COMMUNICATION NETWORKS, COMPLEX INTERDEPENDENCE, AND THE SURVIVAL OF DISAGREEMENT AMONG CITIZENS

COMMUNICATION NETWORKS, COMPLEX INTERDEPENDENCE, AND THE SURVIVAL OF DISAGREEMENT AMONG CITIZENS A RESEARCH PROPOSAL Robert Huckfeldt Department of Political Science Indiana University, Bloomington and Paul E. Johnson Department of Government University of Kansas, Lawrence 2001 Robert Huckfeldt and Paul E. Johnson DO NOT QUOTE WITHOUT PERMISSION Presented at a colloquium at the Workshop in Political Theory and Policy Analysis, Indiana University, Bloomington, on February 12, 2001.

COMMUNICATION NETWORKS, COMPLEX INTERDEPENDENCE, AND THE SURVIVAL OF DISAGREEMENT AMONG CITIZENS The survival of disagreement among and between citizens constitutes an enduring puzzle in the study of democratic politics. On the one hand, a compelling set of arguments suggests that political activation tends to eliminate political disagreement. In their classic study of the 1948 election, Berelson, Lazarsfeld, and McPhee (1954: chapter 7) argue that individuals naturally turn their attention to political issues and controversies in politically high stimulus settings. Politics becomes a frequent topic of conversation, disagreement becomes socially visible, and social conformity processes produce agreement and homogeneity within small groups. According to this argument, communication leads to agreement, and political conformity becomes the dominant condition within closely held networks of political communication. Indeed, more recent agent-based simulations of communication and diffusion suggest that homogeneity is the likely outcome of such diffusion processes (Axelrod 1997), and the preservation of heterogeneity and disagreement is frequently rendered problematic (Johnson 1999). At the same time, a body of accumulated evidence suggests that political disagreement is a frequent occurrence in democratic politics, even within the smallest and most closely held social groups (Huckfeldt and Sprague 1995; Huckfeldt et al. 1998a). While the political preferences of citizens tend to reflect the partisan composition of their micro-environmental surroundings, relatively few citizens reside in politically homogeneous social worlds, protected from the immediate social experience of people holding divergent political viewpoints. Indeed, the experience of disagreement is perhaps the modal condition among politically engaged citizens. And this simple fact, the survival of disagreement, forces a reassessment of the models and mechanisms of communication and influence among citizens. The outcome of such a reassessment is important because a central issue in the study of democratic politics is the capacity of citizens and electorates for tolerating and responding to political disagreement in a creative manner (Barber 1984; Fishkin 1991). The ideal of a free, open, and democratic society is one in which political issues are fully explored and debates are fully aired. In such a society, citizens are open to persuasion, the social boundaries on political viewpoints are fluid and shifting, and individuals encounter the full spectrum of issue positions and political viewpoints. How does this vision of a democratic society correspond to contemporary analyses of citizens and democratic politics? At one analytic extreme, citizens play the role of individually autonomous actors, oblivious to the experience of political disagreement. Individual preferences inform individual choices, and these preferences are idiosyncratic to particular, socially self-contained individuals. Thus, the preferences and choices of one person become irrelevant to the preferences and choices of another, and political disagreement among citizens becomes irrelevant to political outcomes. At an opposite extreme, inspired by the power of social conformity effects (Asch 1955,1956), citizens are sometimes seen as being powerless in the face of an irresistible social influence process. The psychic discomfort of disagreement causes individuals to reduce dissonance through various means (Festinger 1957). In particular, individuals adopt or selectively misperceive prevalent viewpoints and, just as important, they avoid disagreement in the first place by censuring their patterns of social interaction to create politically homogeneous networks of political communication. In summary, we are confronted by a contradiction. Political homogeneity within closely held networks of political communication would appear to be the natural consequence of politically interdependent citizens, but political disagreement is demonstrated to survive on a systematic basis (Huckfeldt, Johnson, and Sprague 2000). Disagreement would appear to suggest that citizens arrive at political judgments independently, but analyses repeatedly demonstrate interdependence among citizens (Pattie and Johnston 1999; Zuckerman, Valentino, and Zuckerman 1994). Our goal is to construct an alternative explanation for the individual and aggregate consequences of interdependent citizens in democratic politics. According to this explanation, citizens are neither the powerless dupes of an irresistible social influence process, nor are they individually autonomous actors. Rather, they are the interdependent participants in a process that is imbedded within horizontal networks of political 2

