Sociology Laboratory
Bill Bainbridge's Sociology Laboratory: Computer Simulations for Learning Sociology (Belmont, CA: Wadsworth, 1987), was intended to be used in lower division undergraduate courses. Despite this, the simulations are theoretically sophisticated, and use many modeling approaches that are considered "state of the art" a decade after the book was published. Rational choice, exchange, and network ideas are used in many models. Both agent-based and systems types of models are used (though, primarily agent-based). And there is an emphasis on some of the evolutionary and learning processes that have become important in later work.
Below, we provide a very brief introduction to each of the models. Bainbridge's book is no longer in print, and you may wish to use this page as a guide to the parameters and settings called for in each of the simulations.
return to the index of this page
Description:
This simulation consists of 46 agents, each of which has two continuously measured traits (smart-dumb, beautiful-ugly). At the beginning of the simulation, traits are randomly assigned to agents. As the simulation proceeds, pairs of actors are chosen at random. Their scores on the two traits are examined, and an exchange of one trait is made if the exchange would move the entire population toward a criterion correlation between the two variables. The experimenter can set the criterion. A correlation of -1.0, for example, would mean that all "beautiful" agents were "dumb."
This model might be thought of as one in which there was a perfect selection pressure operating, and the occurance of interactions of agents was purely random. The agents are also very simple, having only two traits. There is no "mutation" in this model, but only a form of "cross-over." The model can be thought of as an implementation of an evolutionary mechanism of a sort, that operates on pairs of actors (e.g. pairs of people, interacting organizations, etc.) to produce change in the aggregate distribution of traits in the population. The model is one type of evolutionary "population ecology."
Will a model with this sort of algorithm arrive at a steady-state equilibrium? Will that equilibrium be at the goal level set? Will the movement of the population toward the goal be smooth and non-reversing? How long will the process take to reach it's steady state, if it has one. Does the length of time that is taken to reach the goal depend on the goal? In what way? Run a number of "trials" (repititions) with the same goal state. Record the number of exchanges and trails necessary to reach the goal. Change the goal state, and repeat the experiments until you are comfortable with understanding what is going on. Are the results always identical? Why or why not? What does the distribution of outcomes look like (is it normal or skewed).
Bainbridge suggests that this simulation be used to introduce the notions of status markers and status inconsistency. Status inconsistency is seen as a population or structural concept at the macro level. A agent who is smart and ugly (assuming positive correlation between "smartness" and "beauty" is normative) has "inconsistent" statuses in a sense. But, in the Bainbridge model, a person is status inconsistent only if there exists another individual with whom they could exchange a trait to make the overall population correlation between traits stronger.
The simulation program displays:
You will see a scatter-plot of the scores of the cases on the two traits (you may identify the more beautiful half of the cases with a different symbol (menu item: identify), or cause their data points to blink (menu item: blink). You will want to see the value of the population correlation coefficient, and the number of exchanges that have occurred (menu item: analysis).
You may set the following controls:
In addition to the menu items controlling displays, you may use the randomize the scores of actors to begin the simulation (menu item: randomize), and advance the simulation by different amounts of iterations (menu item: proceed).
The only experimental parameter in the system is the goal you set for the population correlation between the
two traits; this is accessed through the menu item: correlation.
Think about:
Can you think of any other sets of labels for the traits of the actors that fit with your areas of interest?
Can you think of any real-world cases where an exchange mechanism or algorithim like the one in this model might actually be operating?
What kinds of changes might you want to make to the algorithms in this model to make it a more useful description of proceses by with the distribution of traits in populations evolve?
return to the index of this page
This is Bainbridge's implementation of a very important early model -- Schelling's (Micromotives and Macro Behavior) study of the dynamics of segregation of populations. The model shows how individual actions, based on local information only, can result in the emergence of macro-patterns (in this case, residential segregation). The model is one of the class of "cellular automata" in that it is an algorithm for changing the traits of locations in a lattice based on the values of those in the "neighborhood" of each point.
Description:
The simulation consists of an 8x8 lattice, with each cell having the value of either "green" or "blue" (that is, a single trait), with four empty cells. The simulation proceeds by having a randomly selected actor examine it's local neighborhood (the up to 8 adjacent cells) and apply one of a set of rules. For example, if a "blue" actor sees that a majority of its neighbors are "green," it will move to a randomly selected empty cell. A round consists of all of the actors having an opportunity to move; the simulation proceeds over as many rounds as you would like.
