yet another way to partition chi-square values that has been used in population genetics is applied by Simek and Leslie (1983) to data from Le Flageolet. The technique tests for homogeneity in populations partitioned by the same categories. A two-way table is formed where one variable has for its values the categories and the other the different populations. The total chi-square value is partitioned into two parts: one which measures heterogeneity among the populations and one which measures the fit of a model to the observed values and pooled over the populations. The authors use material types as the categories and tool types as the populations. They find that all three chi-square values are significant, hence the populations are heterogenous. The model of independence of tool type and material can be rejected and no other model is consistent across the tool types. Spatial Analysis Two kinds of spatial analysis w i l l be distinguished here: (1) characterization of the spatial arrangement of artifacts/features in a region or in a site and (2) statistical modeling of the spatial location of objects, or "predictive modeling." The first of these two topics has had extensive development but no analytical resolution as of yet. Earlier assumptions about the homogeneity of structuring processes across space have been shown to be too simplistic and more recent work has focused on ways to isolate local effects and to dissect spatial arrangements into their component parts. Predictive modeling will only be considered briefty as it has been primarily an application of inferential statistics. This section will end with a brief discussion of an extension of autocorrelation methods for examining spatial arrangements qualitatively as well as quantitatively.

+ + + +

DWIGHT W. READ SPATIAL ARRANGEMENTS

METHODS IN ARCHAEOLOGICAL RESEARCH

51

Intersite Spatial Arrangements


The primary data of archaeology is spatially distributed. The spatial arrangement as recovered by the archaeologist reflects a past society's utilization of space and a society's utilization of space is realized through its internal organization and external relations with other groups. Social organization is affected by material considerations and mediated through the meanings imbued upon space by the ideational system of a past society. Hence analysis of spatial arrangements relates, even if indirectly, to the fundamental processes of a society as a reproducing system. The rationale for undertalang analyses of spatial arrangements is clear; effective means for so doing are less obvious and have suffered from simplistic assumptions about the correlation between material and nonmaterial culture (Hodder 1978a: 24; 1978b: 110-11). One suggested approach for breaking away from simplistic correlations has been to use the following paradigm. Given that such-and-such property of a spatial arrangement is said to have interpretive value, what are the statistical means for measuring whether or not that property is present in a data set? For example, the form of the decrease in the quantity of artifacts of a given type as one moves from a production center (eg., linear, exponential, etc. with distance) allegedly reflects processes of trade; statistical methods such as regression analysis allow one to characterize that drop off and, hence (in t h i s framework), provide the basis for interpretation (e.g., Hodder 1974; Hodder 1978c; Hodder and Reece 1977). In their book, Spatial Analysis in Archaeology, Hodder and Orton (1976) presented statistical methods appropriate to this paradigm (see also Orton 1982, Carr 1984). Topics discussed go from the more particular to the more general, from the single artifact type or site to several distributions and include a variety of techniques and methods: point pattern analysis, models for settlement patterns, regression analysis, simulation of artifact dispersal, trend surface analysis, spatial autocomlation, and association between arrangements. The general theme is one of method, a theme echoed in a number of papers which have attempted to devise or present methods of spatial analysis for archaeological research; e.g., contiguity analysis (Luton and Braun 1977 ,local density analysis ,(Johnson 1984) unconstrained clustering (Whall n 1984), the contiguityanomaly technique (Gladfelter and Tiedemann 1985) and semivariogram analysis (Schieppati 1985). The underlying assumption in this paradigm is that since form - the spatial arrangement - is the consequence of process, understanding the

form suffices for infening process and understanding process suffices for explanation and interpretation - an assumption later rejected by Hodder (1985). Hodder and Orton are aware that the mapping from process to form is many-to-one @. 239); yet, while they ofEter suggestions, they have no method for resolving the ambiguity of the inference problem. S i a r l y , ' they mention the need to show that equations or statistical distributions used to fit data need to have theoretical justification at the level of process, but the implications for analysis aimed at trying to achieve concordance between underlying models for analytical methods and underlying models for data structureare not considered. Fitting curves to quantity/distance plots and then inferring process from the equational form of the curve does not include time as a parameter, so one has as a model: where Q quantity, D = distance (from source or from production center). But processes that operate over space also operate in time, hence minimally the model must include time: Ammerman, Matessi, and Cavalli-Sfona (1978) observed that when a static model of the form Q = F(D) is changed to include a time parameter, one now wants to know the equilibrium values for this dynamic system. They point out that the exponential model for the quantity of a good such as obsidian being passed from one group to the next only has an equilibrium when all groups have 100% of the item. Ammerman et al. introduced a modiiied model of the form: where p the proportion of a group's obsidian passed to the next group, d drop (i.e, discard) rate, to make the model dynamic. The model is given explicitly in the form of a recurrence equation: where Q u ) = amount at site j at time t, p = proportion passed from site to sitej 1, d = drop rate. This has as its solution:

where Qa(0) . at sitej

quantity at site 0 at time t

0, QO)

equilibrium value

For d 0 one obtains the equilibrium solution of all sites having the i l l be linea~ly same quantity, whereas for d > 0 equilibrium values w decreasing with distance versus log(Q 0)). Though elegant, the model depends upon several smng assumptions: (1) p is independent of time and location, (2) d is independent of time and location, (3) the exchange system is linear in structure, (4) the same number of tools are in use at each site and (5) the number of sites is fixed and does not change through time @p. 189-89). In effect, the structure for the exchange system is assumed to be static; only its operation is dynamic. But structures are dynamic, not static. Even at the level of static structures the linear exchange model fails to capture the complexity of exchange systems. After reviewing ethnographic literature relating to the different ways in which goods are exchanged, Hodder (1978d) commented that "It may often be the case that the factors affecting the distributions are so complex and numerous that, whatever the original exchange mechanisms, the composite result reflects basic constraints such as distance, value, etc., but not the particular exchange mechanismsn @. 246). Hence, whether one sees the goal of studying spatial arrangements as linked to understanding the basic consmints," or linked to understanding the "particular exchange mechanisms," will determine the utility to be gained from examining the spatial structure of archaeological data.
Znfrasite SpariolArrangements

DWIGHT W. READ

METHODS IN ARCHAEOLOGICAL RESEARCH

(1) Spatiafiywriforrn processes. In an extensive review of methods for the a r r (198%) sugested that analysis of intrasite spatial arrangements, C researchers dealing with archaeological intra-site spatial arrangements have had four operational goals: (1) description of the fonn of the spatial arrangment, (2) delineation of artifact type specific clusters, (3) comparison of artifact distributions across different types and (4) spatial limits of multi-type clusters. Paralleling the operational goals are three inferential goals: (1) determination of the spatial limits of "activity am&"' (2) organization of tool types into "tool kits"and (3) the spatial dimensions of activities within a site (1985~:305-306). Though not stated in this manner, Cads operational goals Correspond to Borillo's mapping of tht empirical domain onto the symbolic domain, and Cads inferential goals correspond to the interpretation to be made of conclusions derived from mathematicallv based armunents within the m b o l i c domain. ; Carr n o t d that p&ious work on q&tiktive analysis of spatial arrangements has tacitly assumed that the assemblage was the consequence of a single process acting uniformly over the site (1985~:312), hence one assumed identity between analytical result and interpretive significance. Techniques, in this view, would "lead to the discovery of

. . . and thus provide something for the archaeologist to explain" @odder and Orton 1976:241; see also Whallon 1973: 267). With spatial pattern separated out as a property first to be measured and second to be interpreted, analytical emphasis is on method separated from process. For example, Dacey (1973) applied techniques, which were based on a Poisson distribution for frequency counts in a grid system for the site, to first measure whether end-scrapers, carinated scrapers, and burins are randomly distributed across the Sde Divshon site and second, to measure whether they are spatially co-associated. Whallon (1973) had essentially the same goals but used a method developed in ecological studies - dimensional analysis of variance - to decide on the scale for grid squares. Dimensional analysis of variance was used to determine the grid square size which maximizes the variance of the squared frequency count per grid square (see Carr 1984: 144-51 for the equations upon which the technique is bared). Whallon then used this grid square size to construct correlations among frequency counts for the items of interest. Highly intercorrelated items were grouped and given interpretation in terms of subsistenceactivities on the site. To treat spatial pattern@ as if it were a quantitative propem to be measured requires that one have a scale sensitive to variation along a unifomdaggregated continuum. The methods used by Dacey (1973) and Whallon (1973) depend upon frequency distributions derived from counts per grid square and ignore the spatial organization of those frequency wunts. Hence, they are only indirectly a measure of spatial stmcture. Methods have been developed, though, to provide direct measures of spatial structure. Of these, the nearest-neighbor statistic has been the most prominent in archaeological literature (see references given in Orton 1982, Carr 1984). The nearest-neighbor statistic measures deviation from a random distribution and, hence, provides an index that measures the numerical location of a spatial arrangement on a uniform-to-clustered scale. Several papers have discussed technical problems that arise in its application, such as the edge or boundary effect. For example, Pinder. Shimada, and Gregory (1979) gave a correction for the boundary effect pmblem based on a square region and also provide a significance test for the nearest-neighbor statistic. And McNutt (1981) provided a more general correction based on geometric considerations. This trend of going from "rough and ready" measures of non-random spatial pattern based on the Poisson distribution to more careful definition of the f o m that spatial arrangements may take on was extended by Hietala and Stevens (1977). They provided formal definitions of six kinds of correlative spatial patterning ranging from Uniform Aggregation to Uniform Segregation, and appropriate statistical tests for each pattern. The formal definitions are based on the respective magnitudes for the
pttem

DWIGHT W. READ probabilities PA and P, of a grid unit (their region R ) containing artifact type A or artifact type B ; eg., 'Two artifact classes, A and B, are said to be uniformly aggregating if, for every region R, PA (R) = PB(R)" @. 541). The method of analysis used several procedures, including the variance/ mean ratio test used by Dacey (1973), median split tables, and log-linear analysis of three-way contingency tables in which two of the dimensions represent the (x, y) coordinates for the grid system. Only the conclusions drawn from the log-linear tests are presented, however. By including the spatial dimensions in the contingency table, the spatial arrangement of frequencies, not just their pattern in a frequency distribution are compared, hence the method overcomes the non-spatially specific aspect of the variance/mean ratio based on a Poisson distribution for frequency counts. However, it is still a comparison of frequency counts rather than a direct comparison of spatial arrangements and ignores contiguity relations (Gladfelter and Tiedemann 1985). A method for the comparison of spatial arrangements which does not require reduction to frequency counts in grid squares was introduced in a paper by Berry, Mieke, and Kramme (1980) and later modified to allow for unequal artifact class sizes (Berry et al. 1984). The method compares the spatial arrangements of two artifact classes, A and B, by considering all possible subdivisions of the combined set of artifacts into two subsets so as to form a distribution for the average interpoint distances. The interpoint distance for the actual division into classes A and B is compared to this distribution for statistical signifiance. More precisely, let P be the set consisting of both artifact types with, say, a total of N artifacts, and let P, and P, be a partition of P,with n, and n, artifacts in P, and P,, respectively. Let E, and E, be the average interpoint distance for all artifacts in P, and P,, respectively. Let 6 (n,/N)e, (%/N)E, be the weighted average of E, and E,. Berry et al. (1984) derived an approximate statistical distribution for 8. The value of 8' for the partition of P formed by the two artifact classes A and B is compared to this distribution for the likelihood that it could have arisen as a chance partition of P. Rejection of d' as a chance occurrence implies that the artifact classes A and B have distinct spatial arrangements. The procedure does not characterize the nature of the spatial arrangements, though the authors discussed methods for assessing the arrangements in terms of segregation of artifact classes from one another and concentration (where spatial arrangements overlap but dBer in form) . (Berry et al. 1984: 63). The method does, however, depend upon a uniform process having acted to place the artifacts in their current location; a similar assumption underlies the method presented by Hietala and Stevens (1977).