communication and influence (Granovetter 1985; Putnam 1993). And this process systematically gives rise to complex patterns of agreement and disagreement among the citizens who populate these networks. Our research strategy is to pursue an analysis of political communication and influence that combines data collection and analysis with techniques of agent-based simulation. The data collection is essentially complete, based on a study of social communication in the 1996 presidential election campaign as it occurred in the Indianapolis and St. Louis metropolitan areas. In the remainder of this proposal, we construct a plan of analysis. We evaluate the evidence regarding the survival of political disagreement among citizens within their naturally occurring patterns of social interaction, before turning to a description of the Indianapolis-St. Louis study. We then offer an empirical strategy for specifying the structure of agent-based simulation models based on the Swarm toolkit. We consider citizen interdependence based on complex networks of communication the manner in which citizens rely on one another for political information, expertise, and guidance. And we evaluate a dynamic, agent-based model of political persuasion to assess the mechanisms that might sustain disagreement among citizens. The primary questions become, what are the circumstances that give rise to political influence within networks of communication? What are the conditions under which disagreement is likely to be sustained? And what are the consequences, both for individual citizens and for the collective experience of democratic politics? POLITICS, INTERDEPENDENCE, AND THE SURVIVAL OF DISAGREEMENT The classic statement of the socially and politically conservative consequences that arise due to social communication in politics is contained in the work of Lazarsfeld and his Columbia University colleagues, based on their field work in Elmira and Erie County during the 1940s (Lazarsfeld et al. 1944; Berelson et al. 1954). According to their argument, political preferences become individually idiosyncratic as political communication among citizens becomes less frequent during the period of time between election campaigns. In response to the stimulus of the election, the frequency of political communication increases, idiosyncratic preferences become socially visible, and hence these individuals are brought into conformity with micro-environmental surroundings. In this way, social communication creates political stability as it provides a buffer against the political volatility of the external political environment (Huckfeldt and Sprague 1995). The argument presented by the Columbia sociologists is quite persuasive, but carried to its extreme, the logic of group conformity suggests that political disagreement should disappear within networks of social relations. Pressures toward conformity might drive out disagreement in several ways (Festinger 1957; Huckfeldt and Sprague 1995). First, the discomfort of disagreement might encourage people to modify their patterns of social relations so as to exclude people with whom they disagree. Second, people might avoid political discussion with those associates who hold politically divergent preferences. Third, and partially as a consequence of discussion avoidance, people might incorrectly perceive agreement among those with whom they actually disagree. Finally, and perhaps most importantly, individuals might bring their own preferences into correspondence with the preferences that they encounter within their networks of social relations. As compelling as the theory of group conformity argument may be, it suffers from at least one major empirical weakness: campaigns do not extinguish disagreement within networks of social relations. At the end of the 1984 presidential election campaign, Huckfeldt and Sprague (1995) interviewed discussion partners who had been identified by a sample of respondents from South Bend, Indiana. And at the end of the 1992 election campaign, Huckfeldt et al. (1995) interviewed discussion partners who had been identified by a nationally drawn sample of respondents. In both instances, no more than two-thirds of the discussion partners held a presidential candidate preference that coincided with the preference of the main respondent who named them. These measures understate the overall levels of disagreement that exist in the networks in which citizens are situated. Recall that these statistics are based on dyads rather than networks. If the probability of dyadic disagreement within a network is.7, and if the likelihood of disagreement is 3

independent across the dyads within a network, then the probability of agreement across all the relationships in a network with three other discussants drops to.7 3 or.34. In other words, disagreement and heterogeneous preferences are the rule rather than the exception within the micro-environments surrounding many citizens. The pervasiveness of disagreement within networks of political communication leads to a reconsideration of the mechanisms of political interdependence among citizens, as well as a reconsideration of the aggregate implications of political interdependence among citizens. Indeed, the theory of the consequences of social communication for the dynamics of an election campaign might be transformed fundamentally. Rather than serving as a source of insulation from the external political environment, social communication might even serve to magnify the consequences of the external environment by exposing individuals to non-redundant, politically disparate information. THE INDIANAPOLIS-ST. LOUIS STUDY We address these issues on the basis of a unique election study, conducted by the Center for Survey Research at Indiana University during the 1996 presidential election campaign, that was expressly designed to examine the dynamic consequences of the campaign. The primary focus of the study is on political communication and preference formation over the course of the campaign, and thus campaign interviews began early in March of 1996 and ended in early January of 1997. The campaign study includes two samples: a sample of main respondents (N=2,174) drawn from the lists of registered voters, combined with a one-stage snowball sample of these main respondents discussants (N=1,475). Main