An immediate goal of the article
Controlled Simulation of Marriage Systems, published in the online Journal of Artificial Societies and Social Simulation, is to attain valid inferences about actual marriage preferences or rules in a population. The article deals with comparison of actual marriage behavior in any given historical or contemporary population (in which suitable data are available) with simulated random marriage behavior given the demographic constraints in the population. It argues for the possibility, as a more distant goal, of a radical unification of social science theories about norms and behavior in this field that overcomes the divisions between:

  • behaviorist approaches that treat norms as statistically frequent behaviors;
  • subjective idealist approaches treating norms as typical beliefs of individuals or as common patterns inferred from survey questions, etc.; and
  • historical institutionalist approaches that emphasize the coherence of written law or case law, which may also act as a constraint on behavior, or as Durkheimian "social fact" impinging on individual behavior.
Applications of the approach to three case studies show how this unification can be realized (a synoptic outline of the article is only partially completed).

 

1. What makes this a simulation?

The succession of generations is central to the simulation model: as relatives marry in early generations, they lay down patterns of relationships that will define the kinship links of succeeding generations. The historically constructed network, including its demographic characteristics (factors such as size and composition of sibling sets, given legal prohibitions on close blood marriages), affects the likelihood of behaviors in subsequent generations, e.g., that marriages will be with close kin, distant kin, or non-kin, or with particular types of blood kin (e.g., marriage of a man with MoBrDa).
The simulation takes the historical constraints into account in constructing a random baseline model of behavior, the trick being that it is only the behavior in successively "current" generation that is randomly permuted in the baseline model, with the historical network structure and its demographics held constant. Simulation results thus provide a controlled comparison with actual behaviors in each generation, in that both have in common the same set of constraints in the form of the historically prior network structure.

 

2. What are the audiences for this paper?

For multi-agent simulation modeling, complexity theory and readers of JASSS, this approach is complementary to simulations with agent rules and strategies (see for example Dwight Read's simulations in JASSS1.1) in using permutation-tests to find evidence for actual strategies and rule governed behavior in empirical case studies. This approach is concerned with the highly-probable worlds rather than the possible worlds end of the continuum in simulation modeling.

For sociology (e.g., historical) the arguments concerning global structures of large-scale cohesive social networks are particularly relevant to the study of social class formation, elite studies, and the boundary conditions of large-scale social groupings (Sections 7, 8, 10, 11 cases 1 & 3, 12.1/3, 15, 16-18).

For anthropology the identification of local structures (Sections 13) as well as certain global structures (such as dual organization, Section 14) is of interest to the study of social structure, alliance theories, and Lvi-Straussian or structuralist ideas about kinship, with the caveat that the historical simulation approach is capable of identifying discrete epochs in which structural change occurs, and is thus concerned with historical dynamics.

For economics, the new institutional economics recognizes the fundamental importance of norms, and takes "rules" as equivalent to the institutional framework of individual decision-making. What is still left out in this approach is the formation of differential structural positions and emergent groups within the network. It is not just individuals whose actions have consequences.

3. Why are social rules seen as problematic?

Anthropologists, economists (e.g., in game theory or rational choice models), and sociologists pay great attention to social norms and rules, both in terms of what people say and what they do, which are often discrepant (as may be the case as well with legal norms vs. actual behavior). Individual preferences and social "rules," when put into practice, are subject to real-world constraints, unlike idealized subjective statements of norms. This study questions the validity of a common habit of social science studies: to take the raw frequencies of behavior as the relevant evidence for inferring rules or preferences. Looking at what people do, is it even possible, in principle, to find evidence of rules or preferences? The answer being: yes, insofar as we take constraints into account, behavior that follows rules or preferences can be inferred if (1) it departs from what would be expected from "random" behavior under given similar constraints (in this case not only demographic but also legal, as when marriage prohibitions are imposed on the simulation model), and (2) there is empirical evidence supporting the mechanism by which preferential behaviors are implemented.

 

4. What is the relevance of the findings?

Research findings in this field to date have led to two theoretical camps. On one side, studies of social norms are often derided because of the lack of fit with actual behavior. On the other, studies of actual behavior are derided as impossibly complex, in favor of subjective norms or rules ("ideal models").

The research results from the case studies in the present article suggest radically different resolutions of this impasse.

(1) Taking constraints into account to model the preferential or rule-governed component shows a much closer fit to normative models than hitherto recognized in kinship studies and theories of matrimonial alliance.

(2) The relevant "normative rules" or models, however, are rarely those identified by subjective normative theorists!

The bottom line: both camps are wrong, and a much higher level of social science modeling is possible. In this approach, a greater degree of convergence is seen of certain expressed rules or norms with preferential behavior. With this kind of triangulation of research results, we are able to come up with much better models of social or marriage systems than either the behaviorist or the idealist camps on their own. One of the case studies (of a village in the historical kingdom of Kandyan - with an emergent dual organization) also suggests a good fit with legal norms identified by the historical institutionalist approach, although it will not in general be the case that written law is sociologically efficacious (but coherence in institutional case law would be expected to exert itself in constraints on individual behavior).

 

5. Does the approach generalize to other network problems?

Many aspects of this present approach, in a more general context of modeling network behavior under constraint, are already in use in network studies. Some recent examples in the domain of friendship networks include:

Zeggelink, Evelyn P.H. 1993. Strangers into Friends: the Evolution of Friendship Networks Using an Individual Oriented Modeling Approach. Amsterdam: ICS.

Moody, James. 1999. The Structure of Adolescent Relations: Modeling Friendship in Dynamic Social Settings. Ph.D. Dissertation: Department of Sociology, University of North Carolina.

The null hypothesis model of "random behavior under known constraints" is important to network studies and essential to valid statistical inferences about the components of behavior.

 

6. What are the implications and the next steps in this research?

The application of simulation models to the particular problem of population studies -- kinship and marriage networks in the context of history, ethnography, elite studies, studies of social class, etc. is currently being extended to ask questions about "complex system" dynamics and the phenomena of institutional emergence and change due to network processes and critical morphological or density thresholds. These questions require a broader class of probabilistic simulation modeling involving social and demographic constraints, rules and agents in the general family of network heterarchy.

This research was supported under NSF Grant BCS-9978282.

 

Return to methodological abstracts 1997-2000