The simulation program displays:
You are shown the spatial map of your labeled actors after each round. For each round, the number of moves that occur is reported, as well as a measure of goodness of fit: the proportion of each population that are in neighborhoods that satisfy their preferences is reported.
You may set the following controls:
You may experiment with this simulation by selecting different decision rules, the number of actors (and hence, vacant locations), and the composition of the population (how many blues and greens). The decision rules may be different for the two "races." Each race may have "attitudes that range from exclusion of members of the other group from the actor's neighborhood, through indifference, to a mild desire to be in a neighborhood dominated by members of the other group.
Think about:
Seven different preferences (movement rules) are provided for each group. You should think through what you expect will happen for each of the 49 possible combinations.
What do you expect the effects of the amount of "open space" in the simulation will be? Why?
What do you think the effects of different sizes of the two populations might be? Will the effects of population composition be the same, regardless of decision rules?
Do you think that the simulation will reach an equilbrium (that is, a state where there is little or no net change in the pattern of locations over successive rounds)?
Think of one or more applications from your areas of research interest that might be described by something resembling the rules in this simulation.
How might you want to modify this simulation to make it more applicable? What kinds of complexities might you add? What are your hypotheses about the effects of such modifications?
The Schelling model shows how micro-motives (the desires of individual actors) can give rise to macro-structures (stable "communities" of unmoving actors; cyclical alteration or migrations of populations; or more "chaotic" behavior). Do you find this convincing as an explanation of where segregration and other macro patterns come from? Why or why not?
return to the index of this page
Description:
This simulation is in the "systems" tradition, rather than the "agent-based" tradition. It is designed to allow the user to explore the basic principles of population demography. That is, it is concerned with the age-sex composition of population, population size, births, deaths, and immigration. The user may choose different initial population structures, and experiment with differing rates of births, deaths, and immigration.
The simulation program displays:
You may display the age-sex "population pyramid" of your experimental society either as a graphic, or as a table. The simulation steps forward in time by 5-year increments, and you may observe how the population age-sex structure evolves under different sets of assumptions about birth, death, and immigration rates.
You may set the following controls:
You may initialize the simulation in four stereotypical demographic structures (traditional, transitional, industrial, post-industrial). There are also five "test pattern" structures that may be selected. You may also set fertility, mortality, and immigration at levels from "very low" to "very high." These parameters, as well as relative age-sex specific birth, death, and immigration rates remain constant throughout the simulation.
Think about:
One useful simulation is to select a traditional initial condition, but assume that there has been an intervention to reduce mortality (e.g. increased food, better medical care, education, etc.). Examine the behavior of this scenario, versus the baseline traditional society.
Under what conditions is total population stable? Under what conditions is the age-sex structure of a population stable? What are the characteristic patterns of outcomes for this model? What determines which pattern will be observed in a given realization of the model?
How might this model be improved? What aspects of it are particularly unrealistic?
Think about a population of organizations (say "generalist" and "specialist" types). Does this sort of demographic model make sense for studying the dynamics of populations of social structures, as well as of individual persons? How would this model need to be modified to study organizations?
Think about the age-sex structure of a population of interest to you -- e.g. prison populations, social movement participants, etc. -- is the population pyramid a useful tool?
return to the index of this page
Description:
The study of occupational mobility across generations has been a staple of stratification research. Formal models, both statistical and mathematical, are common approaches for understanding and theorizing systems of social mobility. Bainbridge's model is, again, in the "systems" tradition. The model is a very simplified one, but has the same basic architecture as most mobility models. In the model there are four occupations (farmer, bureaucrat, worker, professional). Each young person chooses one of the four on reaching adulthood (age 21) and remains in that occupation until retirement (age 67). The action in the model comes from different theories of how persons choose occupations.
The simulation program displays:
You may see either a graphical or a tabular display of results. The results show the evolution of the occupational structure from year to year as the various sets of controls are applied iteratively.
