SIMULATION DRAWS ITS explanatory power from three main sources. First, it forces the researcher to model a phenomenon of interest. Second, it allows the comparison of outcomes under variable, controlled conditions. Third, it creates an environment in which, as Gilbert and Hammel (1966:72) put it, the researcher can ask the question, What would happen if . . . and can expect a range of quantitatively specified answers.
The validity of simulation of social systems and their evolution rests on two fundamental assumptions: (1) social change is constrained by rules and is not a purely historical, random phenomenon, and (2) even though specific rules may change over time, rule change is systematic and can be understood. Simulation is therefore ideally suited for identifying rules and mechanisms of social change, and also for answering the fundamental question of whether social systems are indeed systematic and rule-bound. If simulation fails to explain or even replicate the evolution of social systems, then both assumptions are likely false.
A model is produced every time a researcher thinks: “If I understand what is going on here, elements Al through Ai, which interact according to rules fl through fi, are responsible for what I am observing.” Building a model of a phenomenon involves identifying the elements responsible for an observed outcome and then positing the rules which regulate the interactions between these elements. In this sense, modeling is identical to Whallon’s (1982:156) concept of explanation, which “specifies both: (1) the relevant critical variables, and, (2) the precise form of the relationships among these variables.” Theoretical explanation involves variables of general applicability, whereas a model specifies values or ranges for those variables, “thereby determining the exact form of the model from the given relationships among these variables” (Whallon 1982:156).
A model is a proposed system of objects and relationships that should produce a certain outcome. A good model should restrict itself to identifying a minimum set of elements and relationships which are sufficient for the production of an observed outcome. A model can be seen as a hypothesis or a system of hypotheses about how parts of the world function. Simulation, aided by the computer, is simply a tool for testing these hypotheses. By adding a time dimension to the model, and by observing the interactions between the objects, the researcher determines whether a simulated outcome has any attributes in common with some real-world observed phenomenon. If the simulated outcome diverges significantly from the real-world observed outcome, the model must be inaccurate at some fundamental level. The set of elements and relationships identified is either inappropriate or insufficient for the replication of the phenomenon under study.
Models useful for the investigation of social systems must be generative, in that they should be designed, using a limited number of clearly abstracted parts, to generate regularities by specified logical operations (Barth 1966:v). The magnitude or interaction of these parts “can be varied, so that one model can be made to produce a number of different forms” (Barth 1966:v).
Descriptive models may be enlightening and may be powerful heuristic devices. They can help us clarify and refine our understanding of what we study. But only generative models can serve for hypothesis testing, since they alone include processes (relationships between elements) which can be observed to unfold over time and produce an outcome. Another critical feature of generative models is that they can surprise us. A descriptive model, by definition, describes a known outcome. A generative model must be observed in operation to find out whether it will produce a predicted outcome.
A model becomes a simulation when the relationships it contains are allowed to unfold over time. This unfolding typically involves calculations and may be aided by computers. Some simulations can be performed without computers. A lithic technology replication study, for example, usually identifies some elements and relationships and makes testable predictions about outcomes. In that case, the simulated outcome is produced not by computation but by the interaction of physical forces. The process is the same. The outcome of the physical experiment may fail to match a real-world outcome observed in the archaeological record, in which case the researcher must reformulate the model and watch it unfold over time again.
Some early anthropological simulation experiments demonstrate the independence of the method and the computer, but also the usefulness of the latter. In 1954, Alice Brues used what she termed the method of “experimental calculations” (Brues 1954:596) to test hypotheses about the evolution of the A-B-O blood group system. She used a desk calculator to watch gene frequencies change in simulated populations over two generations, and then extrapolated the changes in time. By 1963, Brues had updated her work so that a computer could perform her experimental calculations for a large number of generations. She was then able to observe the outcome of the system she had proposed ten years earlier but had then only been able to extrapolate. In theory, any simulation could be performed with pen and paper. In practice, a computer is the only tool that currently allows the simulation of systems that are large and complex enough to be helpful in the study of the evolution of human societies.
The authors of early simulations of social systems were aware of the method’s potential for the resolution of general problems across a wide range of specific contexts. The 1960s saw a movement in demography toward the use of computer simulation in order to better account for cultural and social constraints on human demographic processes in general (May and Heer 1968; Ridley and Shepps 1966). Anthropologists were very interested in the new tool for the same reasons. Gilbert and Hammel (1966:72) proposed early on that simulation of social systems could address a central problem of social anthropology, which is that they lack theoretical null hypotheses for how ecological, demographic, and cultural variables affect social organization, making it impossible to connect rates of occurrence of phenomena with the hypothesized functions that produced them.