respondent samples are drawn from the voter registration lists of two study sites: (1) the Indianapolis metropolitan area defined as Marion County, Indiana; and (2) the St. Louis metropolitan area defined as the independent city of St. Louis combined with the surrounding (and mostly suburban) St. Louis County, Missouri. The pre-election main respondent sampling plan was to complete interviews with approximately 40 main respondents each week before the election, equally divided between the two study sites. After the election, an additional 830 respondents were interviewed, once again divided between the St. Louis and Indianapolis metropolitan areas. Discussant interviews were completed at a rate of approximately 30 interviews each week during the pre-election period, with an additional 639 interviews conducted after the election. In the pre-election period, the discussant interviews for a particular main respondent were completed within two subsequent interview weeks of the main respondent interview. In the modern era of presidential election campaigns, it is very difficult to get in front of the campaign, or even to say when the campaign begins, with candidates lining up support and planning strategy long before the preceding midterm congressional elections. Hence, in order to establish a post hoc baseline for campaign effects on the political activation of citizens, we returned to the field in October and November of 1997 using the same questionnaire and sampling design. During this period we completed 438 interviews with main respondents and 265 interviews with their discussants. In order to collect social network information, every respondent to the survey was asked to provide the first names of not more than 5 discussion partners. A random half of the sample was asked to name people with whom they discuss important matters ; the other half was asked to name people with whom they discuss government, elections, and politics (Burt 1986; Huckfeldt, Sprague, and Levine 2000). The experimental condition imbedded within the design of this name generator allows us to examine the extent to which political information networks are separate from social communication networks more broadly considered (Huckfeldt et al. 1998b). After compiling a list of first names for not more than five discussants, the interviewers asked a battery of questions about each discussant. At the end of the interview, we asked the main respondents for identifying information to use in contacting and interviewing their discussants. Based on their responses 1475 discussant interviews were completed for the campaign sample and another 265 were 4

completed as part of the post hoc baseline, employing a survey instrument that was very similar to the instrument used in the main respondent interview. CAMPAIGN EFFECTS ON ACTIVATION AND HOMOGENEITY An important element of the Columbia model, and indeed a crucial ingredient to any model of social communication effects in politics, is the activation of communication as a consequence of political stimuli. We have examined activation effects on the volume of communication from two different vantage points, by considering both the frequency of political discussion within communication networks as well as the size of the networks (Huckfeldt, Johnson, and Sprague 2000). The only campaign effect on communication volume comes in terms of the post hoc baseline. Both the frequency of political discussion and the size of discussion networks are reduced in the fall of 1997 as compared to the period during the 1996 campaign. And we see no variation in these levels during the course of the campaign. Hence, the campaign does appear to increase the volume of communication among citizens, but the activation effect evidently occurs early and persists throughout the campaign. In contrast, other analyses of the Indianapolis-St. Louis data (Huckfeldt, Sprague, and Levine 2000) demonstrate important activation effects on the effectiveness of communication that occur during the course of the fall election campaign. Indeed, the campaign generates substantial enhancements in the effectiveness of communication, where effectiveness is defined in terms of the clarity and accuracy with which political messages are conveyed among citizens. During the course of the campaign, citizens become increasingly more likely to perceive their associates preferences accurately. They are more confident in their assessments of associates preferences. And their judgments regarding the preferences of others become more accessible they come to mind more readily (Fazio 1995). In short, the stimulus of the campaign produces an activation effect on both the volume and effectiveness of political communication among citizens. Does the campaign also reduce political disagreement among individuals who are connected within these networks of social communication? If the campaign has the effect of eliminating disagreement, then the level of correspondence between individual preferences and surrounding preferences should be enhanced across the campaign. While strong relationships appear between the respondent's perception regarding the partisan composition of the network and the respondent s self-reported partisan preferences, we find no consistent evidence to suggest that these relationships change over the course of the campaign. Moreover, in comparing the self-reported preferences of main respondents and discussants, we find that 63 percent of the dyads agreed during the primary season, 52 percent agreed during the general election campaign, and 60 percent agreed immediately after the election. In summary, we see no compelling evidence to suggest that political disagreement is consistently or dramatically reduced as a consequence of the campaign. The simplest, most direct tests for the homogenizing impact