You may set the following controls:
You may select one of a set of alternative initial occupational distributions of the population. You may specify the relative importance of three sets of factors that affect occupational choice: parental occupation, monetary reward, and random factors. The algortihm for the effect of parental occupation is very simple -- the child enters the same occupation as the parent, with a probability equal to the weight you decide to give to this factor. The monetary incentive effect presents a number of alternative theories. In each theory, the wages in a occupation are determined by the total output of the sector (which is fixed), divided by the number of persons in the sector. You may select different scenarios, however, about the relative outputs of the various sectors. There are two alternative theories of the stochastic component of occupational choice: either each outcome is equally likely, or choices are random, but proportional to the existing occupational structure.
Think about:
As with all the other simulations, you need to spend some time running different scenarios to get a feel for the characteristic behaviors of this model. Explore the effects of the two alternative theories of stochastic effects; explore different scenarios about the relative rewards of occupations. See whether chance factors can lead to unstable outcomes under some circumstances. Does the model always reach an equilibrium? Does it reach a cyclical equilibrium? Does it produce a chaotic outcome?
This is a pretty unrealistic model of actual occupational mobility dynamics -- which the research literature have found to be quite complicated. How might you modify the model to make it somewhat more realistic, without making it too complex to understand?
Many problems are usefully thought of as mobility between states. How might you apply this kind of model to
a problem in your area of substantive interest? What would the states be? What factors would determine the rates
of movement from one state to another?
return to the index of this page
Description:
The "suicide" model is not really a dynamic model at all (in the sense that it does not operate to show a historical unfolding over time, based on iteration). Rather, as Bainbridge describes, it is a fancy calculator. The purpose of the calculation is to show how differing relative risks of events (suicide, in this case) from multiple factors result in differing prevailence. The basic idea is rather like that of a multiple-decrement life table analysis from demography. It is useful to play with this simulation, however. The program itself is really quite clever. And, the model does illustrate how systems with multiple causes can produce the same outcomes (or nearly so) in a wide variety of ways. In a sense, a cautionary tail about trying to reason backwards from observed differences in rates to "discover" the underlying causal model.
The simulation program displays:
You are given a population of 1,000,000 persons. You may sub-divide the population by four (of five) possible criterea: sex, marital status, political stability, migration rates, and religion. In this simulation, each choice is dichotemous. By selecting the relative risk of suicide as a result of each of these variables (e.g. migration is more important than political instability) and the relative risks (e.g. the risk of suicide of males is twice that of females), you can select a scenario of the parameters affecting suicide rates. The simulation then calculates the number of persons in each group, the numbers of suicides, and the rate.
You may set the following controls:
You may select the sizes of populations for each cross-break. You may select the order of importance of the variables (e.g. is marital status or sex more important). You may select the relative risk for falling in one category or the other of each risk factor (e.g. being married reduces risk by 50%, relative to being unmarried.
Think about:
This is a useful simulation for understanding the relationships among relative risks and outcomes. It is also helpful for thinking about how multiple causal factors might be included in equations describing causal processes. As a dynamic model, it is not particularly useful.
How might you go about turning this static representation into a dynamic model?
return to the index of this page
Description:
This model is an agent-based model that uses many of the main ideas of "genetic algorithms." It is intended to explore some of the dynamics of basic socio-biological evolutionary theories. The model is worth a good bit of study because it translates some of the ideas of sociobiological thinking into a formal model where we can explore the implications of these inherently dynamic theories. It is also worth a good bit of study as a (fairly simple) example of the ideas of "genetic models" (i.e. models where agents are represented as a list or string of traits); and, genetic algorithms (i.e. selection of phenotypes or genotypes, mutation, and cross-over).
We begin with a population of two actors (one male, one female) of the same species. The population reproduces by hetro-sexual mating, with each mating producing one offspring. The environment has a carrying capacity of 50 males and 50 females; when this limit is reached, excess agents "migrate" out of the field. All agents age and, eventually, die (if they are not killed first).
Each actor is decribed by four traits ("genes") that describe their genotype with respect to: sex, aggressiveness, deceitfulness, and species. At each mating, the offspring's traits are determined by selection (random?) from those of the parent, plus mutation (if allowed by the experimenter). Sex is not subject to mutation, and is simply randomly selected for each offspring. A new species may evolve, if the investigator allows this possibility, by mutation. The model allows only two species, and they may not mate.