METHODS IN ARCHAEOLOGICAL RESEARCH (1984) has disputed the simplicity of methods that assume spatially uniform processes and has argued instead that spatial arrangements are the product of multiple, spatially non-uniform processes. Heterogeneity in process (see Read 1985) led Whallon to advocate use of methods which make minimal assumptions about data structure. Whallon advocates unconstrained clustering, which "is hardly more than an elaborate approach to a descriptive summary or display of the data, or a series of such summariesand displays" (Whallon 1984: 275). Unconstrained clustering begins by constructing smoothed density contours for each artifact type which are then used to construct, for each data point, a vector of relative densities of the artifact types at that spatial location. Data points are grouped using the relative density vectors, and the found groups are plotted over the site and examined for spatial integrity (in addition to being grouped together through the cluster analysis). These groupings, along with their spatial association and internal composition are then "used for interpretation and reconstruction of the behavior or processes which created the spatial structure or organization of the data" @. 245). The details of the approach are illustrated with data from the Mask site, and Whallon finds reasonable agreement between interpretations made from the analyzed data and the ethnographic report (Binford 1978) on the Mask site. A similar strategy was used by S i e k and Larick (1983), Simek (1984a, 1984b) and Simek, Ammennan, and Kintigh (1985) in an analysis of the arrangement of artifacts in Le Flageolet. Their analyses were based on what Kintigh and Ammerman (1982) referred to as heuristic approaches to spatial analysis (see also Little 1985). The goal was to "develop and apply analytic methods, which provide basic, reduced descriptions of extant spatial patterning within an arrangement, but do not presume interpretation by the nature of the technique themselves . . . .The heuristic approach . . allows form-to-process arguments to be constructed by integrating analytic pathways for defining form in the archaeological record with the human ability to perceive and understand those forms in processual termsn (Simek et al. 1985: 229). The method is perhaps a bit less grandiose; essentially it is based upon using k-means cluster analysis to determine spatial clustering of artifacts (kintigh and Ammerman 1982) and then analyzing clusters for their diversity and heterogeneity ( S i e k et al. 1985: 230). When sample size effects in diversity measures (Kintigh 1984) were taken into consideration for the data from Le Flageolet, diversity measures were discovered to be within the range of values expected for random data ( S i e k et al. 1985: 238-39). Heterogeneity was measured by using Kendall's fau statistic as a nonparametric measure of association for the six artifact classes. The coefficient was measured across the clusters for each of the 2 cluster, 9 cluster and 11 cluster solutions obtained

(2) Spatially mn-unjfotm processes

inductive approaches. Whallon

DWIGHT W. READ

METHODS IN ARCHAEOLOGICAL RESEARCH

through the k-means analysis. Different patterns in the respective correlation matrices were found which they attributed to heterogeneity due to activity patterning @. 242). (3) Spatially non-uniform processes: deductive approaches. In contrast,

Carr (198%) suggested that primarily inductive approaches are insufficient as they still require knowledge about processes structuring the data to interpret multiple representationswhen these are not mutually consistent (1985~:319). In place of a primarily inductive approach, Carr (1985~: 322-23) suggested taking a deductive approach grounded in models for the structuring of data in a site. Carr commented that W o n proposes a geneml technique thought applicable to a diversity of intra-site data structures; Carr proposes a general dara stmcture common to archaeological sites and suggests methods congruent with that structure" (Carr 1985c: 323, emphasis in the original). The "general data structure" is to be "a geneml organizational model of fundamental mathematical character isticsof arnfactput&?ming"@. 326, emphasis in the original). The framework set up by Carr constitutes a break with previous work on analysis of spatial arrangements in that methods of analysis are now to be initially formed and selected so they will be in accord with derived models for spatial arrangements based on postulated processes. While Whallon (1984) recognized the discordance that has existed between analytical method used and fonnation processes, his solution has been to turn to a less constrained, more generalized approach to spatial analysis. C a r r has turned, instead, in the direction of greater spedicity of the presumed processes as a means to develop and/or apply analytical techniques which are sensitive to patterning implied by these processes. The use of a mathematically specified model and an emphasis on logical consistency between technique and data structure (Carr 1985b, 198%: 303) indicate that, for Carr, deduction is not b e i n g used in the pseudo-deductive sense of stating plausible implications of hypotheses as "deductions", but in the mathematical sense of deriving the logical consequences of first premises said to structure the phenomena in question. This would seem to suggest that a formal, axiomatic-like representation (eg, Read and LeBlanc 1978), or at least an approximation thereof, is to be used. However, though mathematical terms are introduced, the mapping from the empirical to the symbolic domain is only partial and sometimes inexact; e.g, calling "amibutes that entities possess" and "list of attributes" concepts from set theory (1985c:'330), or asserting that "activity sets" and "depositional sets" are not "mathematical sets" (1985~: 331). Definitions for mathematical terms are given in their interpreted, applied form rather than as abstract properties to be included in a deductive system. For example, Carr defines symmetry not as a tenn belonging to the domain of mathematical structures (see below for such a

definition), but as a property exhibited by concrete objects: "two types of are said to be symmetrical. . ." (1985~: 336, emphasis added). It is easier for the reader to have abstract ideas expressed in a more familiar idiom. However, there is also the potential for ambiguity b e i i introduced through the implicit implication that it is properties of the concrete situation that are the basis for the deductions, rather than the deductions being based on a system of abstracted relations which are then t o be mapped onto the concrete situation The difference can be seen in confirmation for an argument. Assertions about concrete properties rely on empirical observation for confirmation; assertions about properties of abstract structures rely on l o g i d consistency. Carr's goal is to determine the logically possible arrangements that can occur along a monothetic/polythetic dimension and a non-overlapping/ overlapping dimension for the spatial arrangement of artifacts. Once these are determined, they are linked to fonnation processes that can have these arrangements as consequences. F i y , techniques are isolated that make assumptions consistent with the processes and their associated spatial arrangements. Included in the latter are several similarity coefficients introduced to take into account the different spatial arragements and a new clustering algorithm that allows for overlapping clusters. The methodology is applied to the Pincevent site in France. Carr infers from the analysis that "forms of spatial variation - and the monotheticpolythetic dimension of organization that can be related to them - must be considered when choosing a similarity coefficient for defining site-wide 451); ie., sites are not the product of uniformly depositional sets" (1985~: acting processes.
(4) Dissection of spatial arrangements. Christenson and Read (1977) had pointed out that techniques such as factor analysis make global assump tions about homogeneity of data structure which may not be valid, yet methods to screen data prior to analysis seem to depend on knowing that structure. This they called the "methodological double bid" (1977: 177). Carr (1985~)noted that the same problem arises in the analysis of spatially distributed data: the data are structured by multiple processes with non-uniform sDatial arrangements. Hence thev reauire dissection into spatially homogen&us subsetsprior to use of techkquks that make global, 'homogeneity assumptions. Carr (1985a) discussed Fourier representation a potential and filte& techniques (see & 1985a for references) means to dissect a palimpsest (Fourier methods had been introduced earlier for use with spatial analysis by Graham (1980) who sugested using spectral analysis to characterize spatial patterning.) Or, in the terminology used by Read (1985) to classify modes of discordance, Fourier techniques are b e i i advocated for preanalysis of Mode 3 data: well-definedvariables, not well-defined population. Essentially, Fourier functions allow approximation of a periodic func-

DWIGHT W. READ

METHODS IN ARCHAEOLOGICAL RESEARCH MODELING THE SPATIAL LOCATION OF SITES

tion (defined over all the real numbers) as a sum of sine and cosine terms. For a singledimension the form of the general Fourier function is:
f (t)

L: ancos(2mtlz)+ Pnsin(2mt/t),

where t is the length of a period for a wave, nlz is the frequency of a wave, a,, /3, are the amplitudes of the waves. If f(t) represents artifact density at location t ( i n one dimension) then the Fourier form gives artifact density as a sum of waves with different frequencies. f the cosine and sine terms Filtering is based on the following property. I with n -- no are deleted, then the new function will differ from the original function solely by the absence of any wave with frequency nJr. More generally, the original function may be altered by systematically multiplying the coefficients a, and Pn by specified weights. The weights define the filter and the effect is determined by their values. Thus, with weights = 0 for large n, high frequencies wiU be Wtered out." Noise effects, for example, are typically high frequency phenomena, hence can be filtered addition to a noise filter, .Cam out by choice i f an app&pria& fil&r.-~n discussed filters that could be used for a varietv of artifact arrangements. As Carr commented (1985a: 262; 1987b:*257), the ~ouri; function is not being used as a model for the artifact arrangment, but as a redescription. This is both a strength and a weakness. The strength is that the redescription decomposes the raw data into a sum of curves which can be suitably transformed, the weakness is that the Fourier function, as do all techniques (see Read 1985) imposes an implicit model which may or may not be inherent in the way the data are structured. Here, the assumed structure is one of a periodic effect. Some effects such as noise (e.g., rodent disturbances) may be repeated across the site, hence are already "periodic." Others, such as activity areas, may be spatially unique and so are not periodic in their spatial structure. In either case, inde6nitely repeated periodicity must be imposed by analytically replicating the spatially defined study area across space. The spatial dimensions of the study area to be replicated then determine the lengths of the maximum vertical and horizontal periods (see Carr 1987: 257-258 for a more extensive discussion), and the periodicity necessary for representation via a (finite) Fourier function is now present. In this manner, even effects without inherent periodicity can be described by a Fourier function Fourier decomposition, filtering and signal enhancement techniques are means to begin dissection of data that are the q m of several processes ,into their constituent parts. The coefficients in the Fourier function, however, will only have interpretive value in addition to their descriptive in the orocesses information to the extent that there is inherent ~eriodicitv operating to spatially structure sites. Fourier &ctions pkvide "rescriptions of the outcomes of processes rather than models of the processes that generated these outcomesn(Carr 1987: 257).