Anthropologists at the time were interested in the potential of simulation to tackle the complexities of marriage systems. They also realized that while demographic equations can accurately represent the behavior of large population aggregates, the typically small units studied by anthropologists call for the use of computational methods. Dyke (1981:195) found traditional demographic models to be inadequate for “small populations where random events can have a large effect,” and saw simulation as a way of incorporating specific parameters such as marriage and fertility into demographic models, particularly for small populations.
Human social relations and material systems are embedded in demographic networks. The logical starting point for both social anthropologists and archaeologists was the adaptation of demographic simulation for the study of anthropological problems.
Kunstadter et al. (1963:511) sought to answer the question, “What frequency of the preferred type of marriage can be expected?” in a population with an ideal marriage pattern, under a variety of demographic conditions. Gilbert and Hammel (1966:73), in a similar vein, “wished to demonstrate that at least some portion of that preference was an epiphenomenon of territorial endogamy alone and to estimate how much [of the preference] could be accounted for in this way” (emphasis original). While both studies were loosely based on ethnographic contexts (India and the Near East respectively), they both dealt with a specific problem and attempted to build models which could be applied to any particular context. Taking the opposite approach, MacCluer, Neel, and Chagnon (1971) simulate a particular population in detail in order to see how general demographic problems manifest themselves in a given context. Their simulation takes as a starting point the actual living population of four Yanomamö villages, but its goal is clearly generalizing. The integration of simulation into anthropology’s methodological arsenal fits in with the increasingly general and theoretical concerns of 1960s and 1970s in anthropology, at the expense of particularistic “objective” case studies, a change lamented in bitter quantitative detail by Carlton Coon (1977).
Early simulations of social systems demonstrated the potential of the technique for studying the properties of specific population problems as well as the influence of demographic conditions on particular social contexts. This demonstration paved the way for the use of social simulation as a tool for archaeological research.
Doran (1970) first brought simulation to archaeology’s attention. He proposed it as a tool that could allow controlled laboratory-like experiments (Doran 1970:297), the lack of which seemed to hamper research in the historical sciences. He envisioned a simulation that would specify elements and interactions, and through them produce an artificial archaeological record for an imaginary island. By varying the elements and the rules for their interaction, one could produce different artificial archaeological records and validate proposed models by comparing the attributes and characteristics of artificial records with their real-world counterparts.
This was a very seductive proposition for archaeologists, and it was taken up almost immediately by D. H. Thomas (1972) in his classic simulation study of Shoshonean subsistence strategies. Thomas’s simulated Shoshones produced an artificial archaeological record, and he attempted to validate his model of their subsistence strategies (based on Steward’s 1970 model) by comparing it to archaeological observations.
As Doran envisioned and Thomas demonstrated, the nature of archaeological questions and archaeological data imposes special needs on the design of simulation for the investigation of extinct social systems. Some of these requirements have already been identified in the archaeological literature. They largely apply to any social simulation effort, and they therefore constitute a vital starting point for the simulation of past social systems.
Wobst (1974:151) proposed that computer simulations applied to archaeological social systems but should be general enough to account for the expected range of variation, simple enough to reduce the complexity to “intelligible dimensions,” and should produce testable predictions for future research (see table 16.1).
The generality requirement is important if simulation is to be a strategy that can help evaluate the statement that social organization is systematic rather than idiosyncratic. A simulation that can only deal with a particular context cannot help in this regard. It would not allow the comparison of different outcomes, since all would be the result of unique historical processes. A complete archaeological simulation model should be able to produce any observed form of social organization if the variability of social organization is indeed systematic. If the structure of a given simulation rules out an outcome whose attributes match those of a known or imaginable ethnographic or archaeological instance, the generality requirement cannot be met. The opposite case, in which a model produces forms that are not observed ethnographically or archaeologically does not invalidate it, and may in fact present great advantages.
The simplicity requirement specifies that the model should identify a minimum and sufficient set of elements and relationships for the production of the observed outcome. A model that contains all elements and relationships present in reality is just as hopelessly complex as reality, and therefore no help in understanding it. A model that is cluttered with elements or relationships not required for the production of the observed outcome obscures rather than enlightens.