Bainbridge's substantive interest is in the two other traits: aggressiveness and deceitfulness (my labels, not his). Aggressiveness is genetically carried by both sexes, but realized phenotypically only in the behavior of males. Deceitfulness is genetically carried by both sexes, but is realized phenotypically only in the behavior of females. Males may be non-aggressive (the starting point for the simulation), "attackers" or "superattackers." Attackers will attack and kill non-aggressive males; super attackers will attack and kill non-aggressives, aggressives, and one another. Deceitful females ("trappers") give false signals about their species, and kill males of the other species who attempt to mate with them. Non-deceitful females always give true signals about their species.
The simulation program displays:
You are given a graphic display of the genetic code of each of the actors (sex, aggressiveness, deceitfulness, species) occupying each of the 100 locations in the environment. The simulation begins with two actors of different sex, same species, non-aggressive, and non-deceitful). The number of deaths and migrations is reported as the simulation proceeds. At each step in the simulation, you may request (menu: Information) data on the current composition of the population gene pool.
You may set the following controls:
You may select which traits (aggressiveness, deceitfulness, species) mutate, and the rate of mutation. You may also control the average natural death rate (which is assumed to be the same for all genotypes).
Think about:
The simulation can, most obviously, be used to explore the logic of claims of sociobiologists about the origins of aggressiveness in males. Is aggressiveness favored? Is this always so, or are there some conditions under which it is not a reproductive advantage? Does a favored trait completely drive a less favored trait out of the population by selection?
What is the effect, if any, of the deceitfulness gene?
Is this a useful model for evaluating the logic of sociobiological claims? What modifications of the model might be useful?
Can you think of any applications of the mutation-selection logic to the evolution of the genotypes and phenotypes in populations of organizations or other social structures? How would the algorithm have to be modified to deal with non-biological evolution?
return to the index of this page
Description:
This model grows out of Bainbridge's work on scientific creativity and the effect of network communication structures on innovation. It draws on a number of ideas from the structural (networks) sociology of organizations.
The simulation is an algorithm to find the solution to a problem -- to guess a word or text string that you have picked. This task is undertaken by a set of actors. At each round of the simulation, actors who are inventors seek to identify letters of your word by guessing. If they hit one, they attempt to communicate this finding to the others. As letters are discovered, guesses are made about their order. Again, once a correct guess is made, it is retained and transmitted to others. The simulation ends when the answer has been found and communicated to all members of the team.
The simulation is intended to explore the effects of several variables on the success of the team. One is size: five to nine agents can be used. Another is the network of communication -- not all members of the teams communicate easily with all other members. Another factor is the division of labor. Some actors are better original thinkers (inventors), others are better communicators. Lastly, there is organizational culture. In one scenario, actors use "conventional" thinking (e.g. institutionalism?), in the other, they use a more abstract analytic logic (e.g. "rationality"?).
The model is an interesting one in that it allows for effects of actor characteristics, as well as the embedding of the actors in a network to affect the outcome. The embedding notion is an important one, in that it suggests the possibility that the effects of the characteristics of actors on the behavior of the macro-structure may be contingent on the structure of ties.
The simulation program displays:
You may experiment with one group, or set two groups in competition to solve the same problem. You will see a graphical map of the actors in each group (rectangles), their characteristics (whether an inventor or a transmittor), and the actors to whom each has strong ties (double lines). You will also see the knowledge of the solution that each inventor has at each point in the simulation, displayed in the rectangle.
You may set the following controls:
You may select to study one or two groups. You may then select among a number of pre-set configurations about the structure and size of the networks. Also, you may select a "biased" culture (one in which actors guess letters according to how commonly they occur in everyday speach) or an "even" culture (one in which actors guess letters randomly with equal probability). You may select different divisions of labor -- different configurations of "inventors" and "transmitters." Not every possible configuration is possible, but many very interesting ones are.
Think about:
This is a good simulation to use for exploring the utility of experimental design in studying results. Hold all but one factor constant, and vary the remaining factor across it's range of values. Repeat each a number of times, with the same, and with different stimuli. Fully exploring the range of behavior available in even a pretty simple simulation can be pretty tedious. Got any better ideas about how to do this more efficiently?
What matters more? culture, structure, or division of labor? Or, does it depend?
