A Study of Style and Function in a Class of Tools David J. Meltzer University of Washington Seattle, Washington In this paper the author develops, and then demonstrates, a method for differen- tiating stylistic and functional variability in the morphotos y of stone tools. Style ‘and function are defined in evolutionary terms, then related t0 tool morphology and technology. Within that framework, predictions are made about the spatial and temporal patterning expected if morphology is functional andlor stylistic These expectations are then tested through comparison, using discriminant func- tion analysis, of stone tools fromthe Late Prehistoric in Nw North America and the Late Pleistocene in northern Africa The comparison reveals that the tools, endscrapers, vary only in terms of their ‘amount of wear; their morphology is identical between the samples separated by 10,000 years in time and 5,000 miles in space. Based on the theoretical discus- sion and these observations, it is argued that endscraper morphology is entirely functional, and contains no stylistic component. Morphological variability in these tools indicates adaptation, not time. Introduction The obvious repetition of shapes by the thousands can scarcely be attributed to chance.' While many of our stone-tool typologies are basically ‘morphological in character and derivation,’ their use has, been extended to the interpretation of archaeological activ- ities and t0 the delineation of temporal and spatial groups. In other words, tool morphology is used to indicate, re- spectively, function and style. Yet, because these ancillary uses of the typologies are often made without consideration, of the actual presence of either style or function, the resultant, imerpretations have often seemed contradictory. Consider, for example, the debate on Mousterian varia- bility, The same assemblages are defined as being stylistic, representing cultural groups,? and as being functional, rep- resenting different activity areas.* In the absence of empir- 1. E, Boedes, “Comment,” CA 20 (1979) 10-11, 2. F. Bordes, Typolgie du Paléolhique ancien et moyen (Bordeaux 1961). Tse. Typologe de FEpipaéoihique du Maghreb (Pars 1963) 3. F-Bonles and D. deSonneile-Bordes, The significance of variability in Paleolithic assemblages,” WA 2 (1970) 61-73, 4. L. R. Binfod, "Model Bulding-Parsdigms, and the Curent State of Paleolithic Research,” in L. R. Binford, An Archaological Perspective (New York, 1972) 244-294; L: R. Binford and S. R. Binford, "A Pre liminary Analysis of Functional Variability inthe Mousteran of Levallois Faces." in Recent Studies in Paleoanhropoton, J.D. Clik and F.C. Howell, ed, Amat 68:2, Part 2 (1066) 238-295, ical evidence that specifies the relative amount of stylistic and functional variability within an assemblage and within, a tool class, such debates will continue without clear res- ‘lution. The debates will continue because these analyses are based on morphological typologies, and those mor- phological types allow us to assume only that some form of selective behavior is taking place; they make no allowance for what that selective behavior might be. The a priori assumption that style is present in all ‘shaped’ tools, or in, ‘one or another ofthe attributes of those tools,® while in part correct, is unwarranted logically and empirically. In order to understand the morphological patterns in a class of tools, we have to determine empirically the relations among style, function, and shape. This kind of analysis, ‘must be done on a class-by-class basis because each tool, class contains different amounts of stylistic and functional clements. In this paper I discuss a method developed to separate style and function in tool morphology. I initially consider the general theoretical structure behind the method explain- ing why, in evolutionary terms, the method works. Then I discuss the incorporation of style and function in the tech= nological process of tool production and morphology. This 5. E. Wilmsen, Lindenmeir: a pleisocene hunting society (New York 1974) 98 6. D. Sis, “Paleolithic cultwe and cultural change’ experiment in theory an method,” CA 20 (1999) 1-21: see pp. 3-4314A Study of Style and Function in a Class of ToolsiMeltzer discussion is followed by a sketch of the method itself, specifying the tests and the data requirements. Finally, an example is given for a particular tool class (‘endscrapers’), and the presence/absence of stylistic and functional varia bility is determined, Style, Function and Shape In formal terms, style and funetion are the fundamental sources of variability in archaeological materials.” In defin- ing style in evolutionary terms, one has to be cognizant of ‘ovo separate, but interrelated aspects of stylistic phenom- ena, At one level, style can be defined as those forms in cultural system that do not have detectable selective value: those forms are adaptively ‘neutral’.* On another level, how= ever, because style, like all elements of cultural baggage, hhas a cost in energy, space, and matter, it remains a part of the total selective picture of the cultural system. So in general sense style has a function because, by virtue of its independence from its environment, it can be employed to delineate spatial interaction and demarcate cultural boundaries.” In other words, while particular styles are not directly accountable by natural selection, style in general hhas a role in the cultural system. This difference is primarily one of scale, and can be illustrated as follows. In many instances, the choice between certain kinds of design elements on ceramies is not a func Teal Perspectives.” in New Perspectives in Archaeology, L. R. Binfont and $. R. Binfor, eds. (Chicago 1968) $-32; Binford and Binford, op. cit. (in note 4); RC. Dunpel, Style and Function: 2 Fundamental Dichotomy." Amant 43 (1978) 192-202; A Jelinek, “Form, function and style in lithic analysis," ia Cultural Change and Continuity, C. Cleland, ced. (New You: 1976) 19-34; JR. Socket, “Style, Function and Artifact Variability in Pleolithic Assemblages." in The Explanation of Culture Change: Models in Prehistory, C. Renftew, ed. (Pitsburgh 1973) 5317-325: R. Sacket, The Meaning of Syl in Archaology: a General Mode,” Amant 42 (1977) 369-380, 8, Dunne, op it. (in note 7) 199 In this analysis the term “noua is sed to coanote ems that cant be differentiated by the selective envi ronment, items that can be freely created, replaced. or disposed of inthe ‘altura system ith no deltios impact on the functioning of that system, Tn this sense, two items tha full entcal roles and are in some sense functionally equivalent can be treated as “neural phenoment, because, ‘while each has “Seletive value, they canbe substiuted for one another ‘without changing the selective lance of the system, E, Mayr, Population, Specie, and voltion (Cambridge 1970) 128-127, deals with the anal ‘fous issue of biologically “neural ta 9. Binfor, op. cit. (in ote 4); A. Close, “The identification of style in lithic arfacta.” WA 10 (1978) 223-237; M. W. Conkey, "Style and Fuetion in Cultural Evolution: Toward a Predictive Model for the Paleo- lithic.” im Social Archaeology: Beyond Subsistence and