JOURNAL OF ANTHROPOLOGICAL ARCHAEOLOGY 16, 301–333 (1997) ARTICLE NO. AA970314
Population Structure, Cultural Transmission, and Frequency Seriation
Carl P. Lipo
Department of Anthropology, Box 353100, University of Washington, Seattle, WA 98195-3100
[email protected] Mark E. Madsen
Emergent Media, Inc., 1809 Seventh Ave. E., Suite 908, Seattle, WA 98101
[email protected] Robert C. Dunnell
Department of Anthropology, Box 353100, University of Washington, Seattle, WA 98195-3100
[email protected] and Tim Hunt
Emergent Media, Inc., 1809 Seventh Ave. E., Suite 908, Seattle, WA 98101
Received June 14, 1995; revision received September 30, 1997; accepted October 2, 1997
The task of physics is not to answer a set of fixed been augmented by ethnographic lore and questions about nature, . . . We do not know in anthropological theory. The Americanist ar- advance what are the right questions to ask, and chaeological literature, consequently, testi- we often do not find out until we are close to an fies to a long flirtation with the definition of answer. ‘‘whole cultural’’ units comparable to ‘‘cul- Weinberg (1997:215) ture’’ or ‘‘society’’ as used by sociocultural anthropologists, units themselves not far re- It is not that sociologists are studying the wrong moved from their vernacular counterparts. things, but rather that they are studying them in the wrong ways. . . . the major reason for this ap- Variations of these units include ‘‘ethnic pears to be the way in which sociologists have cho- groupings’’ (e.g., Holmes 1903; Cordell and sen to conceptualize the phenomena of interest to Yannie 1991), ‘‘cultures’’ (e.g., Rouse 1939, them. 1955), ‘‘phases’’ (e.g., Chapman 1989; Krause Willer and Webster (1970: 748) 1977; Lehmer 1966, 1971; McKern 1939; Phil- lips 1970; Phillips and Willey 1953; Williams INTRODUCTION 1954, 1980; Willey and Phillips 1955,1958), ‘‘provinces’’ (e.g., Cordell and Plog 1979; Rather than being theory-driven, archaeo- Plog 1979, 1983), and ‘‘polities’’ (e.g., Ham- logical inquiry has been modeled, albeit of- mond 1972; King and Freer 1995; Peregrine ten unconsciously, from our own social ex- 1991, 1992, 1995; Upham 1982, 1983; Upham periences. In the United States, this basis has et al. 1981; Upham and Plog 1986). While
301 0278-4165/97 $25.00 Copyright ᭧ 1997 by Academic Press All rights of reproduction in any form reserved. AID JAA 0314 / ai06$$$$21 01-05-98 08:02:32 jaaal AP: JAA 302 LIPO ET AL. anthropological interpretations dominate which contrasted starkly with the essential- the use of these units, all are defined by and ist position of Griffin and Phillips. This posi- rest upon similarities and differences be- tion led Ford to break with his co-authors tween archaeological assemblages. What all and analyze spatial variability quantita- of these formulations have in common is an tively in the lower Mississippi valley data in attempt to capture patterns of interaction. a separate publication (1952).1 Ford capitu- The Midwestern Taxonomic Method lated to his co-authors on the matter of time (hereafter MTM), which created a set of five and broke the continuous sequence of events scaled units (component, focus, aspect, represented by seriation into a series of chro- phase, pattern, and base) on the basis of phe- nological periods compatible with an essen- netic similarity (e.g., Sokal and Sneath 1963) tialist view. The combination of periods and between components (more or less homoge- local areas supplied the archetype for the neous assemblages) (e.g., Dunnell 1971; phase (Phillips and Willey 1953; Willey and Krause 1977; Trigger 1989), was an early Phillips 1955, 1958; compare PFG and Phil- attempt to replace ethnographic-based lips [1970] for a concrete example). cultural units with archaeological ones Williams (1954) was first to apply the Wil- (McKern 1939). Relations between units ley and Phillips system anywhere and did were ahistorical in consequence of em- so in the Mississippi valley. Phillips’ study ploying an ad hoc set of set of ‘‘traits’’ to (1970) has had the greatest impact on the assess similarity and were so recognized by area because he defined phases for the entire the architect (McKern 1939). As chronologi- chronological sequence over the whole of cal data became available, the MTM was re- the Mississippi river alluvial valley. Given placed by the nonhierarchical and decidedly this history it is not surprising that ‘‘phase’’ historical system of Phillips and Willey. Of has been the dominant whole cultural unit the units proposed, Phillips and Willey’s used to describe the archaeology of the cen- phase is the most widely employed cultural tral Mississippi river valley (e.g., House unit with both spatial and temporal compo- 1991, 1993, 1995; D. Morse 1973, 1982, 1989, nents (Chapman 1989; Phillips 1970; Phillips 1990; Morse and Morse 1983, 1996; P. Morse and Willey 1953; Willey and Phillips 1955, 1981, 1990; Smith 1990). 1958; Williams 1954). Although rooted strictly in measures of The phases of the central Mississippi river artifact similarity, archaeologists quickly valley have their roots in the Lower Missis- gave phases ethnographic meanings. Even sippi River Valley Survey conducted by though phases, for example, are explicitly Phillips, Ford, and Griffin (1951; hereafter, defined as ‘‘an archaeological unit pos- PFG) in the 1940s. The analytic focus of the sessing traits sufficiently characteristic to PFG study is a series of seriations con- distinguish it from all other units similarly structed by Ford. Similarity was assessed by conceived, whether of the same or other cul- a set of historical types based on types ex- tures or civilizations, spatially limited to or- plicitly created by Ford (1936) to measure der of magnitude of a locality or region and time. Because spatial variation in frequen- chronologically limited to a relatively brief cies of historical types was sufficiently large period of time’’ (Willey and Phillips that no single seriation could be constructed 1958:22). Although it has been routinely sug- for the entire survey area, Ford had to divide gested that ‘‘the equivalent of phase . . . the Mississippi river alluvial valley into a ought to be ‘society’’’ (Willey and Phillips series of ‘‘local areas’’ (St. Francis, Memphis, 1958:49, italics ours; Rouse [1955] expresses Upper Sunflower, Lower Arkansas, Lower Yazoo). These local areas were analytic con- 1 This is why, one suspects, that Ford’s types in PFG trivances, not archaeological discoveries (cf. have become the standard (O’Brien 1996) instead of Ford 1952). Ford took a materialist approach Phillips’ (1970) more detailed types published later.
AID JAA 0314 / ai06$$$$21 01-05-98 08:02:32 jaaal AP: JAA CULTURAL TRANSMISSION 303 a similar view though with different termi- ceramics frequencies used to place assem- nology), early workers tended to be more blages particular phases show that phases cautious about the culture/phase equiva- are not groups (i.e., the members of one lence (e.g., Abbot 1972; McKern 1939; Willey phase are not more similar to members of and Phillips 1953) and realized that such the same group than they are to members equivalencies were accidental rather than of others). structural. Pressure to be ‘‘anthropological’’ While particular formulations may be de- has tended to overwhelm such wisdom in fective because of empirical mistakes (Fox many areas, including the central Missis- 1992), a good deal of the problem arises sippi river valley. Here, many archaeologists through the grouping legacy inherited from treat archaeological phases as if they are the the MTM. Indeed, outside the Mississippi material manifestations of ethnic, political, valley many phases are just renamed foci. or linguistic units. Recent central Mississippi Through Ford’s work (1935a, 1935b, 1936a, River valley research has often focused on 1936b, 1938) the Mississippi alluvial valley relating these phases to historical groups already had a budding chronology by the (e.g., Brain 1978, 1985; Hudson 1985; Morse mid-1930s and thus the ahistorical focus and Morse 1983, 1990, 1996; Phillips 1970; never opened a foothold in this region Phillips et al. 1951; Rouse 1965; Phillips and (Dunnell 1996). To the extent that phases are Willey 1953; cf. Abbot 1972). Not surpris- actually groups (instead of classes) the gen- ingly, there is little discussion regarding eration of new data (ceramic assemblages) their definition (e.g., Eighmy and LaBelle requires the reassessment of similarity be- 1996:56). Indeed, Phillips, one of the main tween all assemblages, not just the new ones. proponents of the phase concept, candidly This was never done; ad hoc extensional admits that the procedures for constructing definition was used to assign new data to phases are ‘‘regrettably non-objective’’ old names (Dunnell 1971). Thus the (1970: 523). ‘‘phases’’ of the Mississippi alluvial valley Still more recently, phase formulations quickly lost their coherence and rationale, and interpretations of central Mississippi producing the mess documented by Fox and river valley archaeology based on them O’Brien. The degree to which such units have been criticized as too simplistic. It has ‘‘worked’’ (i.e., displayed time/space conti- been argued that neither ethnic, political, guity) is a function of the units used in as- or linguistic units have straightforward sessing similarity (i.e., ceramic types). So analogs in artifact distributions (Fox 1992, long as stylistic (i.e., historical or neutral 1998; House 1991; Mainfort 1995; O’Brien [Dunnell 1981]) traits dominated in the as- and Fox 1994). Mainfort (1995), for exam- sessment of similarity, phases would neces- ple, has proposed that while ‘‘some pre- sarily display time/space contiguity. Be- viously-defined late period phases repre- cause phases create boundaries regardless of sent relatively valid units (i.e., statistically the structure of the archaeological record, reproducible), while others are much less phases, their locations and distributions, are robust, and some simply do not exist.’’ inappropriate descriptions of the archaeo- Other reanalyses suggest that phases may logical record when our goal is to study cul- have very little empirical basis. Fox (1998; tural interaction in the past. We must de- also 1992; O’Brien and Fox 1994) have velop new means to describe interaction be- demonstrated that phases such as those tween populations in space and time. created by Williams (1954) have no defin- In their early formulation, cultural histori- ing characteristics and cannot be consid- cal methods that used homologous similar- ered classes (i.e., there are no necessary ity were materialist in perspective and thus and sufficient conditions for membership). are a suitable starting point (e.g., Lyman, et At the same time, cluster analysis of the al. 1997). In particular, seriation measured