Is this model a reasonable starting point for studying the problem solving dynamics of small groups? How might you modify this to make it more useful?
Are there any phenomena in your area of study that might reasonably be thought of as analogs to this model?
return to the index of this page
Description:
Morals is a network simulation designed to highlight two two sociological theories of how communities form. The logic of the theories of community organization is really the same thing as "autopoesis" or "self-organizing" systems from chaos and complexity theory. That is, local action by randomly distributed actors may give rise to clear and obvious patterning - though the processes are highly stochastic and chaotic. The resulting "structures" may be stable, cyclical, or semi-periodic.
The model is a rectangular lattice (not wrap-around) occupied by 190 agents. Non-edge actors have local neighborhoods of the 6 surrounding actors, and may have ties to none, any, or all of their neighbors. Each agent is characterized by three traits: drug user or not; thief or not; and member of a religious sect or not.
At each cycle of the simulation, a random actor is chosen, and one or the other, or both of two algorithms are applied:
Differential association: if this algorithm is used, the focal actor surveys the traits of the actors to whom it has ties, and adopts the trait that characterizes the majority of them.
Balance theory: two alternative versions are available. The "trait" setting causes a focal actor to sever existing ties with actors to whom the focal actor is dissimilar, and form ties with those to whom the focal actor is similar; the "balance" setting causes the focal actor to examine local triads, and form or drop ties to form balanced triads (e.g. focal actor A has a tie to B and a tie to C, but B and C are not tied; the focal actor will break ties with B or with C; focal actor A has ties with B, but not with C; B and C are tied; actor A will form a tie with C).
Following his long-standing interest in religion, Bainbridge adds one more complexity to the model. You may select to have a religious sect present. If you do, it may have members who follow the same algorthms as other folks (weak religion), who are "evangelical" in that they form ties in defiance of the principles of balance theory ("strong" religon), or who are particularly prone to form ties with drug users ("anti-drug" religion).
This model uses neighborhood and/or network approaches to embed actors. It demonstrates the possiblities of self-organization of larger structures (stable neighborhoods or sub-cultures) from simple local action. It also is rather fun to play with, and can generate some rather surprising patterns.
The simulation program displays:
You will see a graphical display of the 190 verticies of the lattice. Each point will be characterized as D(rug user), S(tealer), and/or REL(igious). The ties of actors are represented by double lines.
Using the Information menu, you may see a census of the population at each round (actually, steps of 100 or more iterations are useful), and the triad census. The triad census is an important measure of density and stability from graph theory.
You may set the following controls:
You may select the density of the network ties (which are randomly assigned at initialization). You select the number of drug users and thieves (explore if there are minimum numbers necessary for these traits to persist; whether there are densities that always lead to everyone adopting the traits). You elect whether to apply differential association rules, and balance theory rules (balance or trait). You select whether a religious sect is present, and, if so, it's strength. You also may select a mild or addictive drug (this affects the probablility that a user will change their trait under the influence of association).
Think about:
There is a lot to experiment with here. Explore the effects of differential association all by itself, and the effects of the balance algorithms all by themselves. Get a sense of the characteristic behavior of the model at different initial densities of ties and traits. Then work with both algorithms operating simultaneously. Are the results what you expected? Does tie density and initial population composition matter?
As with the other models, think about how this model might be modified to be a more useful and realistic model of sub-culture structural dynamics. Are the effects of embedding too simple? How would one deal with multiple traits? What other kinds of algorithms might be added?
As with the other models, think about research problems in your particular specialty areas. The ideas of sub-cultures, differential association, and balance are used extensively in most substantive research areas. How might Bainbridge's model be modified to fit the special characteristics of interesting problems in your research area? Make a proposal (maybe Bainbridge -- head of NSF -- will fund you!).
return to the index of this page
Description:
This model is an interesting application of network diffusion. It is an agent-based model (really a network of 120 nodes), and focuses attention on the effects of network structures on multi-state diffusion processes. It shares a good bit in common with the laboratory simulation and the morals simulation.