Dating, C. Redman et. a. eds. (New York 1978) 61-5: H. M. Wos, “Stylistic Behavior and Information Exchange.” in Papers forthe Director: Research Essays in Honor of Janes B. Grif, C. Cleland, ed. (Ann Arbor 1977) 317-382, tional consideration, but rather is historically determined and selectively ‘neutral’, because there is no inherent ad- vantage between one element and the next." The actual presence of the design, however, has a selective value, because that particular design serves to mark a certain in- dividual or group boundary (or whatever other function it may serve). Archaeologists are most familiar with the former instance ‘of stylistic behavior because it isthe historical and ‘neutral’ basis of stylistic attributes that causes them to take on in ‘a temporal dimension that peculiar Markovian ("Battleship Curve") pattern! that makes seriation work. Because stylistic similarity, with the exception of fortuitous circum- stances, is necessarily homologous similarity, the result of cultural contact or common descent," and because particular styles are independent of external conditions, one can obtain the purely historical, non-repetitive classes used to tell time." Function, on the other hand, includes those elements that directly effect the Darwinian fitness of the populations in which they occur." These elements are traits that can be accounted for by selective factors. For this reason, func- tional similarity, while it may be homologous, may also be analogous, the result of similar activities or adaptations in similar environments.'* The replication of functional traits is not dependent on direct character transmission. Function is most often identified empirically on the basis of experimentally or analogically inferred use(s). Tools are recognized as tools because they hold ‘shape’ patterns— ‘and more recently macro- or micro-wear patterns—in com- ‘mon with items whose ethnographic role is known or can bbe replicated through experimentation," In this analysis, function is not treated as a synonym of use. Instead, it is defined solely in terms of the presences absence of wear on the artifact.” This approach is taken for 10, Sackett, op. cit. (in note 7) 378. 11. One sees similar pater, for some of the same reasons, in the ‘phylogenetic paters of clade diversity in many taxa: see S.J- Gould et Al. "The shape of evolution: comparison of rel and random clades. Paleobiology 3 (1977) 23-80. 12, Binford, op. it (i note 7) 8-12; Sackett, op. cit in note 7) 377 13, Duell, op. ct (in noe 7) 199, 14, ia 15. Binford, op. ct. (in note 7) B12; E.R. Service, “Archaeological ‘Theory and Ethnological Fact,” in Process and Patern in Cule, R Manners, ed (Chicago 1964) 364-375, exp. pp. 371-372 16, See e.g, L. H. Keeley and M. H, Newcomer, “Microwear analysis of experimental flint tols: a testcase," JAS 4 (1977) 29-02; S. A Semenov. Prehistoric Tecnology translated by M. W. Thompson (London 1968). 17. The rationale for these arguments an be found in R. C. Dunwll,two reasons. First, knowledge of prehistoric use is not nu cial to the method. One must have members of the same tool class, and this simply requires tools with identical wear patterns." Second, and more important, there is good reason to suspect that wear isthe only valid and empirically ree- gnizable manifestation of prehistoric function. While one can pick up an object and infer a use, that inference is not subject to meaningful tests ofits validity in archaeological data, Nor is this circumstance particularly surprising. After all, how can we expect 0 replicate, or expect the ethno- graphic record to duplicate, all manner of prehistoric be- havior? How could we know it even if it did?” To make these rater esoteric issues more concrete, consider also that the basic reliability of even the better experimental ap- proaches is suspect? Until these issues are resolved, itis best to treat function as wear; a definition of function on the basis of an inferred use is at best a guess, and at worst downright misleading. ‘As mentioned, shape is also indirectly used as a basis for defining function, especially in those instances where prehistoric and ethnographic morphological patterns are matched, and thereby assumed to have identical uses. This use, of course, implies that tools cluster in discrete mor- phological pattems, and thatthe ‘shaped’ tools are directly comparable across temporal and spatial boundaries. Aside from the difficulties this approach presents when there is an absence of analogs," the use of morphology raises an- ‘other issue: what of the tools that are “non-shaped’, those whose morphology does not reflect the traditional ‘shape? We cannot assume, as this position demands, that in order for a variable such as shape to have archaeological signif- icance it must fall into dserete clusters. Shape, like length, Width, thickness, and so on, isa continuous variable whose distribution can show a wide range of variation. Since we do not assume, a prior, that all important archaeological “Archaeological Potential of Antropoogical and Scientific Models of Function,” in Archaeological essays in honor of Png B. Rouse. RC Dunnel and E.S. Hal, es. (The Hague 1978) 41-73. It shoud also be mentioned that the analysis of wear done here is mcro- and not micro 18 R. C, Dunnell and C. Beck, The Caples Site, 45-SA-3, Skamania Coun, Washington (Seale 1979) 89-60, 19, J.A. Sabo, T.W, Beale and A. M. Kurland, “Recent Developments in Archaeology. Annals of the American Academy of Political and Social Science 408 (1973) 108-118, 20. G. Holley and. A. Del Bene, “An evaluation of Kesley's microwear spproach,” JAS tin press, 198. 21. W. Shawcros, “Stone Flake Indes in New Zealand,” JPS 73 (1968) 7-25: 1. P. White, “Typolopes for some Prehistoric Flaked Stone Anttacs in the Austalian New Guinea Highlands", Archaeology and Physical Auhropology in Oceania 4 (1969) 18-46. Journal of Field ArchaeologyiVol. 8, 1981 315 items have to be at least 5 cm. in length, why assume that tools have to be patterned in their morphology? Defining tools by shape forces us to view interval level data in nom- inal terms, with the attendant loss of information. Moreover, defining tools by their morphology assumes that the object itself is a tool and thus ignores variability of the object's attributes (such as wear). An object may well hhave a number of separate and independent tools on it, We cannot, and should not, ignore wear by examining only morphology, nor, for that matter, inflate its meaning with an inferred use. Given these considerations, one can address the relation- ships among shape, style, and function. It must first be recognized that too! morphology is determined by tool tech~ nology, which is in turn determined by the functional re- quirements of the tool class. ‘All tool classes have a set of minimal functional require- ‘ments that must be fulfilled in order for the tool to *work’. To take a more extreme example, projectile points require 4 point, axial symmetry, relatively uniform thickness, and a provision for hafting. These conditions are rarely met in “run-of-the-mill” flakes, and thus these tools almost always exhibit some degree of manufacture. In most instances, however, the