of political interaction do not bear fruit. A more subtle, contingent model of communication and persuasion must be pursued. CONTINGENCIES OPERATING ON POLITICAL INFLUENCE The influence of political communication depends in very important ways on the effectiveness of communication within dyads. Hence, we are deliberately separating the effectiveness of communication from the influence of communication. Unless people communicate clearly and unambiguously, they cannot hope to exercise direct influence within their networks of communication (Latane 1981; Huckfeldt and Sprague 1995). Election campaigns activate networks of political communication, thereby enhancing the effectiveness of communication among citizens. The potential for influential communication is thus increased, but the realization of this potential is problematic due to contingencies operating on political influence within communication networks. Perhaps most important, the impact of dyadic communication is contingent on the configuration of the larger networks within which the communication occurs (Huckfeldt et al. 1998a; Huckfeldt, Sprague, and Levine 2000). For example, respondents are much more likely to recognize the preference of a particular discussion partner accurately when they perceive the preference to be more widespread 5

among other discussion partners. In this way the majority within a particular network realizes an advantage in communicating its preference, and the obvious question becomes whether the majority also realizes an advantage in terms of influence. Are discussion partners more influential if their political preferences enjoy higher levels of support within networks of communication? Analyses of the Indianapolis-St. Louis study show that the political influence of a particular discussant is dramatically enhanced both by the effectiveness of communication and by the distribution of preferences in the remainder of the main respondent's network (Huckfeldt, Johnson, and Sprague 2000). That is, discussion partners are more likely to be influential if they hold preferences that are perceived to be dominant within the network and if it is easier for the respondents to render judgments regarding their political preferences, and the cumulative enhancement effect is particularly substantial. These results are compelling, from both substantive and theoretical standpoints. They suggest that the influence of any particular discussant's viewpoint, as well as the effectiveness with which it is communicated, depends on the views of the remaining discussants in the network. Hence, these processes can be understood as being spatially autoregressive, where space is defined in terms of network structure (Anselin 1988; Marsden and Friedkin 1993; Wasserman and Faust 1994). Fully implementing such a model presents a missing data problem we only interviewed 1475 out of the more than 5000 discussants identified by respondents. And thus one part of our research plan is to employ computationally intensive Bayesian methods to address these issues (Jackman 2000a, 2000b; Gelman et al. 1995). HOW DO MINORITY PREFERENCES SURVIVE? Our analyses add to a significant body of evidence suggesting that political minorities operate under pronounced disadvantages in democratic politics (Miller 1956; Huckfeldt and Sprague 1995; Huckfeldt et al. 1998a; Huckfeldt, Sprague, and Levine 2000). Indeed, the influence of the discussion partner's preference is weighted by majority-minority standing within networks of social communication. Regardless of this cumulatively bleak picture for the communication and influence of minority preferences, there is no evidence in the Indianapolis-St. Louis to suggest that minorities tend to be eliminated within closely held networks of communication (see Moscovici et al. 1994). This raises an important question how is the minority opinion able to survive? Like the gene pool, the opinion pool is subject to a number of different forces which act to both eradicate and sustain diversity. At an aggregate level, perhaps the minority is sustained by something as simple as the Markov principle. A small defection rate operating on a large (majority) population will at some point be at equilibrium with a large defection rate operating on a small (minority) population. We suspect other processes are at work, however, which serve both to maintain individual-level disagreement and to sustain minority opinion in the aggregate. Several simple network structures, as well as their implications for the survival of disagreement, are considered in Figure 1. Individuals are represented as ovals, discussant relationships as connecting lines, and the presence of a particular political preference as the presence or absence of shading in the oval. In Part A of the figure, each individual is connected to each of three other individuals in a self-contained network of relations. In such a situation, the logic of our analysis suggests that disagreement is quite likely to disappear, and only the heroic individual is likely to sustain an unpopular belief. In contrast, Part B of Figure 1 shows two subnetworks of four individuals each, where every individual is connected to every other individual within the sub-network. In addition, one individual within each sub-network is connected to one individual in the other sub-network, thereby providing a bridge that spans a structural hole between sub-networks (Burt 1992). Our analysis suggests that, while agreement will be socially sustained within each of the subnetworks, disagreement will be socially sustained between the individuals who bridge this particular type of structural hole. How important are such networks to the survival of disagreement? One way to address this question is by comparing (1) the main respondent's perception regarding the political composition of the main respondent's network with (2) the discussion partner's perception regarding the political composition of the discussion partner's network. Guided by Part B of Figure 1, we are particularly 6