The requirement of testability ensures that the model can contribute meaningfully to answering specific research questions. At a deeper level, it also ensures that the entire enterprise of social simulation could conceivably fail, thereby rejecting its own core assumptions, a bittersweet but necessary prospect.
It can be argued that the three requirements discussed above apply to any social simulation effort. But what should a specifically archaeological simulation look like? In order for a simulation of past social systems to meet these three requirements of generality, simplicity, and testability, it must meet three other requirements, two of which are specifically archaeological, and one of which is more generally anthropological.
First, the model must be capable of change. Second, it must produce a record that is observable but also dynamic over time. Third, archaeological simulation must deal with unequivocally human social systems. It must therefore make room for agents with cognitive systems, goals, and means to pursue their ends.
| General Requirements | |
| Generality | |
| Simplicity | |
| Testability | |
| Archaeological Requirements | |
| Change | |
| Spatial variability | |
| Macroregional and global transitions | |
| Divergence or regionalization | |
| Record | |
| Deposition | |
| Transformation (natural and social) | |
| Agent | |
| Cognitive system | |
| Learned means of encoded interaction | |
| Vertical and horizontal transmission and modification | |
| Negotiation of individual and collective action | |
| Situated in space and time | |
| Dependent on natural resources | |
If there is one thing on which all archaeologists agree, it is that there has been change over time. Even if we restrict the scope of archaeology to anatomically modern Homo sapiens, the evidence for change in all aspects of human life over time is overwhelming, from technology, to settlement, to subsistence, to artistic expression and ritual. If a simulation model is to claim that it has met the generality requirement, it must therefore be capable of producing change.
Change in the archaeological record exists in two dimensions: spatial and chronological. The simulation model must therefore be capable of local variability (spatial change) and transition (chronological change). In accordance with real-world observations, it must be capable of producing local as well as macroregional and global transitions. At least some of the transitions produced must be abrupt. The model must be able to produce increasing local variability over time, a special kind of two-dimensional transition known as divergence or sometimes as regionalization.
In order to meet the testability requirement, the model must not directly cause variability and transitions to take place through the operation of specific hard-wired rules, but must allow them to emerge by its structure. A model which is programmed for the production of a specific transition (e.g., to change hunter-gatherers into horticulturalists) cannot be of much help in testing hypotheses about the causes of that transition. Only a model from which transition can (or not) emerge will do so. A model that produces an unexpected transition has most value for meeting the testability requirement.
Since archaeology deals with a great time depth, change produced by the simulation must be long term, directional, and as undetermined as possible by the model’s initial structure. It follows that one of the main goals of archaeological simulation design must be the elimination of artificial structure imposed and produced by the rules and initial conditions of a model.
The fundamental nature of the problem of artificial structure in archaeological simulation was recognized early on (MacArthur, Saunders, and Tweedie 1976:322–323). For example, while the use of invariant life tables (listing probabilities of having a certain lifespan) is appropriate for the exploration of most short-range problems in conventional demography, they impose a constant demographic structure on simulated populations and are therefore inappropriate for archaeological simulation dealing with the long term. A system that allows demographic structure to emerge and vary over time, sometimes undergoing radical transformations in response to systemic change, is essential for archaeological simulation.
For example, it is possible to replace the traditional life table by a threat environment. In such a system, each agent has a series of traits that are used to determine its interaction with pathogenic and other environmental threats. These traits are inherited from parent to offspring and subject to random mutation. Each threat has a series of traits that determine the profile of the agents that it can attack. The threats, which in effect are also agents, constantly search through the population for victims that fit the profile in which they are interested. When threats find a suitable victim, they attack, causing some deterioration of health in the agent. The virulence of specific threats varies, and each threat can have biases for age and sex. Furthermore, some threats are subject to random mutation, and any of their traits at start can be modified over time (for a fuller explanation of the model, see Costopoulos 2002). This provides a constantly changing environment to which the agent population is forever adapting, thus causing the emergence of different demographic structures over time.
The emergence of life tables that result from the ongoing interaction between the agent population and the threat population shows the changing demographic structure of a population over time in a single run. Each run begins with a unique agent population and a unique threat environment. In each run, the population adapts to its environment over time, resulting in a reduced number of infant deaths and an overall longer life expectancy. Despite the changes in the threat environment in the form of threat mutations, both populations adapt in the same way. But in subsequent runs, the fifth millennium population experiences some pressure in the fifty-five- to sixty-five-year-old group, a phenomenon that is due to a unique feature of the interaction between agent population and threat environment in that particular run and could not have emerged from a system using a traditional life table.