functional requirements are less severe and they can be satisfied in a common flake assemblage. In general, it can be argued that it is the ratio of required parameters to available parameters that principally deter- ‘mine the degree of manufacture necessary. A low ratio of required parameters to available parameters indicates that simple selection and subsequent utilization ofa flake without ‘modification is probably sufficient. A high ratio of required parameters to available parameters indicates manufacture is unavoidable. While theoretically reasonable, this distinction is not al- ways clear cut empirically because many classes of tools have both manufactured and non-manufactured members Indeed, relatively few classes have solely manufactured ‘members. This observation brings up a couple of important questions. First, why are there so few classes containing exclusively manufactured items? And, second, why is it that in classes where the required parameters are met in a flake assemblage are there any manufactured members? ‘The first question has a rather simple explanation. As Wilmsen argues, flaking techniques themselves were most likely directed ‘towards the production of flakes that could be converted into tools with a minimum of further modi- fication’™®" Most tools and most tool classes should there- fore show litle modification. One would suppose, however 22. B, Wilmsen, Lithc Analysis and Cultural Inference (Tucson 1970) 6-61, 23. i316A Study (and this addresses the second question), that areas low in raw material might exhibit higher percentages of manufac~ tured tools, perhaps indicating tool curation, or, possibly, efforts to make the tools more efficient, either by prolonging their use-life or having them serve a number of different functions.** The results of my own research indicate that there seems to be a relatively high correlation between man- uufactured tools and heavily worn ones, but whether this indicates a more intensive short term use oF a more extensive long term use is unclear.** In either case, however, this high correlation can be interpreted as increased tool efficiency. Itis also not unreasonable to predict that those tool classes ‘that do contain a proportion of manufactured _ members would have the greater potential of including stylistic attri- ‘utes in the tool morphology. Afterall, the circumstances where one can select for stylistic attributes within a flake assemblage are relatively uncommon, though might include such things as selection for particular kinds of raw material, such as exotie material.” The salient points of shape, style, and function can be summarized as follows. Morphological variability can be either functional, or functional and stylistic. That varia- bility can be generated one of two ways: through manufac~ ture or through selection. The choice between manufacture ‘and selection is dependent principally on the relationship between the minimal functional requirements of the tool, ‘and the degree that those parameters are met in a flake assemblage. In more specific terms, one would manufacture largely because the ratio of required parameters to available parameters is approximately greater than or equal to 1, or because manufacture makes the tool more efficient.” The latter route presumably is important in areas where raw material is less abundant. 24. A. Goodyear “A Hypothesis forthe use of Crypocrystlline Raw “Materials among Paleo-Indlan Groups of Nort America,” Research Man script Series No, 156 (University of South Carolina: Columbia 1979) 4, 2 25. DJ. Meltzer, “A Stuy of Style and Function in a Cas of Tools,” Unpublished M. A. research paper, on ile in Department of Ambropoogy University of Washington (1979) 89-90 26, Wilmsen, op cit. (in note 5) 93. 27. R.A. Gould, D. A. Koster and A, H. L Sonte, “The Litic Assen blage ofthe Westem Desert Aborigines of Ausalia." Amant 36 (1971) 149-169; se p. 161 28. The possibility that variability might be soley stylistic an be ignored Since all tools, by definition, are functional to one degre or another 29. My use of a numerical value (ati greater or Hess than 1) inthis ‘iscusson is heuristic, and not an empirical observation about the at ‘hacological record, of 4 reflection of static relation inthe choice of manufacture versus seletion. The rato at which manufsctare willbe invoked will undoubedly vary: of Sole and Function in a Class of ToolsiMeltzer ‘One would opt for selection rather than manufacture in those instances where the ratio of required parameters to available parameters is approximately less than or equal to 1; cases where manufacture is not necessary for the tool to ‘work’. Hence, while some tool classes may have easily satisfied requirements, with manufacture non-obligatory, some individual rools within those classes may exhibit man- facture on those flakes not readily available for use, oF to increase tool efficiency, and so on. The relative abundance of the manufactured versus selected members of the tool class reveals a great deal about the minimal functional re- Quirements, tool efficiency, and raw material availability. From the above discussion it should be apparent that there are predictable temporal and spatial patterns in the stylistic and functional variability in archaeological materials. Ex- actly how that patterning can be determined empirically in the morphological attributes of a tool class is examined in the next section. ‘Methodological Considerations ‘The method developed here, in its most basie form, con- sists of expressing morphological variability as an hypoth- esis, then testing whether that variability is functional, or functional and stylistic. The tests, in this case using This approach, which simply identifies the kind of wear evident macro- scopically on an object, is in some respects cumbersome, since ‘axes’, ‘adzes’, ‘drills’, and other familiar names are replaced by a system of numerically designated functional classes. Nonetheless, it does have the virtue of allowing systematic and quantitative comparison of identically de- {fined tools.* The definition of a 12133 tool is as follows, The kind of wear exhibited is chipping, found along a unifacial edge that is convex in shape. The edge angle is high, ranging between 60 and 90 degrees; the wear intersects the edge perpendicularly. This class clearly incorporates those tools ‘commonly called “endscrapers’, but is not itself a definition Of endscrapers because it includes many of the “unshaped” varieties of this tool, those traditionally relegated to the category of ‘utilized flake” or ‘flake tool’ ‘The Lower Columbia sample consists of 241 tools from a series of Late Prehistoric sites; these date to between ca 1300 A.C. and the period of European contract (1775-1800 in this region). The assemblages that the tools come from seem {0 represent a single activity, the net fishing of stur- ‘ge0n, and probably are one component in a seasonal round. AAs required, the sample was collected and analyzed in an ‘unbiased manner. All tools, regardless of their shape, were subjected to the wear-based functional classification. ‘The second sample consists of 33 tools, the majority from ‘a Late Pleistocene site in northem Africa (29 tools in all)” the remainder are from Kentucky (2 tools), Missouri (1 tool), and West Virginia (1 tool). It is worth mentioning that these tools were collected in the traditional manner, and as a result most of the tools are of the ‘shaped’ variety; there are only a few “flake” tools among them. Yet this will rot have a prejudicial affect on the analysis because, in fact, they exhibit the morphological patterning that is of interest It is advantageous in this instance to have, as it were, a “shaped” sample to compare with a sample that is both “shaped” and ‘non-shaped’ 33, R.C, Dunnell and. K. Campbell, Aboriginal occpation of Hamilton Iuland Washington (Seat 1977) 36 34. R. C, Dunnell, Systematics in Prehistory (New York 1971) 73-74: DDunnell and Beck, op ct (in noe 18) 89 35, Dunnell and Camptell, op. cit. (a note 33) 188 36, Ibid, 34. 