AID JAA 0314 / ai06$$$$21 01-05-98 08:02:32 jaaal AP: JAA 304 LIPO ET AL. continuous variation (Dunnell 1986; Teltser ian thought, we simply assume the general 1995). In the Mississippi valley, Ford’s work applicability of scientific evolution on the (e.g., 1935a, 1935b, 1936a, 1936b, 1938), docu- basis of arguments made decades ago (e.g., ments the continuous evolution of neutral Blute 1979; Dunnell 1978; 1980; Rindos 1980). traits without resort to periodization Consequently, we sketch here only the es- schemes (e.g., foci or phases). Ford’s and sential concepts that extend the theory from most subsequent worker’s, interests in seria- its nonhuman, noncultural origins (e.g., Dar- tion were limited to treating it as an empiri- win 1859). Evolutionary theory explains the cal dating method. There was no theoretical differential persistence of traits in living interest in seriation, i.e., why did seriations forms, i.e., change. Evolutionary theory is deliver chronological information? Spatial the only scientific theory that explains variation in class frequencies were usually change (why rather than how), so our com- seen as distortions, as noise confounding the mitment to evolution arises from a commit- chronological information. Now that prog- ment to an empirical epistemological stan- ress has been made in understanding the dard, nothing more. theoretical basis of seriation in evolutionary The principal mechanism explaining dif- theory (Dunnell 1982; Neiman 1992; Teltser ferential persistence of transmitted variation 1995), the ‘‘noise’’ arising from differences is natural selection. For selection to be opera- in space becomes a potential source of infor- tive, not only must variants be transmitted, mation on the transmission of homologous but at least some of the variants must inter- traits through space just as its use as a dating act with the environment and do so differen- method capitalized on the transmission of tially (i.e., result in difference in fitness.). homologous traits through time. Such variants are commonly referred to as In this paper, we review the linkage be- functional or ‘‘adaptive,’’ though the latter tween interaction, measures of homologous term is fraught with undesirable connota- similarity, space and time as a prelude to tions in the human context. It is with this demonstrating the use of neutral traits to argument in mind that many writers talk map patterns of interaction in the archaeo- about the use of evolutionary theory as tak- logical record. Because the Lower Missis- ing a ‘‘selectionist’’ approach, but it is en- sippi Valley Survey data constitute the his- tirely inappropriate, however inadvertent, torical basis for whole cultural units in use to make such an equation. For decades biol- in this region and because these data were ogists (Crow and Kimura 1970; Kimura 1977, collected (at least roughly) with the intent 1983; King and Jukes 1969) have realized of studying homologous similarity, we have that not all variation results in differential in- reanalyzed PFG’s data for their St. Francis teraction with the environment, that viewed and Memphis local areas. We use a modified from a selectionist perspective, some varia- frequency seriation method designed to ex- tion is neutral. That does not mean, again as amine the effects of space on the structure some writers supposed, such variation can- of homologous similarity. Our findings not be explained by evolution, only that it demonstrate the possibility of understand- is not explained by selection. Such variation ing interaction history using methods like is explained by transmission processes alone deterministic frequency seriation, thus or in combination with sampling (e.g., allowing us to create group-level units of Gulick 1872, 1905; Wright 1931, 1932, 1940, social interaction. 1948, 1949), processes that are Markovian in nature (Dunnell 1978, 1981; Gould et al. EVOLUTION, LINEAGES, AND 1977). CULTURAL TRANSMISSION The extension of evolutionary theory to Although some social scientists remain as archaeological phenomena required two ad- the last bastion of the pre-and non-Darwin- ditional steps. The first, beginning in the
AID JAA 0314 / ai06$$$$21 01-05-98 08:02:32 jaaal AP: JAA CULTURAL TRANSMISSION 305
1950s, was the recognition that an organ- confirmed and extended greatly by the work ism’s phenotype was not limited by its skin, of Neiman (1992, 1995). a view fostered by the museum rather than Two other components are critical. field study approach taken by biologists Dunnell (1970) analyzed Ford’s (Phillips et prior to this time. Behavior, just as much al. 1951: 219–223) assumptions for the use as bones, constituted phenotypic elements. of seriation, all of which were phrased as Indeed, it is even hard to imagine evolution conditions that had to be met by the assem- of morphology without behavior—giraffes blages to be seriated. Dunnell showed that could never have acquired long necks if they Ford’s ‘‘local area’’ criterion, that all of the did not try to eat leaves (Wcislo 1989). assemblages had to come from the same lo- The second step, the recognition that cul- cal area was incorrect. The problem that ture constituted a second (in addition to Ford was attempting to solve with the local genes) mechanism for trait transmission, area notion was an exclusion of space (as was facilitated by the first change inasmuch distance between assemblages) from influ- as most morphological traits are transmitted encing assemblage placement in a seriation. genetically most of the time while many be- Drawing boundaries around local areas was havioral traits are transmitted culturally, not an appropriate solution because culture trait transmission has, necessarily, both spa- even in animals (Lynch 1996; Payne 1996). 2 Given the dominance of culture in the gener- tial and temporal components. While one ation of the human phenotype, it is doubly set of stylistic classes applied to a given area might not detect spatial variation, increasing embarrassing that biologists (e.g., Bonner the amount of space or using another more 1980) were the first to work out these rela- detailed set of classes would include spatial tions. An additional mechanism for trait variability even though each analysis used transmission does not imply a need for addi- the same set of potsherds. Seriation was not tional selective mechanisms as some (e.g., conceived as a theory to explain neutral vari- Boone and Smith 1998; Boyd and Richerson ations by culture historians like Ford, but as 1985; Cavalli-Sforza and Feldman 1981; Dur- a dating method. As a theory, it was, and is, ham 1976, 1979, 1990, 1991, 1992; Graves- dynamically insufficient (Lewontin 1974: 6). Brown 1996; Soltis, et al. 1995) have as- The second element must also be under- sumed. Indeed, positing mechanisms like stood from this historical perspective. Ford ‘‘cultural selection’’ seems to be generated (Phillips et al. 1951) also saw that assem- by a confusion of folk explanation (our cul- blages to be seriated had to be drawn from ture’s ‘‘explanation’’ for change) with scien- the same tradition. This was correctly attrib- tific causation. uted to a condition that the assemblages had Since the connection between neutral vari- to meet but the reasons this should be so ation and the archaeological notion of type are significant. As Dunnell (1970) noted, the or historical types was made (Dunnell 1978, only way you could determine whether a 1980, 1982), it has become possible to explain group of assemblages did constitute a tradi- why seriation worked as a dating method tion or not was whether or not they could and why spatial variation was a confound- be seriated. What Ford meant by ‘‘tradition’’ ing factor in that environment of use (cf. Ford 1954) was that the assemblages be- (Dunnell 1982). If stylistic variation were ing seriated had to be drawn from assem- neutral variation, then the temporal distri- blages that were historically related. In evo- bution of stylistic (historical) types would be lutionary terms, this requirement meant that unimodal because of the probabilistic struc- ture of transmission (Dunnell 1982). This 2 And probably why ‘‘local area’’ was never defined supposition, originally based on Gould et and why no procedures for defining such areas were al.’s (1977) observation in biology, has been given.
AID JAA 0314 / ai06$$$$21 01-05-98 08:02:32 jaaal AP: JAA 306 LIPO ET AL. they assemblages must be samples of the sumptions, allowing individuals to interact same lineage. over great distances, and accumulating Thus, why seriation works (and does not many time slices of such transmission work) is explained by evolutionary theory events. Taking a spatial transect through the and the appropriateness of seriation as a tool space and time ‘‘cube’’ of results is illus- to explore interaction is clearly warranted. trated schematically in Fig. 1. Time is dis- Appropriately expanded, seriation can be played on the vertical axis, with the bottom used just as easily to delineate lineages from of the figure being the start of the simula- which assemblages are drawn as it has been tion, and space along the horizontal axis. to arrange those assemblages in chronologi- The ‘‘threads’’ traveling from bottom to top cal order. That expansion is our task here. are individuals, each displaying the attri- To proceed much further, it is useful to bute currently held along one of several start with Neiman’s (1995) recent formula- dimensions as a letter value. Intersections tion of neutral trait transmission because it between threads are, when viewed at a mo- rests on simple, credible assumptions. In ment in time, interactions between individu- Neiman’s model, all individuals in the pop- als as they move around the space and en- ulation are considered to have equal proba- counter one another. The result of an interac- bility of interaction with each other and the tion can be a switch in traits for both frequencies of traits should behave stochas- individuals, one individual copies the oth- tically due to drift. If this model is translated er’s trait, or neither individual changing into a spatial context, it would mean that their original phenotype. Transmission space imposed no energy costs and that all events are depicted, then, by a symbol individuals would, over the long run, inter- change in a thread exiting one of these inter- act with each other. In panmictic popula- action ‘‘intersections.’’ Using Neiman’s as- tions, we would expect that overall similar- sumptions of global interaction (Fig. 1), we ity between samples of individuals from ran a simulation and tabulated the frequen- across the space would be similar at any cies of three variants for six durations of point in time. Such a model is clearly unreal- time, as indicated between the fine hori- istic, but as Neiman (1990, 1995: Fig. 2) zontal lines. As seen at the right side of the showed, when viewed as relative frequen- diagram, with no spatial biases on interac- cies, transmitted attributes form unimodal tion, the frequencies form unimodal series. curves of the kind seen in frequency seria- In our experiments, we relaxed the as- tion. As Dunnell (1970) notes and as we sumption of panmixis in several ways— would expect, this behavior is due to the both homogeneously (in the form of a sim- absence of a spatial component to transmis- ple distance-decay function [Hagerstrand sion between individuals in the population. 1952; Renfrew 1977]) and inhomogeneously This effect can be most clearly seen by (in the form of clumped distributions of in- simulating interaction using Neiman’s as- dividuals in addition to transmission entail-
FIG. 1. Transmission among individuals with different phenotypic traits, labeled A, B, and C, depicted within a uniform population with no spatial restriction on interaction frequency. When individuals encounter each other there is a probability, not predicted by the form of the trait(s), that the individual takes the trait of the other individual. Although in this simulation the frequency of taking a trait was set at 50%, this frequency can also vary in relation to the frequency of the traits (i.e., density-dependent). Individual letters (A, B, C) represent discard events that are aggregated over a period of time into depositional events (‘‘assemblages’’ 1–6). Trait frequencies are tabulated for each assemblage along the right-hand side. A seriation using these trait frequencies is located at the bottom of the figure. Note that in real cases errors in sampling will yield numbers that do not match the model. Consequently, error terms must be used.