~nthe conclusion to their book, Hodder and Orton (1976:244) mentioned location prediction of yet undiscovered sites as a potential application of analysis techniques, but one with uncertain relevance. Predictive modeling - something of a misnomer since all statistical models are predictive" when extrapolating to unsampled data - has since flourished, though not without considerable concern for its implications in CRM work as noted by Kohler and Parker (1986) in a review article. In its most extreme, purely inferential form, predictive modeling merely assigns a probability density function to coordinate space based on a uaining sample of known site locations with measures of geographical and environmental characteristics used as predictor variables. Modeling is n terms of standard statistical models relevant to the task: multiple done i regression analysis, linear and quadratic discriminant function analysis and logistic regression analysis (e.g, Parker 1985; Kvamme 1985; Custer et al. 1986) are primary examples. Validation need only be through efficacy: Does the constructed probability density function match empirical reality? In this sense it is primarily an application of inferential, descriptive statistics. Descriptive statistics do not require any theoretical underpinnings from the domain in question. If there is complete enumeration of a population, parameters can be measured without error and frequency distributions are exact. When enumeration is not complete, inference is made from sample statistics to population parameters. The validity of these inferences lies in the sampling methods and mathemtical relationships linking sample statistics to population pameters via estimators used to compute estimates for population parameters. Hence descriptive statistics are heavily dependent upon population specification and sampling method for empirical inferences to have the degree of precision, lack of b i a s and the like as specif~ed by statisticaltheory. As Kohler and Parker (1986) discussed, both adequate population specification and sampling methods are problematic. Of necessity, population definition for sampling purposes is coordinate space divided into spatial units, yet measures are often site specific. Spatial auto-corrdation of environmental features violates independence of measures made on sampling units and the sample population is not always the same as the target population. Non-uniform or non-random spatial arrangements increase variance estimates and small sample sizes make statistical tests less powerful. Variables selected may or may not measure the "truen dimensions which a f f d settlement location and the a-level for significance tests depends upon normality assumptions more often violated than not. Kohler and Parker examine the relative efficacy of the several statistical techniques through use of simulated data - with simulation based on

DWIGHT W. READ

METHODS IN ARCHAEOLOGICAL RESEARCH

several different rational choice models for settlement location. Their simulations were based on 150 quadrats for the training sample and 150 quadrats for the test sample. Error rates were computed for each type of arrangement and for each statistical technique. The results are less than comforting: with almost perfect data (no noise, processes operating uniformly over the region, a l l variables relevant to location included in the analysis, etc.) error rates (misclassification of a quadrat in terms of whether it should or should not have a site) for the four statistical techniques and four settlement location models ranged from a low of 8% to a high of 17%. Misclassfication of quadrats with sites classified as quadrats without sites ranged from 7% to 33% with a median of 29%. These are problems that arise when one attempts to do predictive modeling within the sampled region. The matter is likely to become far worse when a predictive model constructed for one population is applied to a new population, for then the statistical inference connection from sample to population is lost. But even within the same population, the assumption of processes acting uniformly over space and equally for all sites is highly problematic. The data are inherently heterogeneous. Hence, stratification is necessary to increase homogeneity in the sample subpopulation, but the criteria for stratification are not very well known initially - though they can be inferred through a two stage sampling process as discussed by Read (1986). Read observes that sampling can be divided into two phases: the first phase is used to obtain initial estimates that are then used to stratify the region for sampling to be done in the second phase. Improvement in predictive modeling would seem to require a different strategy than merely shifting to more complex statistical methods. Kohler and Parker considered work that has been done to develop "deductive" models for settlement location (see Kohler and Parker 1986 for references).They commented that
deductive approaches . force us to focus on the systemic decision context, thereby explicitly incorporating elumnts of anthropolcgical interest [see also Limp and Car 19851. 'Ihe use of these approaches in CRM has not o h been attempted, although there appears to be no reason why deductive models could not be employed in this context (p. 44 1). SPATIAL AUTOCORRELATION AND OUALSTATNE SPATIAL PAlTERNS '

introduced through the assumption of independence for sample values when in fact spatial patterning means that nearby values are correlated and not independent. Carr (1984: 190), for example, did not take this into account in a discussion of a technique suggested by Luton and Braun (1977) for assessing spatial arrangement based on differences, F = f , f , , in the frequency count for a grid-cell, f,, and the frequency, f,, for an adjacent grid cell. C a r r noted that the statistical test used in the procedure is based on the (purported) fact that the variance for D is the sum of the variances for f, and f,.However, if X and Y are random variables and Z = X - Y (or if Z X Y),then

$=$+a:+2cov(X,

Y),

where cov(X, Y) is the covariance between X and Y. Only when cov(X, Y) = 0 will the purported Fact be valid. But the assertion that there is spatial structure implies that covCf,, f,) # 0. Whitley and Clark pointed out that autoregression (discussed in Hodder and Orton 1976) has the potential of resolving the conflict, but caution that the method is complex and "beyond the technical capabilities of the average social scientist" (Whitley and Clark: 391). As an alternative, the authors suggested testing for the presence of spatial autocorrelation in data as a means to determine if inferences made from analyses based on correlationcoefficients are suspect. The presence of spatial autocorrelation can be tested using the I statistic:

..

Spatial autocorrelation refers to situations whdre the value of a variable for an object is correlated with values of the same variable when measwed over objects in close proximity. But, as Whitley and Clark (1985) commented, while spatial autocorrelation is basic to the idea of spatial analysis, the presence of spatial autocorrelation biases regression and correlation analysis used to detect spatial patterning @. 379). The bias is

is a weight assigned to the (i, j ) pair of grid units (e.g., if they where W-,. are contiguous, 0 otherwise), Zi = Xi $ X the variable being measured, A total number of contiguous pairs of units. The numerator consists of a covariance term based on contiguous grid units and the denominator is a variance measure, hence is similar in form to the definition of the Pearson product moment correlation coefficient. The weights Wbspecify a model for the spatial effects. Whitley and Clark applied the I statistic to dates for the Maya lowland sites used by Bove (1981) in a surface trend analysis of these data. They concluded that no spatial autocorrelation is present, hence the results do not support a previously hypothesized northwest-to-southeast pattern in monument erection As discussed by Whitley and Clark (1985) spatial autocorrelation is used as a property expressed in coordinate space; i.e., the variable X is measured over grid units themselves regularly placed in a coordinate system. However, contiguity is also a qualitative relationship. Object A may be considered contiguous to object B if A is the nearest neighbor to

( n x x WqZiq)/(2A 2:Z ),

DWIGHT W. READ

METHODS IN ARCHAEOLOGICAL RESEARCH

B. Voomps and OShea (1987) used this notion of contiguity in their application of spatial autocorrelation to a Mesolithic cemetery. With contiguity defined as a property among objects located in space, rather than a property of spatially defined units, Voomps and O'Shea were able to examine the manner in which grave goods of a given kind exhibited spatial pattern in their contiguity relationships without simultaneously being tied to a metric for the measure of contiguity. In their analysis they first used simulated data to determine that the theoretical statistical distributions computed for contiguity based on spatial proximity ( C M and Ord 1973) also hold for contiguity based on object nearest neighbors. Second, the simulations were used to determine the probability of having more than the observed number of like (alternatively, unlike) nearest neighbors up to the kth nearest neighbor, for values of k ranging from 1 to 50. That is, for each value of k, 1 < k Q 50, where 50 is simply chosen as a reasonable upper bound, counts were made for graves in the Oleneostrovski mgilnik cemetery of the number of times a given grave which contains, say, an effigy, has for its ith nearest neighbor a grave which also contains an effigy, i Q k (alternatively, the number of times the comparison grave does not contain an effigy). The authors established that different kinds of grave goods (effigies, elk, beaver and bear pendants) exhibited clustering at the level of the first 2 or 3 nearest neighbors, but differed markedly around the 39th nearest neighbors. They suggested that the local clustering may mean that "corporate units, probably extended families, were buried in close proximity to one another" @. 520). The secondary clustering suggests that other factors also affected the placement of these goods in graves. Though not mentioned by the authors, the contiguity idea that is being used is an example of the mathematical notion of a relation. Roughly, a relation, R, is a rule that pairs objects together in a specified order. More exactly, a relation R over a set S is a subset of S X S, the set of all ordered pairs of elements from S. For notation, if a is related to b through the relation R, one typically writes aRb (read 'a is R related to b'). Thus, the nearest neighbor defines a relation, call it N, via: aNb if, and only if, a is the nearest neighbor of b and a Z b. One may define a series of such relations based on nearest neighbors by noting whether a is the 1st nearest neighbor (with a Z b), the 2nd nearest neighbor, etc, of b. Thus, let N,be defined by: aN, b if, and only if, a is the ith nearest neighbor of b. Define No to be the nearest neighbor of an object, allowing for the object itself to , be a nearest neighbor. Following Cliff and Ord (1973), the term reflexive is used by Voonips and O'Shea to mean that both aN,b and bN,a (a is the nearest neighbor of b and b is also the nearest neighbor of a). The terminology is unfortunate as the term reflexive has an already established meaning in terms of mathematical relations. A relation R over a set S is reflexive if, and only

if, &a holds for all a in S .The No relation given above is reflexive, for m p l e . The propem used by the authors is that of symmetry: a relation R over a set S is symmetrical if, and only if, whenever aRb is hue then bRa is true. Qualitatively, examination of a spatial arrangement can be extended beyond these distinctions through two other basic properties for relations, w i t i v i t y and equivalence (see also Zubrow 1985 for use of fractals to qualitatively model spatial patterns). Together, these relational properties provide a qualitative basis for a classification. This may be done as follows. A relation R over a set S is transifive if, and only if, whenever aRb is m e and bRc is true then aRc is true. The nearest neighbor relation n, is not, in general, transitive. However, let each N, relation be extended to a relation N: that includes the No relation (that is, N: is defined via: a x b if, and only if, aN, b or aNpb). Then N; is reflexive since aNoa is true. If N: is also a symmetric relabon, then it is transitive: N: symmeaic implies that if both aN, b and bN,c are true then a = c, and N: reflexive implies that aN'c is true. Thus when the relation N: is symmetric it is also reflexive and transitive and hence forms what is called an equivalence relation. It may be shown that for any equivalence relation, R, defined over a set, S, there corresponds a natural partition of the set, S, into what are called equivalence c h a . Equivalence relations thus form a "natural" means to define classes, hence are an alternative to quantitative definitions of classes and their attendant problems (see Dunnell 1971). For example, in the case of a symmetrical relation N,'(that is, where the first nearest neighbor is a symmetrical relation), N,'is an equivalence relation and the equivalence classes defined by N,'will be the pairs of points which are the nearest neighbors of each other.
STATISTICAL MODELING

Quantitative intrasite spatial analysis began with a simple premise - the spatial patterning of artifacts should reflect the spatial patterning of activities - and a simple method - a Poisson test for randomness of frequencies counted over grid squares. The methods became more Poisson distributions led to nearest neighbor statistics and complex nearest neighbor statistics, it is now argued, should be supplemented by spectral analysis (Graham 1980) and Fourier filtering (Carr 1985a; 1987b) - yet it is not evident that increase in methodological complexity alone has necessarily led to commensurately greater understanding the relationship between spatial distributions and structuring processes. Rather, one must have some understanding of that relationship to justify the choice of methods (Carr 1987b: 285). The reasons for introducing more complex methods are not being