37, These anifats were kindly provided by G. T. Jones. 38, These stifacts were kindly proved by R. C. Duel Figure 2, Disgram ofa tol showing measurements taken in analysis 1 Maximum width of tol. 2. Wid of distal edge. 3. Width of worn edge. 4. Distance from point of maximum width to distal edge 5. Distance from wor edge to distal edge. 6. Width af proximal edge 37. Maximum thickness of ool. 8 Distance from point of maximum thiekness to distal edge. 9. Length of too This complementary data set provides the basis for com. parison of widely divergent spatial and temporal locales. ‘The rather peculiar mix of tools, particularly those from the ceastern United States, serves as insurance against the ex- ceedingly small possibility of cultural contact. We can as- sume that there will not be homologous similarity between the Late Pleistocene groups of Libya, and the Archaic and Late Prehistoric groups of the eastern United States and the Lower Columbia, respectively. ‘A few points regarding the data analysis are also worth mentioning. First, only manufacture related to the mor- phology of the object is examined. Manufacture of the worn ‘edge, such as resharpening, is not considered in analysis because it is not directly related to the process of modifying the overall morphology of the tool, and, more to the point, because it cannot easily be distinguished from wear that co- ‘occurs upon that same edge. ‘Second, the ordinal level distinction between “greater* and ‘lesser’ wear is based on the ratio between the width of the worn area and the width of the distal edge, the edge ‘on which the wear occurs. Variation in the amount of wear is, of course, continuous and not discrete; consequently the imposition of ordinal categories is arbitrary. For this anal- ysis, a ratio value of .90 or greater (i.e., greater than 90% of the distal edge is continuously worn) delineates “greater’ ‘wear; below that point is ‘lesser’ wear. This particular cut- point is tempered by empirical considerations. One of the factors in its selection is that it nearly bifurcates the distti- bution of this wear-related ratio value. Third, as mentioned, once manufacture and wear are as- certained, the tools are grouped according tothe contingency table of amount of wear versus presence/absence of man-‘Table 1. Distribution of tools by cel, ‘Cell Number Sample 1234 Tons Lower Columbia -34—«109—=«SS«92S DT Northern Arica’ EEastem United States. 921 tufacture (FIG. 1). Each of the cells in the contingency table is labelled, The emphasis here is primarily on Cells 1 and 2, because these tools exhibit more wear and will yield more information on the stylistic and functional variability in tool, morphology than do tools tabulated in Cells 3 and 4. The distribution of the tools by cell from both samples is given in Table 1 Finally, the set of measurements taken on each tool is| shown in Figure 2. These measurements cover the overall tool morphology, as well as the variation in the worn edge. ‘The acronyms used in the computer analysis, and retained here for the sake of continuity, are listed in Table 2. Analyses ‘There are a variety of statistical techniques that could be used for this kind of study; these include factor analysis, cluster analysis, -tests, and discriminant analysis. The dis- criminant technique was selected because it has a number of features that, either alone or in combination, make it ‘more suitable for answering the kinds of questions being asked, First, the discriminant technique allows the quantitative separation of qualitatively defined groups. The various Table 2. Variable acronyms used in the analysis, ‘VARIABLE, ‘ACRONYM | 1. Maximum width of toot MAXWID, 2, Width of distal edge DISTWD, 3. Width of worn edge WEARWD 4. Distance from point of ‘maximum width to distal edge WIDTDE 5. Distance from worn edge t0 distal edge WEATDE, 6. Width of proximal edge PROXWD, 7. Maximum thickness of tool MAXTHK 8. Distance from point of | ‘maximum thickness to distal | edge ‘THKTDE 9, Length of tool LENGTH 10. Tool weight WEIGHT Journal of Field ArchaeologyiVol. 8, 1981 319 ‘groups, the cells in Figure 1, can be statistically distin- guished on the basis of the morphological attributes of the tools themselves. ‘Second, the discriminant technique reveals the significant underlying variability in tool morphology in a hierarchical fashion, by selecting, in order of their discriminatory power, the variables that best distinguish the cells. These variables, are then weighted and combined linearly such thatthe groups, formed reflect those significant aspects of variability. ‘Third, discriminant analysis creates a set of classification functions that enable the post-hoc identification of the tools, ‘on which the analysis is based. This identification serves as a gauge of the reliability of the procedure, as well as an indicator of within-group homogeneity. Finally, the discriminant technique allows the identifi- cation and classification of additional cases not used in the estimation of the discriminanticlassification functions. Thus ‘one can test independently the ‘fit’ of the functions derived from, say, the Lower Columbia sample, on the tools from the composite northern African/eastern United States sam- pile. Were the tools from all areas a part of the estimation of the functions, classification of the composite northem Africa/easter United States sample would only test the reliability of the analysis, and not reveal valid differences between those tools and the tools from the Lower Columbia Although some of the other techniques would provide certain of these features.” discriminant analysis combines all the information and thus most efficiently yields a solution, The statistical requirements of the technique include, pri- ‘marily, that the data are on an interval level and normally distributed, The later condition need not always be adhered to owing to the robustness of the program and available compensatory statistcs.