AID JAA 0314 / ai06$$$$21 01-05-98 08:02:32 jaaal AP: JAA CULTURAL TRANSMISSION 307
AID JAA 0314 / ai06$$0314 01-05-98 08:02:32 jaaal AP: JAA 308 LIPO ET AL. ing a distance ‘‘cost’’). Figure 2 depicts a ho- mission model with the additions of limited mogenous population with a strong bias to- interaction radii and inhomogeneous popu- wards interaction close to ‘‘home,’’ along lation distribution yields the kind of behav- with an attribute (A) diffusing through the ior we see in real archaeological datasets: population from left to right over an appre- areas within which data form unimodal ciable period of time. Again, frequencies are curves, and larger regions of space over sampled at five points in time marked by which unimodality breaks down because horizontal lines. The global frequencies are several locally interactive groups have been shown at the bottom, while frequencies in artificially lumped together. It is for this rea- three regions of the transect are shown son that Ford had to insert his ‘‘local area’’ across the top. It is notable that the regional criterion for seriation. results show different frequency histories, and different histories of class richness. Importance of Neutrality While all are unimodal (or nearly so, given the effects of sampling), they display peaks An important point not easily captured in and valleys offset from one another, repre- simulations or mathematical models of senting the unique history of transmission transmission is the effects of selection on the in a regional population. eventual distribution of traits. Most pheno- When populations interact differentially typic attributes have the potential of in- across an area and tend to interact within a forming on interaction (Beck 1995), but the short radius relative to the entire region, as degree to which any particular trait is useful in Fig. 3, real differences in the rates of inter- for studying interaction at a particular spa- action between areas of the space and in trait tial and temporal scale will vary depending frequencies appear at any moment in time. upon the strength of evolutionary processes There is a constant density of individuals (other than transmission) acting upon the in each of the panels, but they vary in the trait. If selection or other evolutionary mech- frequency of encounter with others. Even anisms do not act to change trait frequencies, though individuals are uniformly distrib- the geographic distribution of the trait uted throughout the space, variants will dif- through time is determined by the spatial fuse more quickly through the left and right structure of individuals in the environment panels because the interaction frequency is and the frequency with which individuals higher, and will take longer to spread encounter each other (which may vary through and across the central panel. The among communities). In other words, under same results are seen in Fig. 4 where individ- null conditions, the frequency of interaction uals are unevenly distributed in space and is equivalent to the spatial structure of the interaction occurs in frequently between the population. Along a dimension of pheno- two ‘‘clusters.’’ Despite the similarity of pat- typic variability, attributes that do not affect terns of interaction, these two scenarios are fitness, and thus whose frequencies are un- archaeologically distinguishable by studies affected by selection, are neutral or stylistic of the spatial distribution of assemblages. (Crow and Kimura 1970; Dunnell 1978; Ki- Simulations of Neiman’s (1990, 1995) trans- mura and Ohta 1971; King and Jukes 1969).
FIG. 2. Transmission among individuals within a uniform population where the probability of interac- tion is strongly affected by distance. Individuals interact only with local neighbors. In this simulation, trait B moves from left to right through the population. At the bottom of the figure, a seriation that considers the entire space does not meet unimodal expectation of the model. Seriations calculated at the top for spatially limited samples indicated by the vertical lines, however, meet the expectations of the model and form perfect seriations. As mentioned in Fig. 1, real cases require the use of error terms due to the effect of sampling.
AID JAA 0314 / ai06$$$$21 01-05-98 08:02:32 jaaal AP: JAA CULTURAL TRANSMISSION 309
AID JAA 0314 / ai06$$0314 01-05-98 08:02:32 jaaal AP: JAA 310 LIPO ET AL.
Attributes that make either a positive or a fact that frequencies of stylistic types in as- negative contribution to individual fitness semblages are more than simply the results (functional attributes) will be structured ge- of transmission in past populations, but de- ographically not only by the distribution rive their particular character from addi- and connectivity of individuals from lineage tional sources: duration, classification, and to lineage, but by the characteristics of the sampling (Dunnell 1970, 1981; Ford 1949; selective environment. The effects of a Phillips et al. 1951). As Ford was cognizant change in selective environment on stylistic of the duration problem, conscious efforts and functional attributes is depicted in Fig. were made in the field to eliminate long 5. In the right-hand panel the distribution of duration occupations (Phillips, et al. 1951: two neutral traits with respect to an environ- 219). Minimizing the effect of differences in mental gradient (solid line) is random; these duration was accomplished by eliminating traits have distributions that are capable of mixed assemblages or assemblages that informing on interaction between transmit- were anomalously rich compared to other ters. In the left-hand panel the distribution assemblages that were to be seriated (cf. of alternatives is conditioned strongly by the Dunnell 1970). Ford (1949, 1952; Phillips et environmental gradient. If we were to use al. 1951) accounted for variability in the these functional attributes, we might hastily frequencies of stylistic types caused by conclude that there is little interaction be- classification by noting that the unimodal tween the two regions. Thus, in contrast to character of historical classes when viewed simulations, in real archaeological cases we in a seriation was derived from filtering a must be careful which traits we use to delimit continuous stream of transmitted informa- lineages. tion through a classification so that we could count variants. Ford’s explanation, Heritability and the Seriation Method however, is incomplete. Unimodality is created both by the stochastic or Marko- During our simulation experiments, we vian character of neutral trait transmis- noted that panmictic populations whose sion, and by confronting the frequencies of members were free to interact equally over those traits with a set of classes treated as the entire space produced nearly perfect se- a closed array. If there are many classes riations, whereas any restrictions on the ra- forming the closed array, as in a classifica- dius of interaction would destroy our ability tion with many dimensions or attribute to include the entire test population in one classes per dimension, fine temporal and seriation. Frequencies of traits in one area of spatial discriminations can be made the simulated population were out of ‘‘sync’’ (Dunnell 1971). If there are few classes with frequencies in other regions, pre- forming the closed array, the resulting dis- venting any single ordering from yielding a criminations will be coarse in scale. This perfect solution. In the absence of indepen- elementary fact suggests that no single se- dent temporal control over a set of assem- riation can claim to yield groups of assem- blages, we suggest that the failure of a set of blages that comprise a lineage. Rather, se- assemblages to seriate (rather than success) riation can be used to map quantitative might be exploited to test hypotheses about changes in the history of regional and local the history of interaction in a region. interaction. The method we use involves The only difficulty with this notion is the creating seriations with a nested set of
FIG. 3. Transmission depicted within a population varying in interaction frequency across space. Two ‘‘populations’’ of individuals are apparent in this simulation separated by an area of less frequent interaction. A graph of the frequency of interaction is shown at the bottom.