Figure 1. Network structure implications for the survival of disagreement. A. Conformity and the socially heroic holdout. B. Socially sustained disagreement. interested in the composition of the residual networks the networks that remain when the two members of the dyad are removed. Analyses of the Indianapolis-St. Louis data show that, in general, the partisan composition of the main respondent's residual network tends to resemble the partisan composition of the discussion partner's residual network. This similarity is enhanced substantially among main respondents and discussants who report the same presidential candidate preferences. In contrast, for main respondents and discussants who disagree, the partisan composition of their residual networks tends to diverge. In short, agreement within dyads is typically sustained by larger networks of communication that simultaneously support the preferences of both individuals within the dyad. Disagreement also tends to be socially sustained, but by politically divergent networks that serve to pull the two members of the dyad in politically opposite directions. To the extent that networks of communication and influence constitute closed social cells, characterized by high rates of interaction within the network but very little interaction beyond the network, we would expect to see an absence of disagreement among and between associates. Hence, the survival of disagreement depends on the permeability of networks created by "weak" social ties (Granovetter 1973) and the bridging of structural holes (Burt 1992). At the same time that these ties lead to the dissemination of new information (Huckfeldt et al. 1995), they also bring together individuals who hold politically divergent preferences, sustaining highly complex patterns of interaction that produce disagreement, as they expose individuals to alternative political viewpoints. AGENT-BASED MODELS OF POLITICAL INFLUENCE Analyses based on the Indianapolis-St. Louis study provide us with insight regarding the mechanisms that lead to activation, communication, and influence among citizens. At the same time, we cannot hope to untangle the web of endogenous relationships that underlie the interdependence of citizens in the collective deliberations of democratic politics (Achen and Shively 1995). Hence, we turn to agentbased modeling and the Swarm toolkit to consider the aggregate and dynamic consequences of citizen interdependence. We pursue a modeling strategy inspired by Axelrod in his analysis of cultural dissemination (1997a, 1997b, 1997c), where he constructs an agent-based model to explore emergent properties of small scale social interaction. (Also see Kollman, Miller, and Page 1997.) The model conceptualizes interactions among agents in the following way. A square grid of agents, described as 7

villages, is created. Each village has a culture, represented by an array of randomly assigned traits, and each trait varies across a set of alternative "features," representing cultural issue dimensions or topics. The simulation proceeds along the following lines. Each agent (village) is conceived as a unitary actor. An agent is randomly selected and given the opportunity to interact with a randomly chosen neighbor. The set of neighbors is a truncated von Neumann neighborhood. Except for agents on the edge of the grid, the neighbors are found on the on the east, west, north, and south borders. Cells that lie on the outside boundaries are only allowed to look into the grid for neighbors Axelrod does not employ a model in which the space wraps around to form a torus (in contrast, see Epstein and Axtell, 1996). After a random neighbor is selected, an interaction occurs with probability equal to the similarity of the traits of the two agents. If the interaction occurs, then an issue on which the two disagree is selected at random and the agent's opinion on the issue is changed to match the interaction partner. Hence, influence automatically follows whenever interaction occurs. Axelrod made a number of observations on the basis of his model, the most striking being that, over the long run, cultural diversity tends to disappear. While the tendency toward homogeneity is greater for some parameter settings than others, it is powerful in all cases. When diversity survives in the Axelrod model, it is a diversity of the most extreme sort. Different cultural clumps are completely homogeneous and totally isolated from one another. If a village interacts, it interacts with villages that are identical to it. As Axelrod shows, separate groups do not form in some conditions, but they are more likely to form if the number of traits per feature is high. Under those conditions, two agents are less like to have anything in common and so they never interact. He shows that the number of clusters decreases as the number of features increases, and the number of clusters increases as the number of traits increases. Axelrod's conclusion poses a challenge, both to the current modeling exercise and to the larger research project. If we are to formulate a useful model of political communication within small networks of citizens, we must understand the circumstances that give rise to political heterogeneity. One solution is to create agents who are individually resistant to environmental influences, but that is a less than satisfying response to the problem, and it is not the route explored here. Rather our emphasis is on developing a more intricate understanding of the formation of networks and the formulation of public opinion. Using an Axelrod-style model as a baseline, our own analysis turns