*° The data used here, nonetheless, satisfy both requirements. ‘Aside from these overtly statistical requirements, there are other, more subtle requirements and assumptions that ‘must be made, Among the more important are that our initial separation of the data is correct, and that the quantitative variables chosen are reasonable indicators of the desired information. The analysis will give results without question, for comment; it could not point out, for example, that the variables chosen to represent morphology may need refine- ‘ment, of that the 2 2 contingency table of manufacture versus wear may not be a useful means of exploring vari- ability. The technique itself is not inherently infallible; those 39. Factor analysis, for example, would provide a good measure of the underying variability in tol morphology, and a sees of rests would _Bve a resonable measure of differences between pars of groups on a ‘ariable-by-variale bass 40. P. A Lachenbruch, Discriminant Analysis (New York 1995) 40-46,520A Study of Style and Function in a Class of ToolsiMeltzer ‘who use the technique even less so, Hence, while the results of the analyses are, in almost all respects, exactly as pre- dicted, their validity is crucially dependent on the theoretical structure, and not on the procedure, as T. H. Huxley once observed Mathematics may be compared to @ mill of exquisite work ‘manship, which grinds you stuff of any degree of fineness; bu, nevertheless, what you get out depends upon what you putin: and the grandest mill inthe world will not extract wheat-flour from peascod, so pages of formulae will nt get a definite result cout of loose data." With that caveat in mind, 1 would like to rephrase the expectations ofthe theoretical model, but put them in terms of the discriminant technique, itself a mill of “exquisite ‘workmanship’. This rephrasing should clarify any ambiguity regarding how or why this technique is being used, and show what form the results can be expected to take. The basic argument is that if morphology is functional, then the pattem within the too class will be relatively ho- rmogeneous. This homogeneity could be manifest in one of two ways in the results of the discriminant analysis. Ether there will be absolutely no significant morphological vari- ability, in which case the only discriminating variables will be wear-telated, or there will be significant morphological variability, which will result in discrete morphological clus- ters. If manufacture is functional, those clusters will be tniform through time and space, In both instances, of course, there will be morphological identity. In the former instance, it will be manifest in the lack of any morphological variables in the discriminant functions; in the latter it will bee evident in the overlap between tools from the Lower Columbia and the composite northern Aricaleastern United States sample. tis apparent that any absence of morphologically related Variables in the discriminant functions would be of great significance, since the absence would exhibit morphological uniformity, demonstrate that the tools vary only in a di: mension directly related to function (tool wear), and avoid potential analytic ambiguity. That ambiguity would arise in the situation where thete are signifpant morphotogicaldif- ferences in the tools. One then has to ascertain whether or not those are significant enough that one can distinguish ‘manufactured from non-manufactured items, yet still argue that there is enough overlap to warrant a conclusion of ‘morphological identity. The resolution of this issue must necessarily be based on empirical evidence, but one can 44, The observation was made, for those with an intrest inthe history ‘of science, in the context of his debate with Lord Kelvin onthe age ofthe carh. TH. Huxley, "Anniversary Addtess ofthe President." Quarterly Journal ofthe Geological Soviets. London 25 (1868) L. predict certain aspects of the results if the morphological differences are entirely functional First, one would expect to see differences between tools that are manufactured and those that are selected; one would not, however, expect differences within the tools that are ‘manufactured. This expectation holds true for manufactured tools from the same locale, as well a in comparisons be- tween locales. Ifthe morphological variability were stylis- tic, one would again see differences between manufactured and selected tools, yet there would also be important con- trasts within the manufactured tools. Most important, if variability were stylistic, there would be no overlap between the Lower Columbian and northern Africaleastern United States tools Second, if the morphology is functional, and if we can presume that it is geared toward making the tool more ef ficient, then we should see overlap between the Cell | and Cell 2 tools. The morphology in both instances is directed toward function (or a result of function), and it would there- fore be unreasonable to expect differences in kind or degree beyond the simple presence of manufacture.“ Ifthe man- ufacture were stylistic, the Cell 1 and Cell 2 tools would share only the minimal functional requirements, and the ‘morphology in the manufactured items would vary by sty- listic features. ‘The behavior of the Cell 3 and Cell 4 tools will not be an important factor here. Both of these will differ from the Cell 1 and Cell 2 tools in terms of wear and, inthe ease of the Cell 4 tools, in the absence of any patterning that might be common to the tools manufactured or selected for more intensive use In sum, the results of the discriminant analysis should show either absolute morphological identity, evidenced by the selection of only wear-elated variables in the disrim- inant functions; or, the results should show some morpho- logical variability that will be demonstrably functional if tere is overlap between the Cell 1 and Cell 2 tools, within the Cell | tools from the different locales, and within the Cell 2 tools from the different locales. This last expectation ‘would also appear if morphology were stylistic because these tools are not manufactured and thus should show little style; nonetheless, the first two expectations are unique to the position that morphology is functional One final note. If it is the case that the wear-related variables are the only significant discriminators, then there may be significant differences that arise between the Lower 42, Because the Cell 1 tools are manufactured, however, they may more tighly cluster in their morphological pater 43, The Cell 3 tools are not dealt within any deal here since they are ‘rather anomalous Set, mostly broken inthe rocess of manufacture and ‘ths of litle analytic valve. The small sample ofthe Cell also makes interpretation somewhat superfluous,Journal of Field ArchaeologyiVol. 8, 1981 321 ‘Table 3. Comparison of maximal and optimal solutions of discriminant analysis. MAXIMAL SOLUTION OPTIMAL SOLUTION Variable (Step) __Frvalue Significance Variable (Step) __F value Significance WEARWD W 24115 01 WEARWD © Bans ‘01 DIsTwD @) 32737 ‘01 DISTWD, 73.2037 01 MAXWID, G) 2.0058 25 LENGTH 4) 1.6290 25 PROXWD 6) 2.0133 25 WEIGHT 6 Last 25 ‘THKTDE L116 50 WIDTDE 1.1100, 50 WEATDE ©) 1.8010, 25 MAXTHK 10) 5663 75 Discriminant ‘Canonical Discriminant Canonical Function Eigenvalue Correlation _ Function Eigenvalue Correlation _ 1 14967577426 90.8 V 13816176165 97.1 2 1202132758 73 2 alos 19887 29 3 03233-17696 19 Percentage of correct classifications Percentage of correct classifications Lower Lower ‘Columbia Composi Total Columbia Composite Toul 8.4% 3.6% 67.8% 65.1% 75.1% 66.45% Columbia and composite northern Africaleastern United States samples.