In one model run, males show a higher rate of infant death than females by the fifth century. The initial situation can be varied for each run; for example, instead of beginning with no difference between male and female infant death rates initially, a difference can be imposed right from the start of a new run. For this particular simulation, different initial conditions generated similar outcomes within five centuries. This result can suggest hypotheses to test using further runs of the model. A more static and descriptive, more traditional demographic model based on life tables is not capable of generating such change.
From the very beginning, archaeologists have recognized that archaeological simulation must deal with the archaeological record. Doran (1970) made the production of an artificial archaeological record the centerpiece of his vision for archaeological simulation, and even suggested that rules should be included to determine “the extent to which material remains survive to be excavated” (Doran 1970:296).
Thomas (1972) was interested in the simulated physical traces of the behavior of his artificial Shoshones more so than in their actual behavior itself. Wobst (1974:173–176) extensively discusses the implications for the archaeological record of the outcome of his essentially demographic simulation. Wobst (1974:175) concluded that some of the simulated sites may “present only chance accumulations of artifacts deposited outside of occupation areas by repeatedly performed tasks.”
Even a relatively simple simulation of foraging behavior (Costopoulos 2001, 1999) can create highly complex artifact distributions. Figure 16.1 shows the deposition resulting from the simulated foraging behavior of fifty agents over a ten-year period around a lake and its associated river system. Each symbol represents an instance of a different technological material deposited as a result of the exploitation of a particular subsistence resource. From such a database, it is possible to test hypotheses about the constraints that relationships between technological and subsistence resources can impose on archaeological distributions.
Figure 16.1. Simulated foraging deposition from 50 agents over a 10-year period around a lake and its associated river system. Each symbol represents an instance of a different technological material deposited as a result of the exploitation of a particular subsistence resource.
Other early simulation studies focused on the comparison of simulated and observed regional artifact distributions in order to test hypotheses about the nature of prehistoric exchange networks (Elliot, Ellman, and Hodder 1978; Wright and Zedder 1977).
The comparison of artificial and observed archaeological records continues to be a central concern in archaeological simulation (Hazelwood and Steele 2004; Young 2002; Costopoulos 2001; Lake 2000; Varien et al. 1997). However, one of the critical features of the archaeological record has not yet been fully integrated into archaeological simulation. While many archaeological simulations produce an artificial record, few, if any, modify it over time. The simulation of taphonomic processes has been a fruitful area of research in experimental replication studies, as shown most notably by the Overton Down studies.
Doran (1970) proposed that archaeological simulation should include functions for reducing the information available over time, but it is equally important to include taphonomic functions which transform the information available. The archaeological record is not only partially destroyed over time but is modified in other ways as well.
For an approach to archaeological simulation to be complete, artificial archaeological records must be produced by simulated social systems and then modified over time by simulated natural systems as well as simulated social systems. Rules must exist for the interaction of the artificial record with natural and social forces. There have been simulation studies of how natural and social processes (such as plowing) can affect archaeological assemblages (Custer 1992; Yorston et al. 1990), but these have not usually been coupled with a simulation of the production of those same assemblages. Archaeological simulation must face the reality that archaeologists do not study the physical traces of past social systems, but rather their modified and partially preserved traces.
Archaeological simulation has tremendous potential for the study of taphonomy. It provides a means of observing the outcome of posited rules for change over time. It also makes it possible to know the state of the record at each step of its transformation, something that is clearly not available with the real-world archaeological record. Taphonomic processes are one of the blackest of black boxes in archaeology (Leach 1973). Simulation offers a chance to propose mechanisms for the box and observe their functioning over time. While the interface between simulated and observed archaeological records has been an important object of study, that part of archaeological simulation must also eventually meet the criterion of change.
This is the current frontier of archaeological simulation, and this part of the chapter is necessarily more speculative than the rest. It is intended as a stimulus to further discussion.
The machinery exists in the various corners of simulation studies in archaeology for the generation of change, for the deposition and transformation of simulated archaeological records, for the creation of spatial distribution due to exchange, and for the demographic functioning of populations. But the transformation of those populations into functioning, humanlike social systems is in its infancy.