elsewhere to consider the consequences of several other, newly introduced, model features. We have constructed a working prototype model that is implemented in Objective-C using the Swarm Simulation Toolkit (Minor, et al. 1996; we used version 2.1.4.2000-07-26). Swarm is currently being supported by the Swarm Development Group, a nonprofit membership organization (http://www.swarm.org). The modeling project we describe introduces a large number of variables that can be inspected, including the size of the grid, the number of features and traits, the scheduling of agent actions, and so forth. The substantively important additions concern the processes through which others are sought out for discussion and opinions are adjusted. In addition to introducing a number of system and individual level parameters, we also have introduced summary measures for the diversity of opinion (entropy) as well as measures for the individual experience of diversity. These are discussed below (see also, Johnson 1999). We have pursued this agent-based approach as an alternative to cellular automata. Projects by Latane, Nowak, and Liu (1994) and Nowak and Lewenstein (1996) have used a cellular model in which cells are subjected to influence of varying degree from neighbors to demonstrate some interesting emergent phenomena. The agent-based model can incorporate the strengths of that approach, but it can add a variety of new features, perhaps most importantly the movement of agents within and across the grid and the development of individually distinct logics that govern network development. 8

THE BASELINE MODEL AND BEYOND Our simulation approach is designed to isolate the effect of changes in the model design, allowing investigation of various theoretical conjectures. Each agent in the model is conceived as a separate citizen object, with the ability to move about, initiate interactions, develop memories and expectations, and adjust opinions. Axelrod's culture model is embedded within this more general framework. After verifying that the performance of our simulation is statistically indistinguishable from Axelrod's approach, we then proceed to explore changes in the setting. We have already established the basic structure for the computer model, but considerable development work is needed to explore variations on it. The general framework incorporates the following features. 1. The grid allows multiple occupancy more than one agent at a given location on the grid. 2. Several grids can be created, and agents can be located in more than one grid, each of which might represent a workplace, a home environment, or other place of interaction. 3. Agents can move about in a grid or across different grids during the passage of time. 4. Agents can seek discussants either within their current cell or in other cells. 5. Agents can keep records on their experiences and use them to accept or reject either potential discussants or the information that they convey. 6. Agents can employ a number of different procedures to decide whether or not they will change their personal opinion in response to information communicated by a discussant. These changes are accomplished by incorporating a number of software classes that we have designed in combination with the Swarm toolkit's features for "dynamic scheduling" (see Johnson and Lancaster, 2000: Chapter 9.6). Dynamic scheduling is a model design strategy where events inside the simulation determine what actions are scheduled at future time steps. We have structured our model so that each agent plans its activities over the course of a day, a predetermined number of time steps. At the beginning of each day, the agent may be scheduled to go to work, for example, or stay home. And, during a particular time step, the agent may initiate an interaction with another agent in a neighboring cell. Dynamic scheduling raises a number of technical issues, but on a substantive level, the key notion is that the simulation is "event driven" rather than driven by an external clock. As the simulation proceeds, the agents keep records about the others they have encountered. In order to measure the impact of model variations, we collect that information from the agents. Agents keep track of the proportion of others they have met with whom they have a randomly chosen feature in common (we call them "acquaintances"). Among the people selected for interaction, the agent makes note of the proportion of features on which it agrees with the discussant (the degree of "harmony"), and it also notes if the discussant's features are identical to its own. One of the software classes we have created maintains records in the form of averages and moving averages that are used to aggregate individual experience by calculating the various summary statistics. In order to illustrate the fact that our design does reproduce Axelrod's result, we offer a baseline model that is extremely close in spirit to the original Axelrod model. There is one agent per cell, and the agents are fixed in position. The agent looks for a discussion candidate in a bordering cell. Given the randomly chosen neighbor, interaction occurs with probability equal to the similarity of the two agents. When an agent finds a discussant, then the agent will copy one feature on which the two differ from the discussant. The baseline model settings produce a pattern of dynamic adjustment that is consistent with the original Axelrod results. We have introduced measures of individual experience, as described above, and the graph of a representative run of the model is presented in Part A of Figure 2. The most obvious feature of this figure is that all three measures converge to unity. First, the "acquainted" line indicates that, with ever higher likelihood, a randomly drawn discussant will agree with the agent regarding one or more issues. Second, the "harmonious" line indicates the level of agreement between agents who interact, and it shows that the chances of disagreeing about any particular issue are diminished over time. Finally, the "identical" line indicates an average for the extent to which interacting agents are identical with each other across all five issues, and here again we see convergence to political homogeneity. This particular 9