DWIGHT W. READ

METHODS IN ARCHAEOLOGICAL RESEARCH


'

questioned, as the rationale for their introduction is valid. Rather, d i sion in t h i s section will focus on the implications for statistical modeling resulting from the growing awareness that linking data structure with structuringprocesses is a conceptual, not just a methodological, problem In part, the conceptual problem relates to how the archaeological oroblem is conceived; and in part it relates to how statistical modeling can provide a means to extend &chaeological reasoning. The latter dehnds upon finding' congruence between the archaeological and statistical dbmains at Sheir deepest levels, not at the more surface level of data, methods, and hypotheses. To become part of archaeological reasoning, the fundamental concepts upon which statistical methods are built must be shown to be structurally in agreement with conceptualizations that underlie this reasoning. Read (1985, 1987b) discussed two terms basic to application of statistical methods: population and variable. These are constrasted by him with two terms basic to the conceptualfmn?ework which gives statistical methods their meaning: probability space (a set w of outcomes along with a set P of probabilities for the outcomes) and random variable (a mapping from the set o of outcomes to the real numbers, R). The former two terms, he argued, are given equivalence to the latter two terms in normal applications of statistical methods through construction of an m c i a l "experimentw: pick at random an object from the population that has been h i s creates an "experimentn in which each defined for research purposes. T object in the population now becomes a distinct outcome for the experiment. Further, 1/N, where N is the population size, becomes the probability of an outcome. The variable used to assign a value to an object (- outcome) is now a mapping from the outcomes (= objects) to the real numbers, hence satisfies the detinition of a random variable. Read points out this construct c o a c t s with using statistical reasoning as part of archaeological reasoning due to its use of an artificial experiment which has no meaning for the archaeological domain. The statistical arguments are about properties of the experiment for which the probability space is a representation The experiment defined above has no meaning beyond its role as a way to both equate the a space of outcomes and the researcher's researcher's population w i ~ measurement made on objects with a random variable defined over the probability space. For this reason, the statistical results are equally information about circumscribed and so thev will onlv orovide descri~tive . an experiment unrelated-to the s c e s s e s giv@ ;he data their inherent structure. In plaoe of the artificial experiment, Read (1987b) suggested that one may consider the process through which artifacts are produced, for example, to be a natural experiment. Designation of an outcome then determines the modeling process through the use of the terms, probability
t

space and random variable. Thus, if the object being produced is to be the outcome, then one is modeling at the level of the production process. If the emic category (whether this is knowable or not) to which the object belongs is the outcome, then one is modeling at the level of the conceptual system which gives objects their cultural meaning. These two levels for modeling, Read observed, place different demands on what constitutes a random variable. In the former instance, a measure such as length can be construed as a random variable since it assigns to an object (the outcome of the production process) a real number, namely the length of the artifact. The same measure is not a random variable if the outcome is the category to which the object belongs, for here a random variable must assign a value to the category (= outcome); hence the same value would be assigned to each instance of the category. Read (1987b) notes that in this case the following model would apply: where L is the value of the measure for the object (e.g., the length of the object), p is the value assigned to the category to which the object belongs, e is an error tern with E (e) 0. The term p = E (L), under the assumption that length is an emically relevant dimension for the category, can be given interpretation, Read (1987b) argued, as a norm for the category. It would not be a norm in the sense of a statistical average, but in the sense of a conceptual norm; i.e., as an emic property associated with the category. T h i s implies that it is possible to include within statistical modeling qualitative concepts essential for understanding not just the material aspects of artifactual materials and their spatialltemporal distribution, but their underlying ideational and cultural aspects as well. Read then showed via an analysis of projectile points how one may interrelate the groupings found for the objects in a hierarchical structure. Read distinguished four levels in the hierarchy: phenomena, norms, emic models and (emic) concepts @. 177). What is shared as one moves up in the hierarchy is no longer physical attributes, but aspects that derive from the conceptual categorization represented by that level. Reynolds (1986) independently used a similar hierarchical approach in a study of the spatial distribution of materials on the Guila Naquitz living floors. Reynolds argued that a conceptually based characterization of these living floors requires that statistical methods be related to the underlying cognitive/conceptual level, not just to the description of pattern in data. Reynolds observed that some analytical techniques may "produce archaeologically inferred activity areas that bear little resemblance to the cognized activity areas that preceded them and to which they owe part of their existencen and so there is a need to "incorporate formally both behavioral and conceptual postulates into the procedures used to analyze

DWIGHT W. R E A D

METHODS IN ARCHAEOLOGICAL RESEARCH

occupation loorsn @. 386). The goal is to understand "the conceptual bases for statisticallydemonstrable patterns of debris in the cave" @. 396). The analysis is based upon postulating a possible hierarchy of conceptual categories that begins with the kind of object (e.g., a cactus species) and then forms higher level groups which transcend the properties of the lower levels. Thus, at the level above plant species, certain plants may be grouped together on the basis of each b e i i seasonal, so one has the class "seasonal plantsn;at the next level categories such as "seasonal plants" and "cactus plantsn may be grouped together under "plant processing areas," and so on. The higher level groupings are not based on material attributes of objects, but reflect properties they purportedly share as part of a conceptual system @. 386). Analytical methods - measurement of association, Q and R mode analyses using multi-dimensional scaling, etc. - are then utilized in accordance with this hierarchical arrangement. The analysis led to characterization of the living floor in terms of a postulate for how space was conceptualized, not just a description of the spatial pattern of objects. Reynolds concluded that "there seem to have been certain constants in the way work space in the cave was conceptualized, and these constants canied over from one group of occupants to the next" @. 408). Here, statistical modeling has clearly become a means to extend archaeological reasoning. CONCLUSION The "promise" of quantitative and statistical methods to provide significant, new insights has been compromised by lack of its prerequisite: well-shuctured, well-defined arguments connecting data structure with statistical method and archaeological reasoning. This lack is stated in different ways by different authors: discordance between implicit statistical model and archaeological model (Read 1985, 1987b), discordance between data structure and analytical method (Carr 1985b), 1987a; Aldenderfer 1987b) global versus local patterning in spatial distributions (Hietala 1984; Whallon 1984; Can: 1985a, 1987a; Simek et at. 1985), and so o n But whatever the termin'ology used, common to all of these comments is growing awareness that archaeological data have complex structure that is not resolved merely through application of more complex I statistical methods. Indeed, there may not be any method in the statistical sense that can resolve the matter. Data structured by local relationships that are neither globally expressed nor based on locally relevant sets of variables require dissection according to criteria that shift as analysis proceeds (d. Whallon 1972; Read 1974,1985, 1987b) and cannot be specified a priori - else they would have been global properties.
I

Part of what is needed are means to explore data, but not just in the sense of techniques as provided through Exploratory Data Analysis. Rather, what is needed are techniques that allow for interaction between visual representation and quantitative methods, much as Bietti et al. (1985) have discussed. Means are needed to erase the discrepancy noted by S i e k and Larick (1983) for spatial analysis that "The two most common strategies employed are visual inspections of distributions by informed human analysts without a formal guide or technique, and quantitative approaches based on summary statistics such as nearest neighbor" (163, emphasis added). Doran (1987: 85) saw a similar discrepancy when he referred to statistics as not involving the knowledge domain that is explicitly embedded into expert systems and suggested the latter as a way to incorporate the knowledge of archaeologists into their analyses. Though the relationship between statistical analysis and the knowledge of the archaeologist may be viewed in different terms by different authors, common here is the perception that what is needed are means to analytically draw upon the insights that we have about the archaeological domain and incorporate them into methods used to analyze data struc tures. But intelligent insight has no particular method for its realization, no preordained form for its expression. Statistical analyses and quantitative methods should be assessed througb the degree to which they help achieve h i s criterion, quantitarealization and expression of intelligent insight. By t tive methods have had both success and failure; success when they become part of archaeological reasoning and failure when form takes precedence over substance. REFERENCES CITED
Aldenderfer, M. S., ed.1987 Quantitative Research in Archeology.Newbury Park: Sage. Aldendcrfcr, M. S. 1987a Preface. In Quantitative Research in Archaeology. M. S. Aldmderfer, ed. Pp. 7-8 Newbury Park: Sage. Aldenderfer, M. S. 1987b On the Struchlre of Archaeological Data. In Quantitative Rscarch in Arcbology. M. S. Aldenderfer, ed. Pp. 89-113. Newbury Park: Sage Publicatiolls. Alederfer, M.S. 1987c Assessing the Impact of Quantitative Thinkhg on Archaeological Research: Historical and Evolutionary JJI&&~.s. h Quantitative Research in Anhaeology. M.S. Aldenderfer, ed. 4 . 9 - 2 9 . Newbury Park:Sage Publications. Aldenderfer, M. S. and R K. Blash6eld 1984 Cluster Analysis. Beverly Ws: Sage Publicationr Aldenderfer, M. S , ed. 1987 Quantitative b h in Archaeology. Newbury Park: S e e Publications. Ammennan, A, C. Matessi, and L . Cavalli-Sforza 1978 Some New Approaches to the r a d e in the Medikganem and Adjacent Ares. In The Spatial Study of the Obsidian T Orpnization of Culture. I. Hodder, ed. 4.179-96. London: Duckworth. A n & I. 0 . 1976 Stone CXrcles: McgaWic Mathematies or Neolithic Noasease. Mathematical Gazate 60: 189-93.

DWIGHT W. READ Angell, I. 0. 1978a Megalithic Science: Ancient or Modern? In Computer Applications in Archaeology 1978. S. Laflin, ed. Pp. 5-12. B i a m : University of Birmingham. Angd, I. 0. 1978b On Fitting Certain Closed Convex Curves to Archaeological Point Data Using an Interactive Graphics TerminaL Jwmal of Archaeological Science 5: 309-13. A n g a I. 0.1979 Arguments Against t h e Existence of the 'Megalithic Yard'. In Computer Applications in Archaeology 1979. S. Latlin,ed. Pp. 13-19. B i i g h a m University of Bigham. Benfer, R A. 1967 A Design for the SSNy of Archawlogical Characteristics. American Anthropologist 69: 719-30. Berry, K J, K. L . Kvamme, and P. W. Midke 1980 A Pennutation Technique for the Spatial Analysis of the Distribution of Artifacts into Classes. American Antiquity 45: 55-9. W r y , K J., P. W. Mielke, and K. L Kvamme 1984 E m e n t Permutation Procedures for Analysis of M c t Distributions. In Intrasite Spatial Analysis in Archaeology. H. Hietala, ed. 4.54--74. Cambridge: Cambridge University Press. Bertelsen, R 1985 Artifact Panem and Stratification Units. American Archawlogy 5: 16--20. Bieber, A. M, D. W. Brooks, G. Harbottle, and E. V. Saym 1976 Application of Multivariate Techniques to Analytical Data on Agean Ceramics. Archawmetry 18: 59-74. Bietti, A, A. Burani, and L. Zanello 1985 Interwtive Panern Recognition m Prehistoric Archaeology: Some Applications. In To Panem the Past A. Voonips and S. Loving, eds. 4.205-28. Amsterdam: PACT. Binford, L. 1964 A Consideration of Archaeological Research Design. American Antiquity 29: 4 2 5 4 1 . Binford, L. 1975 Sampling, Judgment, and the Archaeological Record In Sampling in Archaeology.J. W. Muder, ed. 4.251-57. Tucson: University of Arizona Press. Biiord, L. 1978 Dimensional Analysis of Behavior and Site Smcture: Learning from an Eskimo Hunting Stand. American Antiquity 43: 330-61. Binford, L. 1981Bones: Ancient Men and Modern MyThs. New Yo& Academic Press. Biiord, L. and J. B. Bertram 1977 Bone Frequencies - and AtUitional Processes, In For 'Iheory Building in Archaeology. L Binford, ed. Pp. 77-153. New York: Academic Press. h e Biiord, S. and L. Binfotd 1966 A Preliminary analysis of Functional Variability in t Mousterian of LcMLlois Facies. American Anthropologist 68: 238-95. Bordaz, V. and J. Bordaz 1970 A Computer-assisted Pattern Recognition Method of Classhication and Seriation Applied to Archaeological Material. In Arch&@ el Calculatwrs. J. C. Gardin, ed. 4.229--44. Paris: mans du CNRS. Bodes, F. 1961 Typologie du Pal&lirhique Ancien et Moyen. Bordeaux: Imprberies Delmas. BoriUo, M. F. 1977 Raisonner, Calculer. 4 Raisomement et Mtthodes Mathbtiques en Archcologies. M. F. Borillo, ed. 4.1-31. Paris: Edition du CNRS. Borillo, M , F. Femanda de la Vcga, and A. Cfuenochc, eds. 1977 Raisomunent et MUhodes Mathhnatiques en Archhlogie. Pairs: Edition du CNRS. Bove, F. 1981 Trend Surface Analysis and the Lowland Q s i c Maya Collapse. American , Antiquity 46: 93-112. Bradley, R and C. Small 1985 Looking for Circular Structures in Post H d e Distributioas: Quantitative Analysis of Two Settlements from Bronze Age England. Journal of ArchaeologicalScience 12: 285-97. Braun, D. P. 1985 Absolute Seriation: A Tme-series Approach. In For Concordance in Archaeological Analysis. C. C a r r ,ed. Pp. 509-39. Kansas City Westport Publisher, I n c