We examine the political affiliations of 120 persons in a village. Each node has up to eight adjacent neighbors
to whom they are tied. Each actor is affiliated with one of five political parties: (left) communists, socialists,
moderates, nationalists, nazis (right). Actors are influenced, at each round, by the affiliations of those to whom
they are tied -- in this simulation, ties are not made and broken, but are set from the initialization. Bainbridge
does not describe the algorithm by which neighbor influence operates -- it is, most likely, an averaging process.
Actors may switch according to one of three rules: equally likely to switch to any other party, likely to move
to an adjacent (left-right dimension) party, or move to adjacent party or nazis (who have universal appeal). The
social class of actors may influence their choices, and the economic performance of the village may be turned up
or down to accelerate or decelerate the rates of changes.
The simulation program displays:
You are given a graphical map of the traits and ties of actors, and may request a census of the affiliations.
You may set the following controls:
You select, first, the overall average network density. Next, you select the network shape (open, two communities,
three social classes, communities and classes, or 30 cliques). Next, select a party switching theory. Then, decide
on what effect social class has from a range of ranked alternatives (from no effect, through only effects at the
top and bottom of the class stratification system, to all class effects). Lastly, you may input an exogeneous
time series of the performance of the economy -- measured in the percentage of the population that is satisfied.
This enable the user to experiment with the effects of exogenous shocks -- often a very useful simulation tool.
Lastly, you select the initial distribution of party affilliations.
Think about:
This simulation shares much with the "morals" simulation, but allows a more interesting set of variations on the effects of network structures on diffusion. See if you can understand what leads to different characteristic behaviors of stable segmented, unstable and cyclical movements, dominance by one group, or stable equilibria. Can you find model settings that produce chaotic behavior?
As always, critique the model and suggest modifications for more realistic study of political party affiliation dynamics
As always, see if you can come up with a creative application of this model (or one rather like it) to your
special area of interest.
return to the index of this page
Description:
The last three simulations (J, K, L) are variations on rational-choice and exchange dynamics -- a frequent theme in the social science modeling community -- particularly in Political Science. The three models are increasingly complex. The first is a simple 3-actor game (simple, but not entirely obvious), where actors learn to select partners from past patterns of outcomes. The second is a classical prisoner's dilemma (or other such) game, but played with 12 actors, and a selection of strategies (this is a very useful simulation for game theorists). The last model adds efforts of the actors to discover underlying patterns and order, and allows experiments with different patterns of communication and solidarity among the twenty-four actors. The models should be studied from simpler to more complex.
All rational-choice, exchange, prisoner's dilemma game type models begin with an assumption of rational individual maximumization as the characteristic orientation of all actors. They seek to understand how stable social order, norms, and institutions can arise by self-organization out of the initial choas of such unstructured situations. The keys are often in how actors learn, the strategies that they use for deciding with whom to exchange, and what exchange strategies they use.
Bainbridge is particularly interested in representing the decision-making structures of actors, and the models try to show how actors store information about the environment, update this information by learning from exchanges, and use the information for strategic choices. Although Bainbridge's models do not, genetic algorithms have also been applied by some to the learning behavior of actors in such models.
In the first simulation, we have a "war of all against all." Each of three actors is distrustful of each the others, and resources are transferred by siezure.
The simulation program displays:
In this simulation, actors seek to steal resources from one another. There are two resource types: energy and supplies. Each actor wishes to maximize their stock of supplies. However, to do so, they must use up energy. On each round, an actor may elect to steal energy or supplies, and may elect one or the other node as a target. The success of each actor on each round is a function of their skill, a random component, their power (the more powerful the actor, the more resources they extract), and the resources available (the bigger the stock raided, the bigger the take).
You may track the progress of the simulation with a table that shows the events of each round, and the existing
stocks of each of the actors. A chart mode allows you to see the history of the distributions of the two resources
over the three actors over time.
You may set the following controls:
You may initialize each actor's initial resource levels (energy and supplies). You may also select differences
in actor skill. You may also elect different sets of rules governing actors behavior. The first set of choices
concerns whether actors are more likely to seek energy or supplies in their raids; the second set concerns how
actors form their expectations about the likely outcomes of raids on the other players; the third choice concerns
whether actors behavior is deterministic (always raid where the greater reward is expected) or stochastic (select
a target with probability proportional to expected outcomes).
Think about:
Does the war of all against all inevitably lead to the permanent dominance of one over the others? Does the model always result in the destruction of all actors (energy = zero)?