AID JAA 0314 / ai06$$$$21 01-05-98 08:02:32 jaaal AP: JAA CULTURAL TRANSMISSION 311
AID JAA 0314 / ai06$$0314 01-05-98 08:02:32 jaaal AP: JAA 312 LIPO ET AL. classes at variable levels in order to see assemblages into a particular seriation that fit how interaction changes with level. If in- the unimodal model formed by assemblages teraction is conditioned solely by distance already in the seriation, with degrees of free- (assuming the types used are strictly stylis- dom in assignment dictated by sampling tic) then varying classification in level (Dunnell 1970, 1981, in press). should produce differences in how assem- All assemblages are samples of a much blages are group that correspond to larger deposit, and because of this fact com- changes in spatial ‘‘boundaries.’’ If, on the parisons between assemblages must be other hand, interactions are structured by treated as hypotheses. Error terms are calcu- barriers (physical or social), spatial bound- lated on the basis of sample size. Two ap- aries will be reflected in inflections, seem- proaches to sampling error were used in this ingly unrelated to distance at some classi- analysis. First, because seriation is sensitive to fication levels while related to distance at the number of classes present in any assem- others (Fig. 6). In the remainder of the pa- blage (richness), and it is well known that in per, we explore this notion using data from ‘‘small’’ assemblages richness may be a func- the Lower Mississippi Valley Survey. tion of sample size (Grayson 1989), we exam- ined the effects of assemblage size on the ANALYTIC METHODS AND number of classes seen in Lower Mississippi TECHNIQUES Valley Survey (PFG) assemblages. We wanted to determine whether there were assemblages The common approach to seriation today that were too small to be reasonably included is a probabilistic technique. In a probabilistic in seriations with the remainder of the PFG technique some form of matrix ordering algo- seriations (i.e., differentiate ‘‘absences’’ due to rithm is used to shuffle assemblages until the sample size rather than population differ- best fit to a unimodal curve is achieved, and ences). Figure 7 shows plots of assemblage the resulting order accepted as chronological richness versus assemblage size for all assem- (e.g., Johnson and Nelson 1990; Love 1993; blages in our study area. This figure demon- Marquardt 1973). The probabilistic approach strates that while there is the familiar trend of is inadequate, as it offers no means of differ- richness increasing as sample size increases, entiating violations of seriation principles there is also considerable scatter. Usual prac- from sampling and measurement error. Prob- tice would be to accept any assemblages that abilistic methods uniformly reduce the fre- fell onto the asymptotic portion of the graph quencies present in data sets to a set of similar- as large enough to have reasonable samples ity measures, which are scaled according to of all classes, but this assumes that all differ- one or another algorithm. Probabilistic tech- ences in richness are due to the effects of sam- niques always produce an order, even with ple size, and excludes possible variation in data of poor quality because information con- richness through time or across space. We did cerning sampling error is not used. We are not want to exclude these possibilities by con- particular sensitive to this issue because we cluding that all assemblages with smaller sam- are interested in using seriation not as a dating ple sizes and low richness were poor samples, technique but as a means of explaining fre- so we used bootstrap methods to examine quency variation. We reject the probabilistic richness and sample size for each assemblage. approach. Consequently, we use a determinis- Bootstrapping is a method for evaluating tic approach to seriation, accepting only those the degree to which a test statistic calculated
FIG. 4. Transmission depicted within a population of varying density across space. Note that the frequency of interaction is the same as in Fig. 5. Archaeologically these scenarios (shown in Figs. 3 and 4) can be distinguished through the study of spatial distributions of assemblages.
AID JAA 0314 / ai06$$$$21 01-05-98 08:02:32 jaaal AP: JAA CULTURAL TRANSMISSION 313
AID JAA 0314 / ai06$$0314 01-05-98 08:02:32 jaaal AP: JAA 314 LIPO ET AL.
FIG. 5. The effects of a change in selective environment on stylistic and functional attributes. In the right-hand panel, the distribution of two neutral traits with respect to an environmental gradient (solid line) is random; these traits have distributions that are capable of informing on interaction between transmitters. In the left-hand panel, the distribution of alternatives is conditioned strongly by the environ- mental gradient. Failure to distinguish functional attributes from neutral attributes in the case of the left hand panel indicates that there is there is little interaction between the two regions. Care must be given when choosing the traits to be used to delimit lineages.
for a particular sample is an estimator of the ences in sample evenness and hence require unknown, underlying population (Berry et greater sample sizes to confidently assess al. 1980, 1983; Efron 1982, 1983; Efron and richness. By plotting the mean and variances Tibshriani 1993; Fisher 1936; Good 1994; of the richness distributions, we were able Mooney and Duval 1993; Noreen 1989; Si- to evaluate the degree to which assemblages mon 1990; Simon and Bruce 1991). The had a sufficient sample size to confidently method relies on the creation of a bootstrap estimate the population richness. We di- distribution of values for the test statistic vided the assemblages into three groups created by repeatedly resampling from the (Fig. 8). The first group contained those as- original data. In our evaluation of richness semblages that had richness mean and vari- and sample size, we created a series of rich- ance values that reached an asymptote with ness distributions constructed of randomly the original sample richness prior to the final drawn resamples that we systematically in- sample size. The second group included as- creased in sample size. For example, with semblages that had mean resample rich- an assemblage of original sample size of 300 nesses or variances (but not both) that did we randomly drew 10 sherds with replace- not approach an asymptote as sample size ment from the original sample and calcu- was increased. Finally, the third group con- lated richness. This procedure was repeated sisted of assemblages that had neither as- 1000 times and a distribution of richness cal- ymptotic mean richness or variance but sug- culated. We then resampled 20 sherds from gested that increased sampling from the the original sample and calculated a richness archaeological record would result in sig- distribution. This process continued with in- nificant increases in the number of artifact creasing resample sizes until the original classes represented. sample size was reached. Significant Finally, when making each seriation at- changes in the richness distribution as sam- tempt, assemblages were added in an iterative ple size increased were caused by differ- fashion, and various orderings treated as a hy-
AID JAA 0314 / ai06$$$$21 01-05-98 08:02:32 jaaal AP: JAA CULTURAL TRANSMISSION 315
FIG. 6. Describing assemblages with a nested set of stylistic classes at variable classification levels permits the examination of how interaction changes with classification level. If interaction is conditioned solely by distance (assuming the types used are strictly stylistic) then varying classification in level corresponds to changes in spatial ‘‘boundaries.’’ Interactions that are structured by physical or social barriers produce spatial boundaries that are detected by inflections in the relationship between distance to classification level. In this hypothetical example, as the number of dimensions in the classification are decreased from four to one, the spatial scales of the resulting groups increase uniformly and are correlated with distance. Using a classification of two or three dimensions, however, consistently pro- duces two spatially coherent groups of assemblages. The disjunction in the measurement of these two assemblage sets cannot be explained by distance and, therefore, may (once sampling and differences in deposition are considered) represent interaction structured by physical or social barriers.
AID JAA 0314 / ai06$$0314 01-05-98 08:02:32 jaaal AP: JAA 316 LIPO ET AL.
FIG. 7. Relationship of richness and sample size for all PFG assemblages, St. Francis and Memphis area data. The diamond (ࡗ), square (), and triangle (᭡) symbols refer to bootstrap assessed types I, II, and II, respectively. pothesis test concerning fit to unimodality. (1945) method: a normal approximation to a Testing the fit involved pairwise comparison confidence interval for a binomial variable was of assemblage frequencies using confidence in- calculated using a type I error rate of 0.005 tervals for class frequencies calculated on the (modified from Cochran 1975:57). Assem- basis of sample size following Beals, et al. blages that fit the increasing or decreasing ex-
FIG. 8. Bootstrap richness and sample size relationship examples, St. Francis and Memphis area data. Assemblages have been split into three types: Type I, assemblages with richness means and variances that reach an asymptote before final sample size in bootstrap evaluations; Type II, assemblages with richness means or variances that reach an asymptote before final sample size in bootstrap evaluations, and; Type III, assemblages with richness means and variances that do not reach an asymptote before final sample size in bootstrap evaluations. Examining the distribution of the bootstrap assessed types in a standard richness and sample size plot (Fig. 7) demonstrates the limitations of these plots; Type I and Type II assemblages are mixed even at ‘‘large’’ sample sizes. Minimum sample sizes, therefore, must be determined for each sample and a ‘‘global’’ sample size may not necessarily be reliably set.
AID JAA 0314 / ai06$$$$21 01-05-98 08:02:32 jaaal AP: JAA CULTURAL TRANSMISSION 317
FIG. 9. Map of Lower Mississippi Valley Survey: Assemblages from the St. Francis and Memphis regions from Phillips et al. (1951). pectations of the unimodal curve model were into various groups while displaying confi- accepted into a given seriation group and as- dence intervals and testing hypotheses of fit to semblages that did not fit, within calculated the model.3 All of the seriation results pre- confidence limits, were rejected and placed in sented here were done using this tool. different groups. Obviously, this technique is not a mechanical algorithm in the spirit of mul- 3 tidimensional scaling but an iterative, operator The program used to perform iterative seriations was written by Tim Hunt as a Visual Basic front-end intensive technique closer to Ford’s (1962) ap- to Microsoft Excel v.5.0. A version of this program is proach. We have built a visual interface that available at our WWW site: http://www.emergent allows one to order and reorder assemblages media.com/archy.
AID JAA 0314 / ai06$$$$21 01-05-98 08:02:32 jaaal AP: JAA 318 LIPO ET AL.