METHODS IN ARCHAEOLOGICAL RESEARCH ~rown,J. A 1987 Quantitative Burial Analyses as Interassemblage Comparison. In Quantitative Research in Archaeology. M. S. Aldenderfer, ed. Pp. 294-308. Newbury park: Sage Publications. Butler, B. H., E. Tchernov, H. J. Hietala, and S. Davis 1977 Faunal Exploitation During the Late Epipaleolithic in the Har Harif, Iq Prehistory and Paleoenvironments in the Cenual Negev, Israel, Vol. LI, The AvdaVAqev Area, pan 2, The Har Harif. A. E. Marks, ed. Pp. 3 2 7 4 5 . Dallas. Southern Methodist University. Carr, C. 1984 The Nature of Organization of Intrasite Archaeological Records and Spatial Analytic Approaches to Their Investigahon. Advanw in Archaeological Method and Thwry 7: 103-222. Carr, C. 1985a Screening Inaasite Artifact Distributions with Fourier and Filtering Methods. In To Panern the Past. A. Voomps and S. Loving, eds. Pp. 249-86. Amsterdam:PACT. Carr, C. 1985b Getting lnto Data: Philosophy and Tactics for the Analysis of Complex Data Smctures. In For Concordance in Archaeological Analysis: Bridgiog Data Suucture, Quantitative Technique, and Theory. C. Cam,ed. 4. 18-44. Kansas City: westport Press. Carr, C. 1984c Alternative Models, Alternative Techniques: Variabk Approaches to I n m i t e Spatial Analysis. In For Concordance in Archaeological Analysis. C. Can, ed. Pp. 3 0 2 4 7 3 . Kansas City: Westport Publishing,Inc. Carr, C. 1987a Removing Discordance from Quantitative Analysis. In Quantitative Research in Archawlogy M. S. Aldenderfer, ed. 4. 185-243. Newbuty Park: Sage Publications. Carr. C. 1987b Dissecting Intrasite Artifact Palimpsests Using Fourier Methods. In Method and Theory for Aetivity Area Researeh: An Ethnoarchaeological Approach. S. Kent, 4.Pp. 236-91. New York: Columb'i University Press. Casteel R. W. 1977 CharacIerization of Faunal Assemblages and the Minimum Number of Individuals Determined from Paired Elements: Continuhg Problems in Archaeology. Journal of Archaeological Science 4: 125-34. cavanagh, W. G and R R Laxton 1981 ?he Structural Mechanics of the Mycenaean Tholos Tomb. Annual of the British School at Athens 76: 109--40. Cavanagh, W. G. and R R Laxton 1982 Corbelling in the Late M i n o a n 'Iholos Tombs. Annual of the British School at Athens 77: 65-77. Camagh, W. G. and R R Laxton 1985 Corbelled Vaulting in M y m e a n Tholos Tombs and Sardinian Nuragbi. In Papers in Italian Archaeology IV-3. C. Malone and S. Stoddart, eds. Pp. 413-33. Oxford: British Archaeological Reports. Cavanagh, W. G., R R Laxton, and C. D. Litton 1985 An Application of CbW-point Analysis to the Shape of Prcchistoric Corbelled Domes; L The Maximum Wcclihood Method. In To Pattern the Past. A. Vwrrips and S. H. Lo-, eds. Pp. 191-200. Amsterdam. PACT. Champion, T. 1978 Strategies for Sampling a Saxon Settlement: A Re?ms@ve View of chelton, In Sampling in Contemporary British Archaeology. J. F. Cherry, C. Gamble, and S. Shennan, eds. Pp. 207-26. Oxford: British Archaeological Reports. Chaplin, R E. 1971 The Study of Animal Bones from Archaeological Sites. New Yo&: Seminar Press. Cherry, J. F. and S. Shmnan 1978 Sampling Cultural Systems: Some Pem@ves on the Application of Probabilistic Regional Survey in Britain. In Sampling in Contemporary British Atchaeology. J. F. Cherry, C. Gamble, and S. Shennao, eds. pp. 1 7 4 8 . Oxford: British ArchaeologicalReports. Cherry, J. F., C. Gamble, and S. Shennan 1978 General IntroduCti~1: Attitudes to Sampling in British Archaeology. In Sampling in Contemporary British Archaedogy. J. F. Cherry, C. Gamble, and S. Shennan, eds. Pp. 1-10. Oxford: British A~~hsoDIogical Repom.

DWIGHT W. READ Chippindale, C. 1985 A Microcomputer Simulation of Variable forms in Prehistoric RockArt In To Pattern the Past. A. Voomps and S. Loving, e d ~Pp. . 303-22. Amsterdam: PACT. c!hrhnsoo, A 1987 Prehistoric Stone Techndogy on Nonhern Black M e s a , Ariwna. Carbondale:Southern Illinois University. Christenson, A and D. W. Read 1977 Numerical Taxonomy, R-Mode Factor Analysis and Archaeological Classification.AmericanAntiqwty 42: 163-79. Christie, 0. H. J. and J. A. Brema 1979 Multivariate Classification of Roman Glasses Found in Norway. Archaeometty 21: 2 3 3 4 1 . Clark, G. A. 1976 More on Contiagmcy Table Analysis, Decision Making criteria, and the Use of Log Linear Models. AmericanAntiquity 41: 259-73. Clark, G. A. 1982 Quantifying Archaeological Rtsearch. Advances in Archaeological Method and Theory 5: 217-73. Clark, G. A. 1984 'On the Analysis of Multidimensional Contingency Tables Using LogL i n e a r Models. In Computer Applications in Archaeology 1984. J. Wilcock and S. LalJin, eds. 4.47-58. B i i University of Birmingham. Clark, G. A. 1987 P a r a d i i and Paradoxes in Contemporary Archaeology. In Quantitative Research in Archaeology Progress and Prospects. M. S. Aldenderfer, ed. Pp. 30--60. Newbury Pa& Sage Publications. Clarke, D. L. 1968 Analytical Archaeology. Londoo:Methuen & Co. Ltd. Cliff, A. D, and J. K Ord 1973 Spatisl Autocorrelation. London: Methuen. Cogbill, S. 1980 Computer Post-Hole Analysis with Reference to the British Bronze Age. Computer Applications in Archaeology 1980. J. Wilcock and S. L a b , eds. Pp. 35--8. B ' i University of B i Coombs, G. B. 1978 The Archaeology of the Western Mojave. BLM: Cultural Resources Publicaiion. Cowgill, G. 1970 Some Sampling and Reliability Problems in Archaeology. In ArchCologie a Calculateurs. J. C Gardin, ed. Pp. 161-172. Pairs: Editions du CNRS. Cowgill, G. 1972 Models, Methods and Techniques for Seriation. In Models in Archaeology. D. L . CLarke, ed Pp. 3 8 1 4 2 4 . London: Methuen & Co. Ltd. Cowgill, G. 1982 CLusten: of Objects and Associations B e Variables: Two Approaches to Archaeological Classification. In Essays on Archaeological Typology. R W. Whallon and J. A. Brown, eds. Pp. 30-55. Evanston: Center for Amuican Archaeology Press. Cuter, J. F., T .Evcleigh, V. Klemas, and I. W& 1986 Application of Landsat Data and Synoptic Remote S e a s 4 to Predictive Models for Prehistoric Archaeological Sites: An Example from the Delaware Coastal Plain. American Antiquity 51: 572-88. Dacey, M. 1973 Statistical Tests of Spatial Association in the Locations of Tool Types. American Antiquity 38: 320-28. De Barros, P. 1982 The Effects of Variabk Site Occupation Span on the Results of Frequency Seriatim. American Antiquity 47: 291-315. de la Vega, W. F. 1970 Quelques Propriet6 des Hiirarchies de Clasikications. In ArchCologie et Calculateurs. J. C. Gardii ed. Pp. 329--40. Paris: Editions du CNRS. de la Vcga, W. F. 1977 Dew Algorithmes de Seriation. In Raisomement et MUhodes Mathhtiques en Arehiologie. M. Borillo, W. F, de la Vega and A. Guenoehe, eds. Pp. 147-7O.Paris: Editions du CNRS. f pbble, H. L . and P. G. Chase 1981 A New Mdhod for Describing and AnalArtifact Sttape. American Antiquity 46: 178-87. Djindjian, F. 1985 Seriation and Topmeriation by Correspondence Analysis. In To Pattern the Past. A Voomps and S. Loving, eds. Pp. 119-36. Amsterdam: PACT. Doran, J. 1970 Archaeological Reasoning and Machine Reasonin& In ArchCologie el Calculateurs. J. C. Gardin, ed. 4 . 5 7 4 6 . Pairs: Editions du CNRS. Doran, J. 1977 Automatic Generation and Evaluation of Explanatory Hypotheses. In