The archaeological record is the product of the operation of social systems. Fundamentally, archaeological simulation must be social simulation. Social systems can be seen as constellations of agents with different goals, beliefs, and access to action. Belief systems are shaped by experience, but they also preexist experience and shape action in pursuit of goals. At least some of the variability and change observed in social systems, past and present, is attributable to interactions between large numbers of agents with individual cognitive systems (in terms of functioning and contents). The archaeological record is partially the result of negotiated action, which reflects both individual and collective goals and ideologies. Archaeological simulation must therefore include networks of interacting agents, with individual cognitive systems, and in constant negotiation, if it is to meet the requirement of generality.
The use of computer simulation in anthropology and archaeology has been associated with processualism and sometimes perceived as an approach which minimizes the role of ideology and action, and adheres to canons of determinism. It need not do so. In his debate with Lansing (2000) over the simulation of Balinese irrigation, Helmreich (1999:249) argued that by rendering complicated political, economic, and social issues into “bounded technical problems,” computer simulation “erases internal community politics and ignores the local and global political economic context in which communities exist.”
Postmodern and contextual approaches to prehistory, which often emphasize agency (see Gardner, chapter 7; Shanks, chapter 9; Koerner and Price, chapter 21), variability, and history at the expense of causality, systemic regularity, and evolutionary process, have contributed powerful critiques to archaeology. These approaches stress that people are not just automatons reacting to an external world but actively create their own social reality as they participate in it (Barfield 1997:4; Dornan 2002:304).
Such critiques pose a significant challenge to explaining, with generally applicable theory involving a limited number of elements and rules, the emergence and evolution of the full range of variability in human social organization. The increasing use of agent-based approaches to archaeological simulation, combined with a reliance on complexity theory (see Bentley and Maschner, chapter 15), is partially a response to this challenge.
In order to address these critiques, social simulation must build networks of agents with individual cognitive systems, knowledge bases, and belief systems. These agents must be capable of action, and their actions must both influence and be influenced by the social and natural context of the system.
Agent-based approaches allow the construction of simulated social systems which can both show broad-scale regularity and local variability, while producing systems characterized by a wide diversity of individual viewpoints, strategic goals, and even belief systems (Costopoulos 2002; Binmore 2001; Lake 2000; Doran 1997; de Vos and Zeggelink 1994). Each individual participant in the social process must have its own cognitive system. It must acquire information and modify its beliefs as processes unfold, but it must also make its decisions within the constraints of the existing contents of its own cognitive system.
The contents of cognitive systems, at least parts of which organize the growth of the system itself, are subject to transmission and modification. An archaeological simulation must have means through which new agents, created in an already functioning social environment, acquire cognitive content through vertical and horizontal transmission (Shennan 2002; see Collard et al., chapter 13).
Minimally, agents in archaeological simulation must need to subsist from resources found in their natural environment and perceived through their cognitive system. They must interact with other agents in their social environment. They must be situated in space and therefore be able to act on a limited portion of their natural and social environment at any one point in time. Through their cognitive system, agents must be aware of a greater portion of their environment than that which is available for immediate action. They must plan their current and future actions on the basis of this wider knowledge.
As a concrete example, figure 16.2 shows how two different agents can have very differently structured perceptions of the same natural environment. In this subsistence simulation (Costopoulos 2001, 1999), agents create memories of resource extraction events, and these memories affect agents’ daily decisions about mobility. Figure 16.2 shows the number of purposeful movements by two agents exploiting the same environment over the same ten-year period. Since purposeful movement requires memories, the distribution of purposeful movement events gives an idea of the way in which the agent perceives (through its memory) the landscape in which it lives. In this case, agent 224’s cognitive landscape is relatively centralized and limited to a few locations, some of which are seldom visited. Using the same resource base in the same environment, agent 18 builds a more decentralized and diverse view of its landscape. It is clearly aware of more locations and uses them more often. Most of the difference between these two cognitive landscapes can be attributed to historical contingency. The order and timing in which the agents encountered resources that made an impression on them is partly responsible for the emergence of their individual cognitive landscape, and that order is in turn dependent on the agent’s location at the start of the run.
Figure 16.2. Distribution of purposeful movement events by two simulated agents exploiting the same environment over the same ten-year period.
Beyond these interactions between agents and environment, agents should interact with their social environment. In the instance above, for example, how would the cognitive landscapes of both agent 18 and agent 224 be modified by information exchange? Ideally, the interactions of agents with their social environment should be encoded in a language-like medium. Agents should have to invent ways of conveying meaning from one to another and should develop lexical systems over time. Two agents from different parts of a simulated social universe should normally have learned different ways of communicating.