SOCIAL ORGANIZATION IN THE the other hand, is partly a measure of the CENTRAL MISSISSIPPI VALLEY: amount of decoration present in an assem- PFG REVISITED blage. Since in general the level of invest- ment in style must be driven by selection During the course of Lower Mississippi (through differential cost), the spatial and Valley survey, PFG collected 346,099 sherds temporal distribution of plain ceramics will from 383 localities. Each of the sherds were be responsive to external conditions, rather tabulated according to culture-historical than being determined purely by the struc- types that passed the test of historical sig- ture of transmission networks. Second, be- nificance.4 Using these tabulated data, Ford cause decoration is not evenly distributed constructed five seriations for the study area across individual vessels, the abundance of in order to minimize the effects of geogra- plain sherds will also be related to the phy that were readily apparent in the latest amount of decoration on vessels, and thus time horizons (Phillips et al. 1951: 224). Fig- breakage will differentially affect the fre- ure 9 shows the survey area and the location quency of plain ceramics derived from dif- of sites included in these seriations and in ferent vessels. Third, variability in biases be- our reanalysis. Sites are numbered sequen- tween collectors also produces non-compa- tially on the basis of the 15 minute quadran- rable assemblages in the frequency of gle in which they are located. decorated sherds. We were fortunate to be able to examine We also excluded types that were present the original data for the Memphis and St. in only one assemblage since they do not Francis seriations.5 Coded on worksheets in inform on interaction frequency. In addition, Ford’s own hand, the original data are in- we eliminated excavated assemblages from valuable for modern analyses of these data, our analysis. These assemblages typically containing as they do the actual sherd have idiosyncratic compositions depending counts and assemblage sizes. The data in- upon the particular excavation location due cluded frequencies for both shell- and clay- to spatial autocorrelation effects arising tempered ceramics, usually interpreted as from related fragments (Dunnell and Mississippian and Woodland, respectively. Dancey 1983). Since archaeological deposits Woodland materials were not the focus of are rarely, if ever, spatially homogenous, fieldwork by Phillips and colleagues and even if described with stylistic (neutral) were recorded only when found underlying types (Dunnell 1981), frequencies calculated or mixed with the shell-tempered ceramics. from spatially limited samples will differ, Thus, the sample of Woodland materials is often greatly, from the frequencies in the as- problematic even given the actual sherd semblage as a whole. As a result, unless re- counts, and in this paper we restrict our fitted (e.g., Newell and Krieger 1949) they analysis to the shell-tempered ceramics. cannot be considered to be representative of We excluded plain ceramics from the tab- the entire deposit. Ceramics obtained from ulations (e.g., Neeley’s Ferry, Belltown the surface, on the other hand, have a greater Plain) because the frequency of plain ceram- chance to exhibit representative frequencies ics are biased in three ways. First, seriation due to post-depositional processes, such as measures relative frequency of historical plowing, that produce a sample of the entire types. The frequency of plain ceramics, on deposit for inspection by archaeologists. We began our analyses of the PFG data by examining the solutions produced by 4 See Krieger (1944), strictly speaking, PFG types are Ford for the St. Francis and Memphis areas not quite this consistent; Ford types approach this con- dition. using only assemblages collected from the 5 The late J. B. Griffin kindly supplied the data sheets surface and tabluating just the decorated, to R. C. Dunnell for which we are most grateful. shell-tempered types. These seriations are
AID JAA 0314 / ai06$$$$21 01-05-98 08:02:32 jaaal AP: JAA CULTURAL TRANSMISSION 319
shown in Figs. 10 and 11. Clearly, there are numerous departures from strict unimod- ality, some probably caused by sampling er- ror, others from the effects of space or as- semblage duration (Dunnell 1970). In the re- mainder of our analysis, we explore the degree to which these potential effects pro- duce deviations from the expected seriation model.
Seriation and Spatial Structure Instead of discarding assemblages on the basis of sample size, we broke the assem- blages into three groups based on the char- acter of the bootstrap analysis. In Fig. 8, we illustrate each of these groups with a repre- sentative bootstrap sampling graph. Conclu- sions based on the assemblages that meet both the mean and variance criteria thus should be relatively meaningful (all other things being equal), while conclusions based uncritically on all of the assemblages (in- cluding the assemblages that met neither the mean nor the variance criteria) should be treated with extreme caution. We began with all of PFG assemblages and seriated the assemblages into the largest groups possible without allowing violations of unimodality within calculated confidence intervals (a Å 0.05). If a seriation containing a group of assemblages displayed significant departures from unimodality, the group was broken into two groups each of which was unimodal. Obtaining seriation solutions in this way is an iterative process, involving series of hypothesis tests about particular or- derings, and we show only the resulting se- riation groups in Fig. 12. The groups shown in Fig. 12 are the largest possible to make within the restriction that departures from Ford’s seriation of the St. Francis area data using only assemblages collected from theunimodality surface and decorated, shell tempered ceramic do not occur. If one maps the assemblages comprising each group, a strong spatial pattern emerges FIG. 10. 6
types. (Fig. 13). The groups within which assem-
6 We can calculate the probability of randomly draw- ing a spatial cohesive set of assemblages by chance alone as the product of series of combinatorial probabil- ities where
AID JAA 0314 / ai06$$$$21 01-05-98 08:02:32 jaaal AP: JAA 320 LIPO ET AL. blages seriate without violations of unimod- ture-historical phases bear a partial relation- ality are spatially cohesive sets; no seriation ship to lineages. Some of the spatial cohesion sets are comprised of assemblages that are of a phase, in our view, comes from real scattered randomly across the region as one structure to the interaction history of popu- might expect if the PFG assemblages were a lations, with the remaining contribution to random sample of a single large interacting frequency variation coming from several population. Neither do many assemblages sources: sampling error (which must be con- seem to belong reasonably to more than one trolled), and the effects of viewing continu- group, with the exception Groups 2A and ous variation through the ‘‘filter’’ of a classi- 2B (Fig. 13). Assemblages that belong to fication. The ethnographic ‘‘groups’’ of the more than one seriation set will occur when- early historic period, to the extent they have ever these assemblages are ancestors to a basis in social interactions are just a fleet- other assemblages which have split into two ing configuration known to us by the virtue or more distinct lineages or are the product of historical accident. As investigative units of ‘‘hybridization,’’ i.e., coalescence of for- they lack any theoretical or empirical war- merly distinct lineages. The spatial groups rant. of assemblages, to a rough approximation, Returning to the seriation groups ob- also resemble the general shapes and posi- tained from using PFG classes with all of the tions of culture historical phases, as one available assemblages (Fig. 13), the question might expect. Parkin phase (P. Morse 1981, arises, are these groups representations of 1990) corresponds to Groups 2A and 2B, No- true discontinuities in interaction frequency, dena (D. Morse 1973, 1990) to Group 1, Walls and thus ‘‘lineages,’’ or are the groups phase (Smith 1990) to Group 5, Kent (House merely artifacts of sampling and other prob- 1991; Smith 1990) to Group 7, and Com- lems with the PFG data? What are the merce phase (Smith 1990) to Group 8. sources of variability that contribute to the That some correspondence to culture his- frequencies and spatial patterns seen in the torical phases exists is not surprising, since PFG data? We must be able to discount all culture historians were generally interested other sources of variability before we can in delineating populations related by homol- conclude securely that the groups seen in ogous similarity. On the basis of transmis- Fig. 13 might represent the structure of pop- sion theory as outlined above, however, cul- ulation interaction in the past. If transmis- sion through the region is basically continu- ous but variable across space, several factors could account for the clumped appearance 1 p Å . of assemblages that seriate well together. N! (N 0 n )! ∗ 1 The first and foremost of these is sampling n1!(N 0 n1)! n2!(N 0 (n1 / n2))! error. To assess the possibility that sampling (N 0 (n / n ))! 1 1 2 rrr error, and in particular changes in richness n1!(N 0 (n1 / n2 / n3))! due to sample size, are affecting the size, posi- n01 tion, and inclusiveness of seriated groups, we 0 ∑ N ni! performed iterative seriations with only those iÅ1 1 n assemblages marked as ‘‘good’’ in bootstrap n1!(N 0 ∑ ni)! iÅ1 analysis. These assemblages each have sample sizes at which no new types are found if addi- N Å total number of assemblages, ni Å number of as- tional samples are taken, and thus are reliable semblages in the ith group. Using the PFG assemblages in their frequency characteristics. The seriation that meet the mean and variance criteria, we can calcu- late that the probability of arriving at this spatially con- results are presented in Fig. 14 and the geo- tiguous pattern by chance alone is very small. (p Å 4.011 graphic distribution of the seriation groups in 1 10012). Fig. 15.
AID JAA 0314 / ai06$$$$21 01-05-98 08:02:32 jaaal AP: JAA CULTURAL TRANSMISSION 321 Ford’s seriation of the Memphis area data using only assemblages collected from the surface and decorated, shell tempered ceramic FIG. 11. types. It is clear, that the neither the seriations in Fig. 10 nor Fig. 11 meet the expectations of the model.
AID JAA 0314 / ai06$$0314 01-05-98 08:02:32 jaaal AP: JAA 322 LIPO ET AL.
What is notable about the clusters de- systematic differences in level can be picted in Fig. 15 is that the groups are re- achieved by simply collapsing dimensions markably similar to those presented in the of variability in an orthogonal manner. We map with all assemblages (Figure 13). Many were able to only approximate this ideal assemblages were dropped, however, due to with the PFG classification. Figure 16 depicts sample size problems, making exact com- the relationship between the original PFG parisons of relationships difficult. Walls types, along the top, and two successive lev- Phase (Group 5) sites, for example, show up els of collapsing based on ever-more inclu- as a cohesive group that seriate well to- sive categories of surface decoration attri- gether, although the relationships between bute classes. For each set of newly collapsed seriation groups and phases becomes less classes, frequencies for each assemblage clear in the area traditionally considered to were recalculated, and seriations were then be the Kent (Group 3) and southern Parkin done with the two collapsed sets of classes. Phase (Group 2) areas. In these areas, the The spatial distribution of seriations are de- small number of available assemblages may picted in Fig. 17. have artificially grouped assemblages from The effects of changing the level of classi- these separate seriation sets together. On the fication for the PFG assemblages in the di- other hand, the phases themselves might be rection of coarser classes (fewer dimen- simply arbitrary spatial and temporal slices sions) is to simply create larger groups of of an interacting population that is much assemblages, and in some cases allowing larger in size (e.g., covering all of 11-N, 12- coalescence of formerly separate assem- N, and 13-N, along with some of 13-O). Dis- blages into more inclusive groups. The tinguishing these possibilities, and under- groups formed, however, still retain a great standing the relationship of the groups deal of spatial contiguity, suggesting that mapped in Fig. 15 to the underlying interac- the possibility exists that the seriations per- tion history of the region, requires that we formed here may reflect structure in the in- examine how the groups and seriations teraction history of the region, and thus viewed thus far are artifacts of the classifi- may map the approximate location of lin- catory scheme used in Phillips et al. (1951). eages. Should this conclusion turn out to be true, the relationship between ‘‘filters’’ of Seriations at Multiple Scales successive sizes suggests that there are not clear social interaction boundaries in the To understand the effects that classifica- area but that interaction is widespread and tion, and in particular the number of dimen- continuous, and that the appearance of dis- sions, has on the size, position, and inclu- crete communities comes from slicing a siveness of seriated groups, it is necessary continuous range of interaction frequencies to perform seriation analyses using a hierar- using an arbitrary measuring unit for inter- chical series of classifications and, if possi- action. If social boundaries existed in the ble, multiple independent classifications patterns of interaction, these should be visi- (Dunnell 1970). Ideally, the classifications ble as discontinuties in the relationship be- used will be hybrids and rigorously dimen- tween the numbers of dimensions and sional in nature (Dunnell 1971), in order that number of groups as the level of classifica-
FIG. 12. Iterative deterministic seriation solution groups for all PFG assemblages. The eight solu- tions (1–8) are the largest seriation groups that can be formed and still statistically meet the expecta- tions of the model.