METHODS IN ARCHAEOLOGICAL RESEARCH Raisomement a Mklhodes Mathhatiqua en Archiologie. M. Borillo, W. F. de la Vega, and A. Guenoche, eds. 4 . 1 7 2 - 8 1. Pairs: Editions du CNRS. Doran, J. 1987 Anthropological Archaeology, Computational Modeling, and Expert Systems In Quantitative Research in Archaeology: Progress and Prospects. M. S. Aldenderfu; cd. Pp. 73-88. Newbury Park: Sage Publications. Doran, J. and F. R Hodson 1975 Mathematics and Computers in Archaeology. Cambridge: Harvard UniversityPress. Dunnell, R. C. 1971 Systematicsin Prehistory. New York: The Free Press. Uisseff, M. V. 1970 DonnCes de Classement Foumies par les Scalogrammes PrivilCgiC. In Archblonies et Calculateurs. J. C. Gardii, ed. 4. 177-83. Paris: Editions du CNRS. Fasham, P. and M. Monk 1978 Sampli for Plant Remains from Iron Age Pits: Some Results and Implittions. In %np@ in Contemporary British Archaeology. J. F. Cheny, C. Gamble, and S. Sheman, eds. Pp. 363-72. Oxford: British Archaeological Repom. FieUer, N. R J. and A. Turner 1982 Number Estimation in Vertebrate Samples.Journal of Archaeological Science 9: 49-62. Fletcher, M. and G. L x k 1984 Post Built Structures at Danebury Hillfort: An Analytical Search Method with Statistical Discussion. Oxford Journal of Archaeology 3: 175-96. Fletcher, M. and G. R Lock 1981 Computerised Pattern Perception Within Post hole Distributions. Scienceand Archaeology 23: 15-20. Fletcher, M, and G. R Lock 1984 A Mathematical Model for the Distribution of Rectangles Wltllin a Random Distribution of Post Holes. Science and Archaeology 26: 28-37. Gardii, J. C., ed. 1970 ArchCologie et Calculateurs: Probltmes Semiologiques el Mathhatiques. Pairs: Editions du CNRS. Gelfand, A. E. 1971 Rapid Suiatioa Methods with Amhamlogical Applications. In Mathematics in the Archaedogid and Historical Sciences. F. R Hodson, D. G. Kendall, and P. Tautu, eds. Pp. 186-201. Edinburgh: Edinburgh University Press. Gero, J, and J. Mazzullo 1984 Analysis of Artifact Shape Using Fourier Series in Closed F o m Journal of Field Archaeology 11: 3 15-22. Geyh, M. A. and P. de Maret 1982 Histogram Evaluation of C14 Dates Applied to the First Complete Iron Age Sequence from West Central Africa. Archaeomcf~y24: 15863. Gibbon, G. 1984 Anthropological Archaedogy. New York: Cdumbia University Press. Gladfelter, B. G. and C. liedemam 1985 The Contiguity-AnomalyTechnique for Analysis of Spatial Variation. In For Concordance in Archaeological Analysis. C. Carr, ed. 4 . 474-501. Kaosas C i t y : Westport Publishin&Inc. Graham, I. 1980. Spectral Analysis and Distance Methods in the Study of Archaeological Distributions. Journal of Archaeological Scimce 7: 105-29. Graham, I., P. Woway, and I. ScoUar 1976 Model Studies in Computer Seriatim Journal of Archaeological Science 3: 1-30. Grayson, D. 1984 QuantitativeZooarchaeology. Orlando: Academic Press, Inc. Grayson, D. K. 1978 Minimum Numbers and Sample S i z e in Vertebrate Faunal Analysis. American Antiquity 43: 53-44. Grayson, D. K. 1979 On the Quantification of Vertebrate Archaeohunas. Advances in Archaeological Method and Theory 2: 199--237. Grayson, D. K. 1981 The Effects of Sample S i on Some Dcrived Measures in Vertebrate Faunal Analysis. Joumal of Archaeological Science 8: 77-88. Groube, L. 1978 Concluding Address. In Sampling in Contemporary British Archaeology. J. F. Cherty, C. Gamble, and S. Shenuan, eds. Pp. 403--08. Oxford: British Archacological Reports. Guenoehe, A. and P. Ihm 1977 Analyse en Composante Rincipale a Analyse Discrim-

72

DWIGHT W. READ

METHODS IN ARCHAEOLOGICAL RESEARCH

73

inante dans le Cas des Don& Incompletes. In Raisomement et MCthodes Math& matiques en Archblogie. M. Borillo, W. F. de la Vega, and A. Guenoche, eds. Pp. 131-38. Pairs: Editions du CNRS. Haselgrove, C. 1978 SpaJial Pattern and Settlement Archaeology: Some Reflections on n Contemporary British Archaeology. J. F. Cherry, C. Sampling Design. In Sampling i Gamble,and S. Shehnan, eds. Pp. 159-76. Oxford: British Archaeological Reports. Hassan, F. A. 1974 The Archaeology of the Dishan Plain, Figypt A study of a Late Palaedithic Settlement. Geological Survey of Egypt. Hayes, A, D. Bmgge, and W. Judge 1981 Traosect Sampling in Cham Canyon - Evaluation of a Survey Technique. In Archaeological Surveys of Cham Canyon. Publications in Archaeology 18.4: Chaco Canyon Studies. A. Hayes, D. Drugge and W. Judge, eds. 4.107-37. Washington, D.C.: National Park Service. Heizer, R F. and S. F. Cook, eds. 1960 The Application of Quantitative Method in Archaeology. New York: Werner Grur Foundation for AnthropologicalResearch Hietala, H. 1984 Variations on a Categorical Data Theme: Local and Global Considerations with Near-Eastern Paleolithic Applications. In Intrasite Spatial Analysis in Archaeology. H. Hietala, ed. 4.44-53. Cambridge:Cambridge University Press. Hietala, H. J. 1983 Boker Tachtit: Spatial Distributions. In Prehistory an Paleoenvironments in the.Ceaual Ne-gev, Israe4, vol. m,The AvdatIAqev Area, Part 3. A. E. Marks, e d . 4 . 1 9 1 - 2 1 6 . D W Southcm Methodist University. Hietala, H. J. and A. E. Qose 1979 Testing Hypotheses of Independence on Symmehical C I S . Journal of ArchaeologicalScience 6: 85-92. Hietala, H. J. and P. Larson 1980 Intrasite and Intersite Spatial Analyses at Bir Tarfawi. In 'Ihe Prehistory of the Easnm Shara: Paleoenvirouments and Human Exploitatioa F. Wendorf and R. Schild, eds. Pp. 379-88. New York: Academic Press. Hietala, H. J. and D. S. Stevens 1977 Spatial Analysis Multiple Procedures in Pattern Recognitionstudies. American Antiquity 42: 539-59. Hinkley, D. V. 1971. 1nfere.u~~ in Two-Phase Regessioa Journal of the American StatisticalAssociation 66: 7 3 6 4 3 . Hodder, 1. 1974 Regression Analysis of Some Trade and Marketing Patterns. World Archaeology 6: 172-89. Hoddcr, I. 1978a Some Effects of Distance on Pattems of Human Interaction. In The SpatialOrganisation of Culture. I. Hodder, ed. 4.155-78. London: Duckworth. Hodder, I. 1978b Social Organization and Human Interaction: The Development of Some Tentative Hypotheses in Terms of Material Culture. In The Spatial Organisation of Culture. L Hodder, ed. Pp. 199-270. London: Duckworth. Hodder. I 1978c Simple Correlations W e e n Material Culture and Society: A Review. In ?he Spatial Organisation of Culture. I. Hodder, ed. 4 . 3-24. London: Duckworth. Hodder, 1. 1978d The Spatial Structure of Material 'Cultures': A Review of Some of the Evidence. In I h e Spatial Organisation of Culture. I. Hodder, ed. Pp. 93-1 12. London: Duckworth. Hodder, 1. 1985 P o s t p r o c ~ l a lArchaeology. Advances in Archaeological Method and Theory 8: 1-26. Hodder, L and E. OkeU 1978 An Index for &ing the Assuciition Between Distributions of Points in Archaeology. In Simulation Studies in Archaeology. L Hodder, ed. pp. f 97-107. Cambridge: Cambridge University Press. Hodder, I. and C. Onon 1976 S p a Analysis in Archaeology. Cambridge: Cambridge University Press. Hodder, I. and R Reece 1977 A Model for the Distribution of Coins in the Western Roman Empire. Journal of ArchaeologicalScience 4: 1-1 8. Hodson, F. R 1971 Numerical Typdogy and Prehiaoric Archaeology. In Mathematics in the Archaeological and Historical Sciences. F. R Hodson, D. G. Kendall, and P. Tautu, eds. Pp. 30-45. Edinburgh: Edinburgh University Press.

Hodson, F. R., D. G. Kendall, and P. Tautu, eds. 1971 Mathematics in the Archaeological and Historical Sciences. Edinburgh: Edinburgh University Press. Hole, B. L. 1980 Sampling in Archaeology A Critique. Annual Review of Anthropology 9: 217-34. Holm-Olsen, I. M. 1985 Farms Mounds and Land Registers in Helgoy, North Norway: An Investigation of Trends in Site Location by Correspondence Analysis. American A r M o g y 5: 27-33. Horton, D. R. 1984 Minimum Numbers: A Consideration. Journal of Archaeological Science 11: 255-7 1. Ihm, P. 1977a Introduction i la Statistique Uni et Multi-Dimensionndle.In Raiso~ement et MCthodes Mathhatiques en Archwlogie. M. Borillo, W. F. ed la Vega, and A. Guenoche, eds. Pp. 34-72. Pairs: Editions du CNRS. Ilun, P. 1977b Deux Transformations Pour le Traitement Mathbtique des Problhes de Senation. In Msomement et M6thodes Mathcmatiques en ArchCologie. M. Borillo, W. F. de la Vega, and A. Guenoche, eds. Pp. 139-44. Paris: Editions du CNRS. Ihm, P. and H. van Gmenewoud 1984 CorrespondenceAnalysis and Gaussian Ordination. Compstat Lectures 3: 5-60. Impey. 0. R and M. PoUard 1985 A Multivariate Mettical Study of Ceramics Made by Three Potters. Oxford Journal of Archaeology 4: 157-164. Inostranaev, k A. 1882 Doistoricheskii Chelovdr K a m e ~ a g 0Veka Poberezh' is LadozhskogoOzera. St. Petersburg: Stasiulevich. Jardine, N. and R. S~bson 1971 MathematicalTaxonomy. New York:John Wdey & Sons. Johnson, 1. 1984 Cell Frequency Recording and Analysis of Met Distributions. In Inrrasite Spatial Analysis in Archaeology. H. Hietala, ed. Pp. 75-96. Cambridge: Cambridge University P r e s s . Jones, M. 1978 Sampling in a Rescue Contwn: A Case Study in Oxfordshire. In Sampling in Contempmy British Archaeology. J. F. Cherry, C. Gamble, and S. Shaman, eds. 4 . 191-206. Oxford: British ArchaeologicalReports. Kendall, D. G. 1963 A Staristical Approach to Flinders Pettie's Sequence Dating. Bulletin International StatisticalIostitute. 40: 565-70. Kendall, D. G. 1971 Senation from Abundance Matrices. In M a h a t i c s in the Archaeological and Historical Saences. F. R Hodson, D. G. Kendall, and P. Tautu, eds. Pp. 21 5-52. Edinburgh: Edinburgh UniversityPress. Kimball, L. R 1987 A Consideration of the Role of Quantitative Archaeology in Theory Coasnuctioa In Quantitative R c h in Aarchaeology. M. S. Aldendetfer, ed. Pp. 114--25. Newbury Park: Sage Publications. Kintigh, K. W. 1984 Measuring Archaeological Diversity by Comparison with Simulated Assemblages. Ame~ican Antiquity 49: 44-44. Kohler, T. A. and S. C. Parker 1986 Predictive Models for Archaeological Resource Location In Advances in Archacdogical Method and Theory. M. Schiffer, ed. 4 . 397-452. New Yo& Academic Press. Krantz, G. S. 1968 A New Method of Counting Mammal Bones. American Journal of Archaeology 72: 286-88. Krause, R A. 1984 Modeling the Making of Pots: An EthnoarchawlogicalApproach. In The Many Dimensions of Pottery. S. van der Ltcuw and A. Pritchard, eds. Pp. 615706. Amsterdam:Univeniteit van Amsterdam. Kuhl, F. P, and C. R Giardina 1982 Elliptic Fourier Features of a Closed Contour. Computer Graphics and Image Rocessiag 18: 236-58. Kvamme, K. 1985 Detcmhbg Empirical Relationships Betwen the Natural Environment and M t o r i c Site Locations: A HuntaGathertr Example. In For Concordance in Archaeologid Analysis: Bridging Data Struchlre, Quantitative T&que, and Theory. C. Cam, ed. 4.208-38. Kansas City: Westport Prrss. Landau, J. and W. F. de la Vega 1971 A New Serialion Algorithm Applied to European