The actions of each agent must be taken in response to its context, as filtered through the current state of its cognitive system (as Childe anticipated). Those actions must be perceived by other agents and become part of their context. Ideally, these literal interactions can lead to the emergence of superagents in the form of cohesive networks of agents with lower variance in viewpoints, goals, and beliefs than can be found in the social system as a whole.
Several such cohesive networks in a single system, each one centered on a different point and forming a relatively tight but permeable cloud in the multidimensional space of viewpoints, goals, and beliefs, could be construed as simulated societies. Those simulated societies, through rules for the translation of social action into material traces and the transformation of those traces over time, can produce a simulated archaeological record useful for the study of past social systems and their evolution.
The central problem of archaeological simulation remains the conjugation of the elements discussed above for the production of sufficient variability to meet the requirement of generality. Ideally, any social simulation should be able to produce an Icelandic blood feud, an Hxaro exchange system, a Roman republic, as well as a Ramayana. An archaeological simulation should be able to give us a taphonomically transformed record of each of those things.
Some will suggest that this sets the sights very high indeed. I will argue that it sets them high, but not too high. The simulation of social systems is an extremely intensive enterprise in terms of design and implementation time, equipment, running time, analysis of huge amounts of data, and so on. Anyone who has seriously considered its use can attest to the large investment even a simple simulation exercise entails. Under those conditions, only the highest possible return makes simulation a reasonable proposition. Whether the goal can in fact be reached or not, the return is the same: either the core assumptions of social simulation will be rejected, or they won’t. Each of those outcomes has a scientifically equivalent value.
The goal is not within sight, but there has been enough good work done on meeting each of the individual criteria outlined above for a powerful synthesis to be within our grasp. That synthesis, if it is effected, can start moving us toward the goal. Beyond simulation in demography, there is, after all, demographic simulation. There exists a core of theory and methods that all demographic simulations share (Buikstra and Konigsberg 1985). Any simulation in demography that is missing one of the central processes affecting population structure (births, deaths, migration) cannot qualify as demographic simulation. The same is true of flight simulation, and materials engineering simulation. A simulation without lift and gravity, while it may describe the flight-like movement of an object, is simply not a flight simulation. It does not generate flight.
There are currently no criteria by which simulations in archaeology are evaluated, and there need to be. There is no unified theoretical and methodological core to what should be a subdiscipline. The elements are there and have been worked on separately for at least thirty-five years.
If archaeological simulation is to exist in the same sense as demographic simulation, each simulation engine should, in one form or another, from its own perspective, meet at least the criteria outlined in this chapter. Then, out of the multitude of existing simulations in archaeology, we will see the emergence of archaeological simulation.
Barfield, Thomas (ed.). 1997. The dictionary of anthropology. Oxford: Blackwell.
Barth, Fredrik. 1966. Models of social organisation. Occasional Paper 23. London: Royal Anthropological Institute.
Binmore, Ken. 2001. How and why did fairness norms evolve? In Gary Runciman, ed., The origins of human social institutions, 149–170. Oxford: Oxford University Press.
Brues, Alice. 1954. Selection and polymorphism in the A-B-O blood groups. American Journal of Physical Anthropology 12: 559–597.
———. 1963. Stochastic tests of selection in the ABO blood groups. American Journal of Physical Anthropology 21: 287–299.
Buikstra, Jane, and Lyle Konigsberg. 1985. Paleodemography: Critiques and controversies. American Anthropologist 87: 316–333.
Coon, Carleton. 1977. Overview. Annual Review of Anthropology 6: 1–10.
Costopoulos, Andre. 2001. Evaluating the impact of increasing memory on agent behaviour: Adaptive patterns in an agent-based simulation of subsistence. Journal of Artificial Societies and Social Simulation 4(4). www.soc.surrey.ac.uk/JASSS/4/4/7.html.
———. 2002. Playful agents, inexorable process: Elements of a coherent theory of iteration in anthropological simulation. Archeologia e Calcolatori 13: 259–265.
Custer, J. F. 1992. A simulation study of plow zone excavation sample designs: How much is enough? North American Archaeologist 13: 263–279.
De Vos, Henk, and Evelien Zeggelink. 1994. The emergence of reciprocal altruism and group-living: An object-oriented simulation model of human social evolution. Social Science Information 33: 493–517.