Are there any real-world scenarios that you can think of that are characterized by the sort of pure opportunism
built into this model?
return to the index of this page
Description:
The second model in the series is fundamentally different, in that it allows for reciprocal exchange, rather than simple theft. However, actors may elect to be opportunistic by offering a bargain, and then "defecting" from the bargain.
The Deimos simulation is a classic "prisoner's dilemma" game (actually, you may set different payoff matricies to play different types of game theoretic scenarios). Here, however, the game is played with a population of 12 actors (rather than classical two-person game theory). Actors also have the option to refuse an exchange -- a modification found very productive and useful by later game theorists.
The main action of the simulation is in the selection of actor's strategies. Five of the six available strategies are fixed, but the sixth allows actors to learn which of two strategies has higher utility.
The simulation program displays:
You may track the progress of the game either with tables that show each actor's resource levels, or with graphics showing refusals, exchanges, and payoffs.
You may set the following controls:
The key things to play with in this simulation are the five basic strategies: "always cooperate" "always defect" "random" "cooperate then tit-for-tat" and a modification of tit-for-tat that adds 20% random defection.
Once you get a sense for these, examine whether outcomes change is actors are allowed to learn whether past interactions with specific others have been rewarding or not (one learning theory modification that is often added to iterated prisoner's dilemmas). Examine the effects of giving actors the right to refuse to exchange with those who have ripped them off in the last exchange.
Think about:
It is difficult to overstate the importance of the prisoner's dilemma game and it's major modifications, to the rational choice and exchange theory schools. The simulation here allows you to explore most of the major variations on strategy (but less on learning and embedding). Unlike most games, there are more than two actors in this simulation, and the outcome patterns can be quite rich and complex.
Consider how the game models may be too simple in representing iterated exchange games. How might you modify them, while keeping the basic ideas of individual utility maximization.
Again, see if you can apply a model of this type to a problem in your area of research interest. Iterated exchanges
would seem to be quite a common feature of most social structures, and, with a little imagination, you should be
able to come up with some interesting applications.
return to the index of this page
Description:
This is the last of Bainbridge's series of exchange models, and is the most complex. In many ways, it is very similar to the later work of Axtel and others on artificial societies. Bainbridge puts a particular emphasis on the examining the role of belief (religion) and communication as factors in the emergence of stable exchange relations. Other modelers seek to produce order in exchange, and to show the development of norms and culture without these complexities.
We have a population of 24 actors. The actors are differentiated into two, three, or four types (according to what resources they are capable of producing). However, no actor knows the type of any other actor, how many types there actually are, and what distinguishes the types of actors. As the actors play out their exchanges, one of the major outcomes is that they try to learn about the social structure (how many groups? what does each group do?) so that they can make choices of exchange partners that are more likely to result in successful outcomes.
Each actor must obtain energy, water, food, and air. Each actor, according to their type, may produce one or more of these commodities. Bainbridge allows you to add a "meta-commodity" of personal salvation -- held either by joint belief, or by a charismatic leader, if you would like.
At each round, actors seek exchange partners who can provide what they have least of, in return for what they
have most of. The payoffs from their exchanges are proportional to the fit of their demand and supply schedules
with their exchange partner. With each exchange, the actor learns about the partner, and seeks to build a general
model of groups, their characteristics, and the location of other actors in the social landscape. The more successful
the actors are in learning, the more productive their exchanges. In order for exchanges to occur, there must be
excess resources. The simulation allows for periods of "resource making" between rounds of exchanges
as a way of dealing with this.
The simulation program displays:
The simulation is complex, and a lot of data. The graphical map shows exchanges occuring, and the resource accumulations as they change over time. You may view the output for each actor, or mean levels for the groups.
You may set the following controls:
The controls are rather simple. The user must select the number of underlying groups, and the mix of productive
capacities of each. The user may also elect to study the effects of religion (none, social shared belief, or charismatic
leader), and the pattern of communication among the actors.
Think about:
The key insight of this simulation is in the more complex model of learning. Here, actors seek to develop a cognitive map of the social differentiation of their fields in the abstract, as well as to form expectations about particular exchange partners. Study, particularly, how the results differ when the social structures that are being learned differ.
return to the index of this page