AID JAA 0314 / ai06$$$$22 01-05-98 08:02:32 jaaal AP: JAA CULTURAL TRANSMISSION 323
AID JAA 0314 / ai06$$0314 01-05-98 08:02:32 jaaal AP: JAA 324 LIPO ET AL.
FIG. 13. Geographic distribution of solution clusters of all PFG assemblages. Numbers (1–8) refer to the seriation solutions from Fig. 12. tion is changed. Differentiating these alter- analysis reflect interaction history and thus natives will require new, more complex community structure, we must consider classifications, and much larger sample less easily controlled sources of variation. sizes (Lipo and Madsen 1996) to support For example, we know that Phillips and col- the use of such a classification. leagues (1951) attempted to gain uniform coverage of the study area, but there is a SUMMARY strong possibility given their focus on large, mound sites that a bias exists in the survey Before we can conclude that the groups data. We cannot, from the PFG data alone, of assemblages formed throughout this determine whether the seriation groups we
AID JAA 0314 / ai06$$$$22 01-05-98 08:02:32 jaaal AP: JAA CULTURAL TRANSMISSION 325 Seriation solutions of PFG assemblages with sample sizes that meet asymtoptic means and variance criteria (Type I) as measured by FIG. 14. bootstrap assessment. Six solutions were created (1–6).
AID JAA 0314 / ai06$$0314 01-05-98 08:02:32 jaaal AP: JAA 326 LIPO ET AL.
FIG. 15. Geographic distribution of seriation clusters of PFG assemblages that meet the asymtoptic means and variance criteria (Type I) as measured by bootstrap assessment.
detected are a reflection of past interaction ity between PFG assemblages that could ex- history or the consequence of simply ‘‘miss- plain portions of the variability we see, and ing’’ assemblages intermediate in space thus nullify an interpretation that seriation (and in class frequencies) between those an- groups represent lineages intersected by an alyzed here. arbitrary measure of interaction density. Given the ways in which the fieldwork Any factor that creates differential represen- was done and samples collected, there are tation of ceramic types across several assem- also a host of issues concerning comparabil- blages will potentially structure the distribu-
AID JAA 0314 / ai06$$$$22 01-05-98 08:02:32 jaaal AP: JAA CULTURAL TRANSMISSION 327
FIG. 16. Three hierarchical levels of classification used to assess effects of classification level on group membership. Here we have aggregated the original PFG types (the bottom level) to create two coarser classification levels. tion of assemblages across a seriation inde- Our goal has been to examine ways in pendently of space or time and thus yield which archaeologists can create popula- either systematic or random errors in tion-scale descriptions of the kind needed mapped representations of communities. If, for evolutionary explanations. Starting for example, assemblages differ in the over- with a simple theory of cultural transmis- all frequency decorated of ceramics (a factor sion, we rationalized deterministic fre- almost certainly under the control of selec- quency seriation as the best method for tion), or if differences in settlement patterns examining interaction history in the ar- or geomorphology create differences in the chaeological record. The nature of cul- kinds of sites available in the surface record, tural transmission leads us to expect that some of the variability in class frequencies lineages will not be discrete bounded enti- between assemblages will be due to these ties but anomalously dense clouds of in- factors rather than transmission. Solving teraction within a background of continu- these issues requires additional fieldwork in ous transmission. As suggested through- the study area. out this paper, this fact places enormous Nevertheless, if we can measure the ef- data requirements on efforts to detect lin- fects of such possibilities using newly col- eage structure from aggregate frequency lected data to calibrate existing assemblages, data. We found, looking at the large data- we can finally test the hypothesis that the set offered by Phillips et al.’s Lower Mis- seriation groups discussed here represent sissippi Valley Survey, that seriation de- lineages or communities measured indi- tected clear clusters of assemblages that rectly via the frequency of interaction. Even are spatially coherent, but that explana- if there are serious problems with the com- tion of these clusters required many parability of PFG assemblages and frequen- sources of data simply not available from cies (and there almost certainly are), the fact their information. Solving the problems that the groups outlined here match culture inherent in traditionally collected data historical phases closely suggests that both such as PFG’s is a task that will take many culture historical methods and our approach years of dedicated and rigorous work. to seriation are measuring interaction inten- Our preliminary analyses, however, sug- sity at least indirectly. gest strongly that the effort will yield val-
AID JAA 0314 / ai06$$$$22 01-05-98 08:02:32 jaaal AP: JAA 328 LIPO ET AL.
FIG. 17. Geographic distribution of maximal seriation solution clusters of PFG assemblages that meet asymtoptic means and variance criteria as measured by bootstrap assessment using three differ- ent classification levels specified in Fig. 16. The geographic background has been removed for clarity. The effects of changing the level of classification for the PFG assemblages in the direction of coarser classes (fewer dimensions) is to simply create larger groups of assemblages, and in some cases allowing coalescence of formerly separate assemblages into more inclusive groups. The groups formed retain a great deal of spatial contiguity. This result suggests that the seriations performed here may reflect structure in the interaction history of the region and thus may map the approximate location of lineages. More detailed classifications and larger sample size, however, are needed to confirm this conclusion.
uable results. Although the task is diffi- fining population-scale units, a critical cult, we are now in a position to see the step in formulating rigorous evolutionary shape of methods and techniques for de- explanations.
AID JAA 0314 / ai06$$$$22 01-05-98 08:02:32 jaaal AP: JAA CULTURAL TRANSMISSION 329
REFERENCES CITED Cordell, L., and F. Plog 1979 Escaping the confines of normative thought: A reevaluation of Puebloan prehistory. American Abbott, D. N. Antiquity 44:405–429. 1972 The utility of the concept of phase in the ar- Cordell, L. S., and V. J. Yannie chaeology of the southern northwest coast. Sy- 1991 Ethnicity, ethnogenesis, and the individual: A esis 5:267–278. processual approach toward dialogue. In Pro- Beals, R. L., G. W. Brainerd, and W. Smith cessual and postprocessual archaeologies: Multiple 1945 Archaeological studies in Northeast Arizona. ways of knowing the past, edited by R. Preucel, University of California Publications in Ameri- pp. 96–107. University of Southern Illinois can Archaeology and Ethnology 44(1). Press, Carbondale. Crow, J. F., and M. Kimura Beck, C. 1970 An introduction to population genetics theory. 1995 Style, function, technology, and chronological Harper Row, New York. units. Paper presented at the 60th Annual Meeting of the Society for American Archaeol- Darwin, C. ogy, Minneapolis, MN. 1859 On the origin of species by means of natural selec- tion, or the preservation of favoured races in the Berry, K. J., K. L. Kvamme, and P. W. J. Mielke struggle for life. John Murray, London. 1980 Permutation techniques for the spatial analysis Dunnell, R. C. of the distribution of artifacts into classes. 1970 Seriation method and its evaluation. American American Antiquity 45:55–59. Antiquity 35:305–319. 1983 Improvements in the permutation test for the 1971 Systematics in prehistory. Free Press, New York. spatial analysis of the distribution of artifacts 1978a Archaeological potential of anthropological into classes. American Antiquity 48:547–553. and scientific models of function. In Archaeo- Blute, M. logical essays in honor of Irving Benjamin Rouse, 1979 Sociocultural evolutionism. Behavioral Science edited by R. C. Dunnell and E. S. Hall. Mouton, 24:46–59. The Hague. 1978b Natural selection, scale, and cultural evolution: Bonner, J. T. Some preliminary considerations. Paper pre- 1980 The evolution of culture in animals. Princeton sented at the 77th Annual Meeting of the University Press, Princeton. American Anthropological Association., Los Boone, J., and E. A. Smith Angeles, California. 1998 Is it evolution yet? A critique of evolutionary 1978c Style and function: A fundamental dichotomy. archaeology. Current Anthropology 39(1). American Antiquity 43:192–202. Boyd, R., and P. J. Richerson 1980 Evolutionary theory and archaeology. In Ad- 1985 Culture and the evolutionary process. University vances in archaeological method and theory, edited of Chicago Press, Chicago. by M. B. Schiffer, pp. 35–99. vol. 3. Academic Press, New York. Brain, J. P. 1981 Seriation, groups, and measurement. In Ma- 1978 The archaeological phase: Ethnographic fact or nejos de Datos Y Methods Mathematicos de Ar- fancy? In Archaeological essays in honor of Irving qualogia, edited by G. L. Cowgill, R. Whallon B. Rouse, edited by R. C. Dunnell and E. S. Hall, and B. S. Ottaway. Union Internacional de pp. 311–318. Mouton, New York. Ciences Prehistoricas y Protohistoricas, Mex- 1985 Introduction: Update of De Soto studies since ico DF. the United States De Soto Expedition Commis- 1982 Science, social science and common sense: The sion Report. In Final report of the United States agonizing dilemma of modern archaeology. De Soto Commission., edited by R. Swanton, pp. Journal of Anthropological Research 38:1–25. xi–xlix. Smithsonian Institution Press, Wash- 1986 Five decades of American archaeology. In ington. American archaeology past and future, edited by Cavalli-Sforza, L. L., and M. W. Feldman D. Meltzer, D. Fowler and J. Sabloff, pp. 23– 1981 Cultural transmission and evolution. Princeton 49. Smithsonian Institution, Washington, D.C. University Press, Princeton. 1996 Griffin’s ‘cultural change and continuity’; the synthesis of time and space. Paper presented at Chapman, C. H. the Paper presented at the First Annual G. R. 1989 The archaeology of Missouri, volume II. Univer- Willey Symposium on the History of Archae- sity of Missouri Press, Columbia. ology, 61st Annual Meeting of the Society for Cochran, W. G. American Archaeology, April l996., New Or- 1975 Sampling techniques. Wiley, New York. leans.