DWIGHT W. READ Protohistoric Anthropomorphic Statuary. In Mathematics in the Archaeological and Historical Sciences. F. R Hodson, D. G. KendaU, and P. Tautu, eds. Pp. 255-62. Edinburgh: Edinburgh University Press. Laporte, G. 1976 A Comparison of Two Norms in Archaedogical Striation. Journal of Archaeological Sciences 3: 249-55. Laxton, R R. 1976 A Measure of Pre-q-ness with Applications to Archaeology. Journal of Archaeological Science 3: 43-54. LeBlanc, S. 1975 Micm-Seriation: A Mahod for F i e Chronologic Differentiation. American Antiquity 40: 22-38. Leese, M. N. 1985 nx. In To Pattern the Pasl A Voomps and S. Loving,Pp. 000--00. Amsterdam: PACT. Leese. M. N. and S. P. Needham 1986 Frequency Table Analysis: Examples From Early Bronze Age Axe Decoration,Journal of Archaeological Science 13: 1-1 2. Lennan, C. 1977 M&we de Proximit6 Entre Structures Algbriques de M h e Type sur un Ensemble Fini: Application la C l d c a t i o n Automatique. In Raisonnement et MCthodes MathCmatiques en Archwlogie. M. Bodlo, W. F. de la Vega, and A. Guenoe,eds. Pp. 100-30. Pairs: Editions du CNRS. Lerman, M. L C. 1970 H-clsssificabilid In W o g i e et Calculateurs. J. C. Gardin, ed. 4.319-25. Paris, Editions du CNRS. Lie, R W. 1980 Minimum Number of Individuals from Osteological Samples. Norwegian Archawlogical Review 13: 24-30. Lightfoot, K . G. 1986 Regional Surveys in the Eastern Unit4 States: The Saengms and Weahessm of Implementing Subsurface Testing Programs American Antiquity 51: 484-504. Limp, F. and C. C a n 1985 The Analysis of Decision Making: Alternative Applicatioils in Archaeology. In For Concordance in Archaeological Analysis: Bridging Data Smctun, Quantitative Technique, and Theory. C. C a r r ,ed. 4.128-72. Kansas City: Westport Press. Lingoes, M. J. C. 1970 A General Nonparamebic Model for Representing Objects and Attributes in a Joint Mehic Space. In Archblogie et Calculateurs.J. C. Gardin, ed. Pp. 277-96. Paris: Editions du CNRS. Little, B. 1985 A Comparative Analysis of Spatial Pattern. Amxican Archaeology 5: 34-40, Linon, C. D. and J. Rcstorick 983 Computer Analysis of Post Hole Distributions. ' i Computer Applications in ArclLzedogy 1983. S. La&, ed. Pp. 85-92. B University of B ' i . Luton, R M. and D. P. Bram 1977 A Method for Testing the Significance of Aggregation and Association in Archaeological Grid Cell Counts. Unpublished Paper Prescated at the Annual Meetingof the Society for American Archaeology. Main, P. L . 1978 The Storage, Retrieval, and C l a d b t i o n of Artefact Shapes. In Computer Applications in Ar&aeology 1978. S. L&n, ed. Pp. 3 9 4 8 . Bimhgham: University of Bhmhgbm. Main, P. L. 1979 Desirable Amibutes for a Data-BaaL of Archamlogical Shapw. In Computer Applications in Archatology 1979. S. Laflin, ed. Pp. 5-12. Birmiogbam: University ofBhh&am r Malina. J, J. Stdcl, 2 .Vasicek, and T. Velimsky 1975 Some Problems on Description and Uassification of Archaeological (Stone) A r t e f a c t s .Arrhaeographie 5:45-54. Marquardf W. 1978 Advances in Archaeological .%%tion Advances in Archaeological Method and Theory 1; 257-314. Marquardt, W. H. 1979 A Factor Anal* Approach to Seriatim PLains Anthropologist 24: 309-28. McNutt, C. M. 1981 Nearest Neighbors, Boundary Effect, and the Old Flag Triclr: A General Solution. American Antiquity 46: 571-91.

METHODS IN ARCHAEOLOGICAL RESEARCH Moberg, C. 1971 Closing Address. In Mathematics in the Archawlogical and Historical Saences. F. R Hodson, Kendall, D. G., and P. Tauhl, eds. Pp. 551-62. Edinburgh: Edinburgh University Press. Moberg, C. A. 1977 Question sur I'Outillage Mathhatique d'une ArchMogie Sociologique. In R ~ ~ S O M Q I Iet U ~MCthodes ~ Mathbtiques en Archwlogie. M. Borillo, W. F. de la Vega, and A Guenoche, eds. Pp. 183-98. Paris: Editions du CNRS. Mueller, J. W. 1975 Archaeological R c h as Cluster Sampling. In Sampling in Archaeology.J. W. Mueller, 4 . Pp. 3 3 4 4 . Tucson: University of Arizona Ress. Myers, 0. H. 1950 Some Applications of Statistics to Archaeology. Cairo: Government Press. Nance, J. D. 1978 Regional Subsampling and Statistical Inference in Forested Habitats. American Antiquity 43: 172-76. Nance, J. D. 1983 Regional Sampling in Archaeological Survey: The Statistical Perspective. Advance in Archaeological Method and Theory 6: 289-356. Nance, J. D. 1987 Reliability, Validity, and Quantitative Methods in Archaeology. In Quantitative Research in Archaeology. M. S. Aldenderfer, ed. Pp. 244-93. Newbury P a . & Sage Publications. Nance, J. D. and B. F. Ball 1986 No Surprises? The Reliabity and Validity of Test Pit Sampling. American Antiquity 51: 457-83. OShea, J. M. 1985 Cluster Analysis and Mortuary Patterning: An Experimental Assessment. In To pattern the Past. A. Voorrips and S. Loving, eds. Pp. 91-100. Amsterdam PACT. Orton, C. 1971 On the Statistical Sorting and Reoomtruction of the pottery from RomanoBritish Kiln Site. In Mathematics in the Archaeological and Historical Sciences. E R. Hodson, D. G. Kendall, and P. Tautu, eds. pp. 451-59. Edinburgh: Edinburgh univeriay Press. Orton, C. 1978 Is Pottery a Sample?. In Sampling in contemporaryBritish Archaeology. J. E Cherry, C. Gamble, and S. Shenan, eds. 4 . 3 9 9 4 0 2 . Odord: British Archaeological Reports. Onon, C. 1982 Stochastic Recess and Archaeological Mechanism in Spatial Analysis. Journal of ArchadogicalScience 9: 1-23. Onowsy, B. 1974 Cluster Analysis of Impurity Pattern in Armoricao-British Daggers Archaeometry 16: 221-31. Parker, S. 1985 Predictive Modeling of Site Settlement Systems Using Multiwuiate Logistics In For Concordance in Archaeological Analysis. C. Cur, ed. Pp. 173-207. Kansas City Westport Publisher, Inc. Peacock, W. 1978 Pmbabilirtic Sampling in Shell Middens: A Case Study From Oronsay, Inner Hebrides. In Sampling in Contempomy British Archeology. J. E Cherry, C. Camble, and S. Shenan, e . Pp. 177-90. Oxford: British Archological Reports. Pehie, W.M. F. 1899 Sequences in Prehistoric Remains. Journal of the A n ~ l o g i c a l Institute 29: 295-301. Phagaa, C J. 1985 Dolorn Archedcgical Program Projectile Point Analysis. Report !I DAP-255: Dolores Archadogical Project. Philip, G. and B. S. Ottaway 1983 Mixed Data Cluster Analysis: An Illustration Using Cypriot Hooked-Tang Weapons. Amhaeometry 25: 119-33. Pinder, D., I. Shimada, and D. Chgory 1979 The Nearest-Neighbor Statistic Archaeological Application and New Directions. Amexkau Antiquity 44: 430--45. Pitts, M. W. 1978 On the Shape of Waste Flakes as an Index of Teehnol4gical Change in Lithic Industries. Journal of Archaeological Scieace5: 17-37. Flog, F, M. Weide, and M. Stewart 1977 Research Design in the S U N Y - B ' i t o n Contract Rogram. In Conservation Archawlogy. M. Schiffc~ and G. Gumerman, eds. Pp. 107-20. New York: Academic Press.

DWIGHT W. READ PIog, S. 1978 Sampling in Archaeological Surveys: A Critique. American Antiquity 43: 280-85. Powell, M. J. D. 1964 An Efficient Method for F i n d i n g the Minimum of a Function of Several Variables Without CalculatingDerivatives. The Computer Journal 7: 155-62. Rao, C. R 1971 Taxonomy in Anthropology. In Mathematics in the Archaeological and Historical Sciences. F. R Hodson, D. G. Kendall, and P. Tautu, 4 s . 4. 19-29. Edinburgh: Edinburgh Univmity Press. Read, D. W . 1974 Some Comments on Typologes in Archaeology and an Outline of a Methodology. American Antiquity 39: 2 1 6 4 2 . Read, D. W. 1982 Toward a Thaory of Archaeological Classification. In Essays on Archaeological Typology. R Whallon and J. A. Brown, eds. Pp. 56-92. Evanston: Colter for American Archeology Pr. Read, D. W. 1985 The Substance of Archaeological Analysis and the Mold of Statistical Method: Enlightenment out of Discordance. In For Concordance in Archaeological Analysis. C. Carr,ad. pp. 45-86. Kansas City Westport Press. Read, D. W. 1986 Sampling Procedures for Regional Surveys: A Problem of Representstivoless and Effectivems. Journal of Fidd Archaedogy 13: 103-22. Read, D. W. 1987a Some Comments on the Utility of Mathematical Constructs in Building Arcbaedogical Theory. Paper Presented at the commission 4 Symposium, 11th ccagms of tht UIPPS. Read, D. W. 1987b Archaeological Theory and Statistical Methods: Discordance, Rsolution, and New Directions. In Quantitative Research in Archaeology. M. S. Aldenderfer, ed. 4.151-84. Newbury Park: Sage Publications. Read, D. W. 1988 Intuitive Typology and Automatic Classitication: Divergence or Fuli Circle?. Journal of Anthropological Archaeology (in press). Read, D. W . and A Christenson 1978 Comments on 'Cluster Analysis and Archaeological Ckification'. Americanhtiquity 43: 5 0 5 4 6 . Read, D. and S. LeBlanc 1978 Descriptive Statements, Covering Laws, and Theories in Archaeology. Currmt Anthropology 19: 307-35. Redman, C. L. 1973 Multi-Stage Fiddwork and Analylical Techniques. American Antiquity 38: 61-79. R@r, M. S. 1970 Non FCcondiGC du Mod& Statistique GCnCral de l a Classification Automatique. In Archtologie et Calculateurs. J. C. Gardin, ed. Pp. 301-06. Parig Editions du CNRS. Regnier, S. 1977 Seriation des Niveaux de Plusieurs Trances de F o d e dans une Zone Archaeologique Homoghe. In Raisomement et Mithodes Mathhtiques en Arehblogie. M. Bodlo, W . F. de la Vega, and A. Guenoche, eds. Pp. 146--56. Paris: Editions du CNRS. Reynolds, R G. 1986 Multidimensional ScaJing of Four Guila Naquia Living Floors. In Guila Naquis Archaic Foragiog and Early Agriculture in Oaxaca, Mexico. K . V. Flannuy, ed. Pp. 3 8 5 4 2 3 . New York AcadcmicPress. Robinson, N. S. 1951 A Method for Chronologically Ordering Archaeological Dcposik. Amai~an Antiqiquity 15:293-301. Rowlett, R M. and R B. Pohac 1971 Multivariate Analysis of Mamian La Tene Cultural Group. In Mathematics in the Archaeolagical and Historical Sciences. F. R Hodson, D. G. Kendall, and P. Tautu, eds. 4.46-58. EdinEdinburghUniversity Press. k m y , J. G. 1978 -1ogie de 1'Epipaleolithique (Mesolithique) FraumBelge. Bulletin de la Socitt6ArchCologique. Champcnoise.Special Volume. Salmoa,M.1982Philosophy and Archaeology. New Yo*: Academic Pnss. Schep, S. 1982 StatisticalBlight & m a i m Antiquity47: 836-51.