“ This result would yield useful information ‘on the use-life of these tools in the different areas and, of course, on any wear-related size differences between the tools. Results ‘The discriminant analyses were conducted using the BMDP7M Stepwise Discriminant Program. The results of the analysis that are of the most interest here are, first, the determination of the critical variables in the discriminant functions; second, the a posteriori classification of the data;, and third, the plots of the tools and the mean values for the cells (called the group centroids) in discriminant space. The results follow below in that order, For heuristic purposes, two solutions were computed. The first used all 10 of the variables, to insure that if any mor- phological variability were present it would have a chance ‘of appearing in the functions. This is the maximal solution 44. IF these differences appear along with morphological diferences, whether those morphological differences are functional of sist, they are easly factored out by examining the canonical variates ofthe dsr inant functions. 45. BMDP Biomedical Computer Program P Series (Los Angeles 1979) The second used only those variables that proved signifi- cantly to sort the groups. This is the optimal solution.** ‘These results are in Table 3. In the optimal solution, two variables are used to derive ‘wo discriminant functions, In the maximal solution three functions are derived using all 10 variables. The contribu- tions ofthe particular variables tothe discriminant functions and group separation can be measured in at least two ways (1) the step number of their inclusion in the function, and (Q) their approximate F statistic.” While intuitively one 46, The diflerences between the maximal and optimal solutions ae based ‘on the F sttstics ofthe variables. The F static esentialy tvs the all hypothesis that there are no differences between two Or more groups using ‘punicular variable, A high F score ona pticla variable indicates there isa significant difference inthe mean ofthe two groups. If that F score is higher than a re-sabished"Ftocntr™ then the variable i incladed inthe diseriminant fonction. Ifthe F score is lower than the “Foner the vaable isnot included. By manipulation of the “F-orenter™ the analyst can contol the number and kind of variables inclided inthe ds riminant functions. A low “Fio-enter” allows most variables in the function, and resis inthe maximal solution ("F4o-enter” of 0.5 in this ‘nalyss) high “Focenter selects only those variables that best sor the groups. and is the optimal solution ("Fo-enter™ of 3.0 in this analysis) 44. There are, of couse, exer ways, such a8 the canonical corelation ‘eetween a variable and the discriminant function, though these need not ‘once us here522A Study of Style and Function in a Class of ToolsiMeltzer ‘would suspect thatthe discriminant functions based on more variables provide a better analysis of the data, such is not the case Note, first of all, the decreasing F values in the maximal solution. By the third step in the analysis, the variables have ‘ceased to improve the discrimination of the groups in any significant manner. Every variable entered beyond that point is essentially ‘noise’: “certainly no variable that enters after a nonsignificant entry should be considered”. In this case, the magnitude of the drop in significance leaves litle to quibble about Next, observe the statistics that describe functions, es- pecially the third discriminant function in the maximal so- lution. The Eigenvalue, canonical correlation, and percent of variance explained by that function all suggest the func- tion is superfluous. More variables resulting in more func tions do not necessarily yield more and/or better results While the Eigenvalue totals differ slightly, each of the so- lutions explains all the common variance, but the optimal solution is equally effective with a fraction of the variables. Finally, consider the percentages of correct classifications for each of the solutions. Obviously there is little loss of information from the maximal to optimal solutions. While there is a decrease in the total percentage of correct clas- sifications in the optimal solution, a decrease of 1.4% rep- resenting 4 tools not correctly classified, this ‘loss’ of information is hardly sufficient to warrant use of the max- imal solution, For these reasons, but primarily because there are only two variables that significantly discriminate the groups, only the optimal solution is examined from this point on. Keep in mind which two variables proved significant: WEARWD. and DISTWD, wear-related variables.” In Table 4 the classification matrix forthe optimal solution is shown. These results are informative in a number of ways, ‘but most important for the analysis is the relative ‘confusion’ 448. Lachenbruch, op cit. (in note 39) 19, emphasis in orginal 49. The fact that these two measures are taken from the same edge isnot 4 mater of gret statistical concer, While created, that corlation fn empirical fac, ot definitional relation, as in the case whee, say ‘one analyzed series of variables sich as Length, Width, and rato of Length’ Width. There, beease the ratio Length With iS defintionslly related (autocorelated) to variables Length and Width, one would have A significantly titer than random chance af pein the values of some ‘when the values ofthe ater are known. Because there is no definitional reltion between DISTWD and WEARWD one cannot be predicted from the other @ prior), and because WEARWD can vary independent of DISTWD (lege, smaller, the same size, in no deterministic elation). the ‘variables canbe treated 2 independent observations. Any coreation be teen the two becomes 4 matter of empirical significance, as it would imply a functional relationship between the variables, For a careful con ‘dertion of topics of seril correlation, se P. Gould, “s Stars Iferent the Geographical Name for a Wild Goose?” Economic Geography 46 (1970) Par 2 Table 4, Classification matrix forthe optimal solution. [Number of Cases Actual Chasified Into Group Percent Group 1234 Comet Tower [Cell 19 15 0 0 35.9% Columbia | Cell2 377206. 