AID JAA 0314 / ai06$$$$22 01-05-98 08:02:32 jaaal AP: JAA 330 LIPO ET AL.
Dunnell, R. C., and W. S. Dancey 1952 Measurements of some prehistoric design elements 1983 The siteless survey: A regional scale data col- in the southeastern states 44. Anthropological Pa- lection strategy. In Advances in archaeological pers of the American Museum of Natural His- method and theory, vol. 6, edited by M. B. Schif- tory, New York. fer, pp. 267–287. vol. 6. Academic Press, New 1954 The type concept revisited. American Anthro- York. pologist 56: 42–54. Durham, W. H. 1962 A quantitative method for deriving cultural chro- 1976 The adaptive significance of cultural behavior. nology. Technical Manual, No. 1. Pan American Human Ecology 4:89–121. Union. 1979 Toward a coevolutionary theory of human bi- Ford, J. A., and G. R. Willey ology and culture. In Evolutionary biology and 1941 An interpretation of the prehistory of the East- human social behavior: an anthropological perspec- ern United States. American Anthropologist 43: tive, edited by N. Chagnon and W. Irons, pp. 325–363. 39–58. Duxbury Press, North Scituate, MA. Fox, G. L. 1990 Advances in evolutionary culture theory. An- nual Review Of Anthropology 19(V19):187–210. 1992 A critical evaluation of the interpretive framework of the Mississippi period in Southeast Missouri. 1991 Coevolution: Genes, culture, and human diversity. Unpublished Ph.D. dissertation, Department Stanford University Press, Stanford, Calif. of Anthropology, University of Missouri-Co- 1992 Applications of evolutionary culture theory. lumbia. Annual Review Of Anthropology 21(V21):331– 355. 1998 An examination of Mississippian-period phases in Southeastern Missouri. In The archae- Efron, B. ology of the Central Mississippi River Valley, ed- 1982 The jackknife, the bootstrap and other resampling ited by R. C. Dunnell and M. J. O’Brien, Uni- plans. SIAM, New York. versity of Alabama Press, Tuscaloosa. 1983 Computer-intensive methods in statistics. Sci- Good, P. entific American May:116–130. 1994 Permutation test: A practical guide to resampling Efron, B., and R. Tibshirani methods for testing hypotheses. Springer-Verlag, 1993 An introduction to the bootstrap. Chapman and New York. Hall, New York. Gould, S. J., D. M. Raup, J. J. Sepkoski, T. J. M. Schopf, Eighmy, J. L., and J. M. LaBelle and D. S. Simberloff 1996 Radio carbon dating of twenty-seven plains 1977 The shape of evolution: A comparison of real complexes and phases. Plains Anthropologist and random clades. Paleobiology 3(1):23–40. 41(155):53–69. Graves-Brown, P. Fisher, R. A. 1996 In search of the watchmaker: Attribution of 1936 Statistical methods for research workers. Oliver agency in natural and cultural selection. In and Boyd, Edinburgh. Darwinian archaeologies, edited by H. D. G. Ford, J. Maschner, pp. 165–179. Plenum Press, New 1935a Ceramic decoration sequence at an old Indian York. village site, near Sicily Island, Louisiana. Loui- Grayson, D. K. siana State Geological Survey, Department of Con- 1989 Sample size and relative abundance in archae- servation, Anthropological Study 1. ological analysis: Illustrations from spiral frac- 1935b An introduction to Louisiana archaeology. tures and seriation. In Quantifying diversity in Louisiana Conservation Review 4(5):8–11. archaeology, edited by R. D. Leonard and G. T. 1936a Archaeological methods applicable to Louisi- Jones, pp. 79–84. Cambridge University Press, ana. Proceedings of the Louisiana Academy of Sci- Cambridge. ences 3:102–105. Gulick, J. T. 1936b Analysis of Indian village site collections from Lou- 1872 On diversity of evolution under one set of ex- isiana and Mississippi. Department of Conserva- ternal conditions. Journal of the Linnean Society tion, Louisiana State Geological Survey, An- of Zoology 11: 496–505. thropological Study, 2. New Orleans. 1938 A chronological method applicable to the 1905 Evolution, racial and habitudinal. Carnegie Insti- Southeast. American Antiquity 3:260–264. tution of Washington, Washington, D.C. 1949 Cultural dating of prehistoric sites in Viru´ Val- Ha¨gerstrand, T. ley Peru. American Museum of Natural History, 1952 The propagation of innovation waves. Royal Uni- Anthropological Papers 5: 263–266. versity of Lund, Dept. of Geography, London.
AID JAA 0314 / ai06$$$$22 01-05-98 08:02:32 jaaal AP: JAA CULTURAL TRANSMISSION 331
Hammond, N. 1971 Introduction to Middle Missouri Archeology. An- 1972 Location models and the site of Lubaantun: A thropological Papers 1. National Park Service, classic Maya center. In Models in Archaeology, Washington. edited by D. L. Clarke, pp. 757–300. Methuen, Lewontin, R. C. London. 1974 The genetic basis of evolutionary change. Colum- Holmes, W. H. bia University Press, New York. 1903 Aboriginal pottery of the Eastern United States. Lipo, C. P., and M. E. Madsen Bureau of American Ethnology, 20th Annual 1996 Problems and solutions in the study of social orga- Report., Washington, D.C. nization. Paper presented at the Society for House, J. American Archaeology, New Orleans, LA. 1991 Monitoring Mississippian dynamics: Time, settle- ment, and ceramic variation in the Kent Phase, Love, M. W. Eastern Arkansas. Unpublished Ph.D. disserta- 1993 Ceramic chronology and chronometric dating: tion, Department of Anthropology, Southern Stratigraphy and seriation at La Blanca Guate- Illinois University, Carbondale. mala. Ancient Mesoamerica 4:17–29. 1993 Dating the Kent Phase. Southeastern Archaeol- Lyman, R. L., M. J. O’Brien, and R. C. Dunnell ogy 12(1):21–32. 1997 The rise and fall of culture history. Plenum Press, 1995 Investigating Mississippian settlement variability New York. in the Lower St. Francis Basin, Eastern Arkansas. Lynch, A. Paper presented at the 52nd Southeastern Ar- 1996 The population memetics of birdsong. In Ecol- chaeological Conference, Knoxville, Tennes- ogy and evolution of acoustic communication in see, November 9, 1995. birds, edited by D. E. Kroodsma and E. H. Hudson, C. Miller, pp. 181–197. Cornell University Press, 1985 De Soto in Arkansas: A brief synopsis. Arkansas Ithaca, New York. Archeological Society Field Notes 205:3–12. Mainfort, R. C. Johnson, K. W., and B. C. Nelson 1995 Cluster analysis and archaeological phases: An 1990 The Utina: Seriations and chronology. The Flor- example from the Central Mississippi Valley. ida Anthropologist 43(1):48–62. Paper presented to the Midwest Archaeologi- Kimura, M. cal Conference, Beloit, Wisconsin, 1995. 1977 Preponderance of synonymous changes as evi- dence for the neutral theory of evolution. Na- Marquardt, W. H. ture 267:275–276. 1973 Advances in archaeological seriation. Advances 1983 The neutral theory of molecular evolution. Cam- in Archaeological Method and Theory 1:257–314. bridge University Press, Cambridge. McKern, W. C. Kimura, M., and T. Ohta 1939 The mid-western taxonomic method as an aid 1971 Theoretical aspects of population genetics. Univer- to archaeological study. American Antiquity sity of Princeton Press, Princeton, New Jersey. 4:301–313. King, A., and J. Freer Mooney, C. Z., and R. D. Duval 1995 The Mississippian Southeast: A world-systems 1993 Bootstrapping: A nonparametric approach to statis- perspective. In Native American interactions: tical inference. Quantitative Applications in the Multiscalar analyses and interpretations in the Social Sciences 95. Sage Publications, New Eastern Woodlands, edited by M. S. Nassaney York. and K. E. Sassaman, pp. 266–288. The Univer- Morse, D. F. sity of Tennessee Press, Knoxville. 1973 Nodena: An account of 75 years of archaeological King, J. L., and T. H. Jukes investigation in southeast Mississippi County, Ar- 1969 Non-Darwinian evolution: Random fixation of kansas. Arkansas Archeological Survey Re- selectively neutral mutations. Science 164:788– search Series, 4, Pine Bluff, Arkansas. 798. 1982 Regional overview of northeastern Arkansas. Krause, R. A. In Arkansas archeology in review, edited by N. L. 1977 Taxonomic practice and Middle Missouri Pre- Trobowitz and M. D. Jeter, pp. 20–36. vol. 15. history: A perspective on Donald J. Lehmer’s Arkansas Archaeological Survey Research Se- contributions. Plains Anthropologist, Memoir ries, Fayetteville, Arkansas. 13:5–13. 1989 The Nodena phase. In Nodena: An account of 90 Lehmer, D. J. years of archaeological investigation in southeast 1966 Fire Heart Creek site. Publications in Salvage Ar- Mississippi County, Arkansas, edited by D. F. cheology 1. Smithsonian Institution, River Ba- Morse, pp. 97–113. vol. no. 30. Arkansas Ar- sin Surveys, Lincoln, NE. cheological Survey, Fayetteville, Arkansas.