METHODS IN ARCHAEOLOGICAL RESEARCH Schieppati,F. 1985 Intrasite StruchlralAnalysis Using Semivariograms. American Archaeology 5: 47-54. Schiffer, M. B. and J. B. House, eds. 1975 The Cache River Archeological Project: An Experiment in Contract Arche.ology. Arkatisas Archeological Survey Publications ArcheologicalResearch Series 8. Schiffer, M. 1976Behavioral Archeology. New York: Academic Press. Scollar, I, B. Weidner, and 1 . Herzog 1985 A Portable Seriatim Package with Dynamic Memory Allocation in PASCAL. In To Pattern the Past A. Vwmps and S. Lovhg, eds. Pp. 149-40. Amsterdam: PACT. Sibson, R 1977 Multidimensional Scaling in Theory and PracLice. In Raisonnement et M6thodes Mathhtiques en Archblogie. M. Borillo, W. F. dc la Vege, and A. Guenoche, eds. Pp. 73-98. Paris. Editions du CNRS. Simek, J. 1984a A k-means Approach to the Analysis of Spatial Structure in Upper Palaeolirhic Habitation Sites: Le FIageoIct I and F'incevent 36. Oxford: British Archaeological Reports. S i e k , J. 1984b Inkegraw pattern and Context in Spatial Archaeology. Journal of Archaeological Science 11: 405-20. Simek, J. F., A. J. Ammerman, and K.W. KinIigh 1985 Explorations in Heuristic Spatial Analysis:Analyzing the Structure of Material Accumulation Over Space. In To Pattern the Past A. Vwrrips and S. Loving, eds. Amsterdam: PACT. S i e k , J. F. and R R. Larick 1983 The Rcccgnition of Multiple Spatial Pattern A Case Study from the French Paleolithic.Journal of Archaeological Science 10: 165-80. Simek, J. F. and P. W. Leslie 1983 Pattitioning Chi-Square for the Analysis of Frequency Table Data: An Archaeological Application. Journal of Archaeological Science 10: 79-85. Slachmuylder, J. L. 1985a A Seriation Through Correspondence Analpis with Construction of a Continuous Gradient In Procesdings Symposium on Data Analysis. E. Diday, ed. Pp. 131-54. B~ssels:PlenumPrc~s. Slachmuylder, J. L. 1985b Seriation by Correspondence Analysis for Mesolithic Assemblages. In To Pattern the Past A Voorrips and S. Loving, eds. 4. 1 3 7 4 8 . Amsterdak PACT. Sparck-Jones.K. 1970 The Evaluation of Archaeological ClassiIications. In Archblogie et Calculateun.J. C. Gardin, ed. Pp. 245-74. Paris: Editions du CNRS. Spuldibg, A. 1953 Statistical ~ & h n i ~ u e for s the Discovery of Artifact Types. American Antiquity 18: 305-13. Spaulding, A. 1960 Statistical Description and Comparison of Artihct Assemblages. In The Application of Quantitative Methods in ArchaeoIogy. R. F. Heizer and S. F. Cook, ads. Pp. 60-92. New Yo*: WcunerGren Foundation. SpaulcIing, A. 1971 Some Elemcats o f Quantitative Archaeology. In Mathematics in the AdmeoIogical and Historical Sciences. F. R Hodson, D. G. Kendall, and P. Tautu, eds. 4.3-16. Edinburgh: Edinburgh University Press. Spaulding, A. 1977 On Growth and Form in Archaedogy. Journal of Anthropological R e s e a r c h 33: 1-15. Stafford, C. R and B. D. Stafiord 1983 The Functional Hypothesis: A Formal Approach to Usewear Experhunts and Settlement-Subsiitcnoe. Journal of Anthropological W c h 39: 351-75. Thorn, A 1967Megalithic Sitesin Britain. Oxford: Clarendon Press. Thomas, D. 1976 F i l i n g Anthropology.New Yo*: Holt, Rinehart and Winston. Thomas, D. H. 1978 The Awhrl Truth About Statistics in Archaeology. American Antiquity 43: 2 3 1 4 4 .

DWIGHT W. READ Tyldesley, I. A., J. S. Johnson, and S. R. Snape 1985 'Shape' in Archaeological Artifacts: s i n g a New Analytical Method. Oxford Journal of Archaeology 4: Two Case Studies U
19-30.

A Survey of Human Biodemography


KENNETH M. WEISS
Departmenl ofAnrhropology

Vierra, R K. and D. L. Carison 1981 Factor Analysis, Random Data, and Patterned R e s u l t s .American Antiquity 46: 272-83. Voonips, A. 1987 Formal and Statistical Models in Archaeology. In Quantitative Research in Archaeology. M. S. Aldenderfer, ed. Pp. 61-72. Newbury Park: SakePublications. Vwmps, A. and S. H. L o & & eds. 1985 To Pattern the Part. Amsterdam:PACT. Voorrips, A., D. m r d , and A. Ammerman 1978 Toward an Evalmtion of Sampling Strategies: Simulated Excavations Using Stratified Sampling Designs. In Sampling in Contemporary British Archaeology. J. F. Cherry, C. Gamble, and S. Shennan, eds. Pp. 227-62. Oxford:British Archaeological Reprots. Voorrips. A, S. Loving,and J. Smckee 1985 The Gamma Mix Density: A New Statistical Model for Some Aspects of System Organization. In To Pattern the Past. A. Voorrips and S. Loving,eds. 4.323-48. Amsterdam: PACT. Voorrips, A. and J. O'Shea 1987 Conditional Spatial Patterning: Beyond the Nearest Neighbor. AmericanAntiquity 52: SOU-21. Werner, J. 1953 Das Alnmannische Graberfeld Von Bulach, Basel. In Monographien Zur Urund Fruhgeschichtc der Schweiz. Whallon, R 1972. A New Approach to Pottery Typology. American Antiqwty 37: 13-33. Whallon, R 1973 Spatial A m + % of Occupation Floors I: Application of Dimensional Analysis of Variance. kmrican Antiquity 38: 266-78. W o n , R 1977 The Application of Fonnal Methods of Typology in Archaeological A n a l y s i s . In R a i ~ ~ W O ret nt Mlthodes MathCmatiques en Archhlogie. M Bodlo, W. F. de la Vega, and A. Gueaoche,eds. 4.201-13. Pain: Editions du CNRS. W o n , R 1984 Unconstrained Clustering for the A n a l y s i s of Spatial Distributions in Archaeology. la Instrasite Spatial Analysis. H. J. Hierala, ed. Pp. 242-77. Cambridge: Cambridge University Press. W o n . R. 1987 Simple Statish. in Quantitative Research in Archaeology. M. S. Aldenderfer, ed. 4 . 135-50. Newbury park: Sage Publications. White, T. E. 1953 A Method of Calculating the Dietary Percentages of Various Food Animals Utilized by Aboriginal Peoples. American Antiquity 18: 396-98. Whitley, D. S. and A. V. Clark 1985 Spatial Autocorrelation T e s t s and the Classic Maya Collapse: Methods and Inferences. Journal of Archaeological Science 12: 377-95. Wid, C. J. and R K. Nichol 1983 Estimation of the Original Number of Individuals from Paired Bone Counts Using Estimators of the Krana Type. Journal of Field Archa6 ology 10: 337-44. Wilkinson, E. M. 1971 Archaeological Seriation and t k Travelling Salesman Problem. In Mathematics in the Archaeological and Historical Sciences. F. R Hodson, D. G. Kendall, and P. Tautu, eds. Pp. 276-83. Edinburgh: Edinburgh University Press W i l k i n , e. M. 1974 Techniques of Data Analysis Seriation Theory. Archaeophysica 5: 7-142. Zubmw, E. 1985 Fractals, Cllltural Behavior and Prehistory. American Archaeology 5:

and
Gmduate Program in Genehcs

Pennsylvania Shue Universily UniversityPark, PA 1 W 2 U.S.A.

ABSTRACX Human biodemography, the biological study of human populations, has undergone profound changes in recent years. Many of these changes have bem brought about by developments in statistical methodology and computing, coupled with the availability of new biological data more directly relaled to the phenomena of interest. Methods of assaying biological tissues,.many based on new DNA and immuwlogical techniques, have allowed us to address questions we could previously only speculate about Along with these technical developments has come a shih from a determinisic view of human population biology, variation, and evolution, to a more realistically stochastic view. This view takes into account probabilistica s p of genetic evolution and ddemgraphy, as well as field sampling and data collection. Some very recent trends suggest that even current probabilistic approaches may under-evaluate the importance of heterogeneity among individuals or of truly non-repetitive events, suggesting the need to develop a new class of models. The sophisticated analytical tools and laboratory techniques that are becoming available augur well for the future, but will require more thorough technical training for anthropological population biologists.
KEY WORDS: demography, genetics, population, biodemography

63-76.

There is probably no problem in natural or social science that cannot be thought of in intuitive or qualitative terms. Many problems, however, cannot be addressed adequately without recourse to quantitative methods. Sometimes these methods are analytical in nature; that is, they can be solved theoretically with formal models without the need for statistics, simulation, or numerical approximation, or they involve variables and their inter-relationships which directly represent the natural world. Such analytical methods can be used for fonnal deduction. For example, we may wish to test whether genes are passed from parent to offspring in mendelian fashion. Often, however, the theory is too complex to be worked out analytically, or the role of stochastic (probabilistic) factors too great for tractable solution. Models related to such phenomena are essentially descriptive and approximate, and inferences based on them involve testing the acceptability of quantitative hypotheses rather than
Journal of ~ ~ ~ ~ ~ A h1 (1989) p o 79-1 l o 51. g 8 1989 by Kluwcr Academic Publishers. ~

You might also like