Doran, James. 1970. Systems theory, computer simulations, and archaeology. World Archaeology 1: 289–298.
———. 1997. Distributed artificial intelligence and emergent social complexity. In James MacGlade and Sander van der Leeuw, eds., Time, process, and structured transformation in archaeology, 283–297. New York: Routledge.
Dornan, Jennifer. 2002. Agency and archaeology: Past, present, and future directions. Journal of Archaeological Method and Theory 9: 303–329.
Dyke, Bennet. 1981. Computer simulation in anthropology. Annual Review of Anthropology 10: 193–207.
Elliot, Kevin, D. Ellman, and Ian Hodder. 1978. The simulation of Neolithic axe dispersal in Britain. In Ian Hodder, ed., Simulations studies in archaeology, 79–87. Cambridge: Cambridge University Press.
Gilbert, John, and Eugene Hammel. 1966. Computer simulation and analysis of problems in kinship and social structure. American Anthropologist 68: 71–93.
Hazelwood, Lee, and James Steele. 2004. Spatial dynamics of human dispersals, constraints on modeling, and archaeological validation. Journal of Archaeological Science 31: 669–679.
Helmreich, Stefan. 1999. Digitizing “development”: Balinese water temples, complexity, and the politics of simulation. Critique of Anthropology 19: 249–265.
Kunstadter, Peter, Roald Buhler, Frederick Stephan, and Charles Westoff. 1963. Demographic variability and preferential marriage patterns. American Journal of Physical Anthropology 21: 511–519.
Lake, Mark. 2000. MAGICAL computer simulation of Mesolithic foraging on Islay. In Stephen Mithen, ed., Hunter-gatherer landscape archaeology: The Southern Hebrides Mesolithic Project, 1988–1998. Cambridge: McDonald Institute for Archaeological Research.
Lansing, Stephen. 2000. Foucault and the water temples: A reply to Helmreich. Critique of Anthropology 20: 309–327.
Leach, Edmund. 1973. Concluding address. In Colin Renfrew, ed., The explanation of culture change: Models in prehistory, 761–771. Pittsburgh: University of Pittsburgh Press.
MacArthur, Norma, Ian Saunders, and Richard Tweedie. 1976. Small population isolates: A micro-simulation study. Journal of the Polynesian Society 85: 307–326.
MacCluer, Jean, James Neel, and Napoleon Chagnon. 1971. Demographic structure of a primitive population: A simulation. American Journal of Physical Anthropology 35: 193–208.
May, David, and David Heer. 1968. Son survivorship motivation and family size in India: A computer simulation. Population Studies 22: 199–210.
Read, Dwight. 1998. Kinship-based demographic simulation of societal processes. Journal of Artificial Societies and Social Simulation 1(1). www.soc.surrey.ac.uk/JASSS/1/1/1.html.
Ridley, Jeanne, and Mindel Shepps. 1966. An analytic simulation model of human reproduction with demographic and biological components. Population Studies 19: 297–310.
Shennan, Stephen. 2002. Genes, memes, and human history. London: Thames & Hudson.
Steward, Julian. 1970. The foundations of basin-plateau Shoshonean society. In Earl Swanson, ed., Languages and cultures of western North America. Pocatello, Idaho: Caldwell.
Thomas, David. 1972. A computer simulation model of Great Basin Shoshonean subsistence and settlement patterns. In David Clarke, ed., Models in Archaeology, 671–704. London: Methuen.
Varien, Mark, and James Potter. 1997. Unpacking the discard equation: Simulating the accumulation of artifacts in the archaeological record. American Antiquity 62: 194–213.
Whallon, Robert. 1982. Variables and dimensions: The critical step in quantitative typology. In R. Whallon and J. A. Brown, eds., Essays on archaeological typology, 127–161. Evanston: Center for American Archaeology Press.
Wobst, Martin. 1974. Boundary conditions for Paleolithic social systems: A simulation approach. American Antiquity 39: 147–178.
Wright, Henry, and Melinda Zeder. 1977. The simulation of a linear exchange system under equilibrium conditions. In Timothy Early and Jonathan Ericson, eds., Exchange systems in prehistory, 233–253. New York: Academic.
Yorston, Ron, Vincent Gaffney, and Peter Reynolds. 1990. Simulation of artefact movement due to cultivation. Journal of Archaeological Science 17: 67–83.
Young, David. 2002. A new space-time computer simulation method for human migration. American Anthropologist 104: 138–158.