AID JAA 0314 / ai06$$$$22 01-05-98 08:02:32 jaaal AP: JAA 332 LIPO ET AL.
1990 The Nodena phase. In Towns and temples along Payne, R. B. the Mississippi, edited by D. H. Dye and C. A. 1996 Song traditions in Indigo Buntings: Origin, im- Cox, pp. 69–97, University of Alabama Press, provisation, dispersal, and extinction in cul- Tuscaloosa. tural evolution. In Ecology and evolution of Morse, D. F., and P. A. Morse acoustic communication in birds, edited by D. E. Kroodsma and E. H. Miller, pp. 198–220. Cor- 1983 Archaeology of the Central Mississippi Valley. Ac- nell University Press, Ithaca, New York. ademic Press, New York. 1990 The Spanish Exploration of Arkansas. In Co- Peregrine, P. lumbian consequences, edited by D. H. Thomas, 1991 Prehistoric chiefdoms of the midcontinent: A ed. Pp. 197–210, Vol. 2. Smithsonian Institu- world-system based on prestige goods. In Core/ tion Press, Washington. periphery relations in precapitalist worlds, edited 1996 Northeast Arkansas. In Prehistory of the Central by C. Chase-Dunn and T. Hall, pp. 193–211. Mississippi Valley, edited by C. H. McNutt, pp. Westview Press, Boulder. 119–135. The University of Alabama Press, 1992 Mississippian evolution: A world-system perspec- Tuscaloosa. tive. Monographs in World Prehistory 9. Pre- Morse, P. A. history Press, Madison, Wisconsin. 1981 Parkin. Arkansas Archeological Survey Re- 1995 Networks of power: The Mississippian world- search Series No. 13, Pine Bluff. system. In Native American interactions: Multi- 1990 The Parkin site and the Parkin phase. In Towns scalar analysis and interpretations in the Eastern and temples along the Mississippi, edited by D. Woodlands, edited by M. S. Nassaney and K. E. Dye and C. Cox, pp. 119–134. University of Sassaman, pp. 247–265. University of Tennes- Alabama, Tuscaloosa. see Press, Knoxville. Neiman, F. Phillips, P. 1990 An evolutionary approach to archaeological infer- 1970 Archaeological survey in the lower Yazoo Basin, ence: Aspects of architectural variation in the 17th- Mississippi. Peabody Museum, Cambridge, century Chesapeake. Ph.D. dissertation, Yale MA. University. Phillips, P., J. A. Ford, and J. B. Griffin 1992 Stylistic variation in evolutionary perspective: Im- 1951 Archaeological survey in the Lower Mississippi Al- plications for Middle Woodland Ceramic diversity. luvial Valley, 1940–1947. Papers of the Peabody Paper presented at the 57th Annual Meeting Museum of American Archaeology and Eth- of the Society for American Archaeology, Pitts- nology 25. Harvard University, Cambridge. burgh, PA. Phillips, P., and G. R. Willey 1995 Stylistic variation in evolutionary perspective: 1953 Method and theory in American archaeology: Inferences from decorative diversity and inter- An operational basis for culture-historical inte- assemblage distance in Illinois Woodland Ce- gration. American Anthropologist 55:615–33. ramic assemblages. American Antiquity 60(1):7– 36. Plog, F. 1979 Prehistory: Western Anaszi. In Handbook of Newell, H. P., and A. Krieger North American Indians, Volume 9: Southwest, 1949 The George C. Davis site, Cherokee County, Texas. edited by A. Ortiz, pp. 109–130. vol. 9. Smith- Society for American Archaeology and the sonian Institution, Washington, D.C. University of Texas, Menasha, Wis. 1983 Political and economic alliances on the Colo- Noreen, E. rado Plateaus, A.D. 400–1450. In Advances in 1989 Computer intensive methods for testing hypotheses. world archaeology, edited by F. Wendorf and Wiley, New York. A. Close, pp. 289–330. vol. 2. Academic Press, New York. O’Brien, M. J. 1994 Cat monsters and head pots: The archaeology of Renfrew, C. Missouri’s Pemiscot Bayou. University of Mis- 1977 Alternative models for exchange and spatial souri Press, Columbia. distribution. In Exchange systems in prehistory, edited by T. K. Earle and J. E. Ericson, pp. 71– O’Brien, M. J., and G. L. Fox 90. 1994 Assemblage similarities and dissimilarities. In Cat monsters and head pots: The archaeology of Rindos, D. Missouri’s Pemiscot Bayou, edited by M. J. 1980 Symbiosis, instability, and the origins and O’Brien, pp. 61–93. University of Missouri spread of agriculture: A new model. Current Press, Columbia. Anthropology 21:751–772.
AID JAA 0314 / ai06$$$$22 01-05-98 08:02:32 jaaal AP: JAA CULTURAL TRANSMISSION 333
Rouse, I. Upham, S., K. G. Lightfoot, and G. M. Feinman 1939 Prehistory in Haiti. Yale University Publications 1981 Explaining socially determined ceramic distri- in Anthropology 21. bution in the Prehistoric Plateau Southwest. 1955 On the correlation of phases of culture. Ameri- American Antiquity 46:822–833. can Anthropologist 57:713–722. Upham, S., and F. Plog 1965 The place of ‘Peoples’ in prehistoric research. 1986 The interpretation of prehistoric political com- Journal of the Royal Anthropological Institute plexity in the Central and Northern Southwest: 95(1–15). Toward a mending of the models. Journal of Scollar, I. Field Archaeology 13:223–238. 1995 Bonn archaeological software package. 4.5 ed. Wcislo, W. T. Bonn Archaeological Software, Cologne. 1989 Behavioral environments and evolutionary change. Annual Review of Ecology and Systemat- Simon, J. L. ics 20:137–169. 1990 Resampling: Probability and statistics in a radically different way. Unpublished manuscript avail- Weinberg, S. able from Resampling Project, College of Busi- 1997 The first elementary particle. Nature 386(6632): ness, University of Maryland, College Park, 213–215. MD 20742. Willer, D., and M. Webster 1970 Theoretical concepts and observables. Ameri- Simon, J. L., and P. Bruce can Sociological Review 35:748. 1991 Resampling: A tool for everyday statistical Willey, G. R., and P. Phillips work. Chance 4(1). 1955 Method and theory in American archaeology. Smith, G. P. II: Historical developmental interpretation. 1990 The Walls Phase and its neighbors. In Towns American Anthropologist 57:723–819. and temples along the Mississippi, edited by D. 1958 Method and theory in American archaeology. Uni- Dye and C. Cox, pp. 135–169. University of versity of Chicago Press, Chicago. Alabama Press, Tuscaloosa. Williams, S. Sokal, R. R., and P. H. A. Sneath 1954 An archaeological study of the Mississippian cul- 1963 Principles of numerical taxonomy. Freeman, San ture in Southeast Missouri. Unpublished Ph.D. Francisco. dissertation, Department of Anthropology, Soltis, J., R. Boyd, and P. J. Richerson Yale University. 1995 Can group-functional behaviors evolve by cul- 1980 The Armorel Phase: A very Late Complex in tural group selection? An empirical test. Cur- the Lower Mississippi Valley. Southeastern Ar- rent Anthropology 36(3):473–474. chaeological Conference Bulletin 22:105–110. Wright, S. Teltser, P. A. 1931 Evolution in Medelian populations. Genetics 1995 Culture history, evolutionary theory, and fre- 16:97–159. quency seriation. In Evolutionary archaeology: 1932 The role of mutation, inbreeding, crossbreed- Methodological issues, edited by P. A. Teltser, pp. ing and selection in evolution. Proceedings of 51–68. University of Arizona Press, Tucson. the Sixth International Congress of Genetics Trigger, B. G. 1:356–366. 1989 A history of archaeological thought. Cambridge 1940 The statistical consequences of Mendelian he- University Press, Cambridge; New York. redity in relation to speciation. In The new sys- Upham, S. tematics, edited by J. S. Huxley, pp. 161–184. 1982 Polities and power: An economic and political his- Oxford University Press, Oxford. tory of the Western Pueblo. Academic Press, New 1948 On the roles of directed and random changes York. in gene frequency in the genetics of popula- 1983 Intensification and exchange: An evolutionary tions. Evolution 2:279–294. model of non-egalitarian socio-political organi- 1949 Population structure in evolution. Proceedings zation for the prehistoric Plateau Southwest. of the American Philosophical Society 93:471–478. In Ecological models in economic prehistory, ed- Zeier, C. D. ited by G. Bronitsky, pp. 219–245. vol. 29. Ari- 1980 The Willey and Phillips system revisited: A zona State University Anthropological Re- proposed expansion of the paradigm. Plains search Papers, Tempe. Anthropologist :29–36.
AID JAA 0314 / ai06$$$$22 01-05-98 08:02:32 jaaal AP: JAA