1% Sample | Cell 3 0 1 4 1 66.7% cell 0 9 2 614% Composite | Cell 1 70200 0 7% Sample | Cell 2 4 1 1 0 76.1% cls 0 0 T 65.7% ‘among the various cells. For example, the misidentification of the Cell 1 and Cell 2 tools from the Lower Columbia sample indicates, particularly in light ofthe variables in the discriminant functions, that there is no significant mor- phological difference between tools that are manufactured and those that were simply selected and used. The only differences that exist are in terms of the amount of wear, and even these are equivocal. Just over half of the Cell tools fom the sample are correctly classified as Cell 1 tools; the remainder are all classified as Cell 2 tools. They are not mistaken for either Cells 3 or 4. In faet, of the 15 Cell 1 tools misclassified as Cel 2 tools, 14 are correctly classified by the second probability function. The Cell 2 tools, ex- pected to mimic the Cell | tools, do so: some 34% (37 tools) are classified as Cell I tools, though the remainder are correctly classified. Again, not only do these two cells mimic each other in terms of morphology, but also in terms of the amount of wear. The only apparent differences within the Lower Colum- bian sample are between the Cell | and Cell 2 tools and the Cell 3 and Cell 4 tools. These differences are, as predicted, in the amount of wear. Its admitedly surprising that here there were no morphological differences, particularly since these cells comprise a wide range of both morphology and ‘wear. OF course, the fact that there are no significant mor- phological changes between Cells 1 and 2 versus Cells 3 and 4 simply reinforces the argument that all morphological variability is determined by function and, presumably, the basic process of tool technology. Classification of the composite northern Africafeastern United States sample conforms exactly to expectations. The Cell 1 tools of the composite sample are virtually all (7 of 9 or 78%) correctly identified as Cell 1 tools by the Cell 1 definition of the Lower Columbia tools. In other words, not only isthe morphology uniform, but so too are the wear patterns. OF course, one could argue that if there were morpho-‘Table 5. Mahalanobis distances and F statistics between group centroids. “Mahalanobis Distance Gls 12 8 cen2 3924 Cel3 1.9313 1.6514 Cell 4 6.9861 5.5380 1.7073 Cell $+ 2496 1.2673 2.7677 8.7899 Cell 9579 “1285 2.0546 5.4420 2.1778 | Cell 3.9034 2.3398 1.0549 9817 5.7599 2.0163 F statistics (Note: non-significant differences at 05 level underlined) Cells 12 3 4 Cell 5.048 | Cell3 4.795 4.654 Cell 86.017 137.452 4.758 Cell +866 5.222 4.598 35.666 Cell & 6,100 1.225 4.602 46.103 6.615 Cell 5.226 3.384 908 1.410 5.832 2.526 Nove “Indicates Cell | of the composite nother Africfcastem United Sates sample “Indicates Cell 2 of the composite norther Afriewestem United States sample “Indicates Cell 4 of the compte norhem Africaeatem United States sample logical changes between these samples, they would be ‘masked because the classification functions of the Lower Columbian tools included only WEARWD and DISTWD. ‘The obvious testis to examine the maximal solution, which included morphological variables. That solution reveals that, when the entire set of variables is considered, all 9 tools (100%) of the composite sample are correctly classified as Cell | tools. There is no significant difference in tool mor- Phology between the samples. The Cell 2 tools from the composite sample are identified both as Cell | tools and Cell 2 tools, once more on the basis of the Lower Columbia tool definitions. Even the Cell 4 tools of the composite sample are recognizably Cell 4 tools in Lower Columbian terms, The similarities and differences that do exist within and between the samples are even more clearly distinguished in Figures 3 and 4. It should be apparent that, as predicted, the distributions ofthe tools overlap to a significant degree, that the overlap is in terms of wear as well is all the more significant. The figures ‘assume’ morphological identity, and simply depict wear variability. The plots themselves, Journal of Field ArchaeologyiVol. 8, 1981 323 of course, are more intuitively satisfying than statistically informative, but one can calculate the actual distances in- volved between groups or cases, and compute the F statistic that determines the significance of those distances, *® These results are in Table 5. ‘The Table demonstrates thatthe visual overlap among the tools is not an illusion. There are no significant differences between the Cell 1 Lower Columbia centroid and the com- posite northem Africa/eastern United States Cell | centroid, Similarly, there are no significant differences between the respective Cell 2 and Cell 4 centroids. Indeed, there seems to be more overlap between tools separated by 10,000 years in time and 5000 mies in space, than there is between tools belonging to the same cultural tradition (e.g., the F value between Cells 2 and 4 of the Lower Columbia sample) ‘The evidence from the classification matrix (TABLE 4), the plots ofthe tools and the group centroids in discriminant space (FIGS. 3-4), the significant differences and overlap between the group centroids (FIG. 4, TABLE 5), and most especially the variables selected for inclusion in the dis- criminant functions (TABLE 3), provide a solid basis to argue that there are no significant morphological differences within this class of tools, and that all variability is func- tionally determined. This conclusion holds true whether the ‘observation is made on a local scale, or across widely sep- arated points in time and space, The only difference that does appear is related to the wear on the tools, This result, of course, is not expected. It simply suggests that these tools. ‘were used for different periods of time or intensity The relative absence of evidence of morphological var- ibility within the tool class brings us back to a point raised 50. The formula for calculating Mahalanobis distance, DY, between two _roupsis rom JV, MeKeon, "Canonical analysis: some relations between canonical comelation, factor analysis, diseriminant function analysis and ling theory." in Psychomeric Monographs 13 (1966) 9. The formula sas fllows, D where CV 3 Gir ~ Cin sanoniel variates ofthe two group cents the smaller of(g~1) and (p), where the numberof groups, and the numberof variables hand j ® the two groupe of interest. ‘The formula for calelating the commensurate F value is from Lachen- ‘uch, op cit. (n note 39) 10, a follows 2 utn—k-D atm wth where = of group 1 1 = nof group? = number of variables Dé = Mahalanobis distance between wo groups ith Degrees of Freedom = kand ny +m — kf ‘A. Olshn called these formulas my attention, and also provid advice ‘on runing the programs.324 A Study of Style and Function in a Class of Tools/Meltzer Overlap of different groups indicated by * 2 eiqeuen jeoluoueD, sa.a8 i Canonical variable 1 Figure 3. Distribution of the tools in discriminant space. A. Cell tools Lower Columbia sample (1 indiats group ‘cengoid).B. Cell 2 too, Lower Columbia simple (2). C, Cel 3 tools, Lower Columbia sample (3). D. Cel # tools, Lower Columbia sample (4) E. Cell | tools, composite norter Afrcseastem United States sample (5) F. Cell 2 tols, composite nontern Afieweastem United States sample (6). G, Cell 4 tots. composite norter Afrcueatem U earlier about defining a tool in terms ofits ‘shape’. | argued that while in some cases morphological variables can cluster in recognizable and archacologically meaningful patterns, this can only be ascertained after an examination of the ‘morphological patterning in the emtire class of tools. De- fining a tool by its ‘shape’ without considering the variation within the total tool class is problematic at best. In this instance, morphology has been used to define a particular kind of tool—endscraper’. Yet when the entire range of tools with similar wear patterns are examined, the mor- phological basis of the ‘endscraper’ is lost. The traditional definition of the tool is more an artifact of our conception Of the archaeological record than a reflection of the record itself. id Stats sample (7 Moreover, the basic morphology is solely functional, re- gardless of whether or not the tools exhibit manufacture beyond basic technological flake production. Modification is not even sufficient to change overall tool morphology from any other form of that same kind of tool. These results seem to affirm Wilmsen’s argument that flaking techniques are directed towards the production of tools with a minimal amount of modification.*! They also suggest that selection of a flake for use as a tool is a rather systematic process and/or that there is litle variation in flake production. What- ever the process, the product is a class of tools with re- ‘markably uniform morphology. SI, Wilmsen, op. cit in note 22) 66-67.Overlap of different groups indicated by * Z alqeuen jeo1uoueD, sre ss0t saat ete Journal of Field ArchaeologyiVol. 8, 1981 325 00 Canonical variable 1 Figue 4. Distibution of group centroids in discriminant spac. 1. Cell | centroid, Lower Columbia sample. 2. Cell? cenirog, Lower Columbia sample. 3, Cell 3 certo, Lower Columbia sample, Cell cenroi, Lower Colambia sample, 5.Cell | centroid, composite sample. 6. Cell 2 centroid, composite sample. 7. Cell 4 ceawoid, composite sample, ‘The homogeneity of morphology among the manufac- tured and non-manufactured tools, and between the tools from the Lower Columbia and from the composite northern Aftica/eastern United States sample, negates any argument for the presence ofa stylistic component in the morphology Of these tools, Conclusions ‘There has been an increased interest in recent years in the stylistic and functional attributes of archaeological ma- terials.* It has been generally recognized that: 52, Dunnell, op. ct (in note 7); Jlinek, op. cit in nae 7: Sacket, op it Gn note 7) to distinguish between these kinds of properties is ‘of primary concern in any study that attempts fo treat artifacts 25 a part of a functioning cultural system or to relate the par- ticular industries of two oF more prehistoric cultures in a mean- ingful way beyond spatil-temporal distribution." Yet, despite this truth, I would argue that there have been relatively few attempts by lithic analysts to go beyond the ‘obvious recognition of stylistic and functional variability in artifacts to an empirical analysis of variability in particular tool classes. There are, of course, exceptions to this gen- 53. Jelinek, op. cit (in note 7) 19-20,326A Study of Style and Function in a Class of Tools/Meltzer cralization.** Nonetheless, I think a significant chasm exists between the theoretical analyses of style and function and ‘mainstream lithic studies.*> The observation might also be ‘made of the gap between lithic studies and general archae- ological theory, although this gap is largely a function of the fragmented nature of the discipline and not because of divergent tendencies on the part of lithic analysts. At least all are agreed on the need to analyze stone tools, but if the reaction to Stiles” recent work’” is a reasonable barometer Of the situation, then there is little agreement on how we should go about the analysis. Not surprisingly, one area of agreement among some of the commentators on Stiles* is the need to grapple with issues of style and function ‘The definition of functional attributes themselves is not altogether a complex issue because most functional varia- bility is directly observable on the tool as wear. And, be- cause functional attributes are adaptive, they are, in some sense, redundant and thus predictable. Stylistic variability, fon the other hand, is a much more elusive element in tools. There are no predictable empirical manifestations of style, except to say that itis a historical phenomenon, In this analysis I have attempted to unravel style and function in one class of tools. I recognize the preliminary nature of the study, and that these issues and data require, ‘even demand, long-term analysis.” The examination must be done on a class-by-class basis, and it must include not ‘only morphology, as examined here, but also other avenues of variability, particularly those that might include stylistic elements. I chose to focus on morphology because, while central to most of our typologies, there appears to be very little understanding of the roles played by style and function within it So far as I can tell, given the variables I selected, the sample size, and the particular time/space coordinates of data, there is no stylistic component in the morphology of the tools examined, Considering the wide dispersion of the comparative samples, one would suppose that the conelu- sions reached here would hold true for the morphology of all 12133 tools. This proposition remains to be tested. It should be pointed out, however, that this is not to say that all 12133 tools, and endscrapers, are lacking in stylistic 4, See the work of Close, op cit (in note 9: idem, The Identification (of Spl in Lic Artifacts from Northeastern Africa (Cato 1970. 55, Jelinek, op. ct (in now 7) 27 56. R.C. Dunnell, “Amercanst Archaeology: the 1979 Contribution, AJA 84 (1980) 463-478; M. B. Schifer, "The Place of Lithc Use-Wear Studies in Behavioral Archaeology” in Lthic Use-Wear Analysis, B Hayden, ed. (New York 1979) 15-25, 57. Ses, op. cit. (in note 6), ‘58. See comments by Ammerman, Benéer, Bordes, and Jelinek in iid 59, Close, op. cit (in note $4) 258. attributes. For example, examining retouch might reveal a stylistic component.** This possiblity, to0, remains to be tested. ‘A question was raised, a number of years ago, regarding “why itis that endscrapers as a class exhibit relatively litle variability either spatially or temporally compared to pro- jectile points or some other class of items?" This question is part of a larger problem: whether our stone-tool typologies ‘measure style or function. ‘As Ihave argued here, “endscraper’ morphology is simply a reflection of functional variability, and thus it will be relatively uniform (ahistorical) through time and space. It should be apparent that the use of these tools as temporal ‘markers, a common practice in many chronologies, is mis- leading. These tools indicate function, and only ina limited sense can function tell time. 60. Close, personal communication, May 1980; Jelinek, op, ct (in note 7) 32 makes a similar suggestion 61. LR Binford, 4m Archaeological Perspective (New York 1972) 454 (62. Binford, op. cit. in note 7) 28. (63. This research as benefited from a great deal of helpful advice and titi from frends and colleagues. Fr this version in particular, I would ‘specilly lke to acknowledge the aid of Drs. R. C. Dumell, D. K. Gayson, B. D. Stafford, and R, Stafford, Suzanne Siegel, as always, made merlless comments on everything from my grammar to my late night hallucinations about endsrapes David J. Meltzer is @ graduate student at the University ‘of Washington. His interests include Early Man in North America, the history of archaeology, archaeological ‘method and theory. and evolutionary theory. Recent research, beside that reported here, has focused on the 19th century development of Early Man studies, and on Late and Postglacial environments in eastern North America.