simulation is not significantly different from the other 100 runs we have conducted with five issues( features) and three positions (traits) for each issue. Figure 2. Agent-based models of communication and influence. A. Baseline model. B. Network mediated influence. average proportion 0.0 0.2 0.4 0.6 0.8 1.0 identical harmonious acquainted average proportion 0.0 0.2 0.4 0.6 0.8 1.0 identical harmonious acquainted 0 1000 2000 3000 4000 0 200 400 600 800 period period Note: The simulation terminates if no opinion change is observed for 10 consecutive days or 100 time steps. Thus, our baseline model produces the same homogeneous outcome as the Axelrod model, and we proceed to introduce variations. The motivation for this effort is that the conformity predicted by the model does not match the empirical evidence regarding communication within networks. Political homogeneity and the absence of disagreement are not typical features of the political landscape. Hence, we explore alternative assumptions about the way that individuals initiate conversations, the manner in which they respond to others who initiate discussions, and the circumstances under which they adjust their opinions. After exhaustive analyses informed by the Indianapolis-St. Louis study, we hope to arrive at a better understanding of the conditions that are necessary and sufficient for the maintenance of heterogeneity in the presence of ongoing interaction. These research activities are described below. THE IMPACT OF SELECTIVE INTERACTION The first variation that we propose to investigate concerns the creation of political networks and discussion patterns. To what extent does the elimination of diversity depend on the ability of individual agents to select discussion partners? In Axelrod's simulation, it appears that diversity within localities is extinguished because agents never interact with others with whom they differ in all features. We have investigated that observation by relaxing the Axelrod's assumption, pursuing a variant assumption in the spirit of Coleman's early work (1964: chap. 16). In his effort to relax assumptions of random mixing in patterns of interaction within populations, Coleman introduced a parameter that allowed individuals to reject interaction, with some probability, across group boundaries (Huckfeldt 1983). We have adapted Coleman's logic to the current context. As in the Axelrod model, the agent chooses a candidate at random from the neighborhood. The candidate will be accepted with probability equal to the similarity of the agents. However, if that discussant is rejected, then with a given probability (which we call the Coleman parameter), an interaction occurs. Hence, as the Coleman parameter grows larger, the diversity of interaction increases. If the interaction does not take place, the individual repeats the search process until an interaction partner is located or ten efforts have been made. The intuition 10

guiding this change was that the homogenizing consequences of interacting with others might be ameliorated by raising the level of interaction with people who are quite different. In a manner totally counter to our original intuition, reducing the individual's ability to ignore others who are different produces a more rapid convergence to political homogeneity. The attenuation of self-selection does not change the fact that, over the long haul, disagreement disappears. We observe that agents who avoid interaction with politically disagreeable encounters are acting to sustain their own cluster of beliefs, and that tends to preserve small clusters and delay the process of political homogenization. Weakening selectivity only serves to accelerate convergence toward a politically homogeneous outcome (Bartholomew 1982). We spend precious space describing the results of this effort to reiterate that all is not always as it seems with complex models of social interaction. By subjecting expectations to evaluation via an agentbased model, we are able to validate or affirm conclusions that are derived from our empirical analysis. As a part of this research agenda, we propose to explore further variations on the selective interaction hypothesis. We intend to explore models that work in the opposite direction of the Coleman model. Instead of exposing agents to wildly differing points of view, we will make them more selective. We wonder what happens if agents can develop relationships and seek out interaction on the basis of their past experience. If the agents can keep records on their previous interactions with specific individuals, they can decide whether to interact with a stranger or a known person (according to various criteria). According to our experience with adjustments in the baseline model, we expect this is likely preserve diversity, but at the price of eliminating interaction among people who disagree. The challenge, then, is to make other changes in the model that re-introduce interaction among people who are not identical. This can be achieved by scheduling agent movement among various environments. SEPARATING PERSUASION FROM INTERACTION The baseline model assumes persuasion will automatically result from interaction. For many purposes, this is a wholly adequate model. If you need information regarding web sites for vacation alternatives, you might indeed seek out information from neighbors and take whatever information they provide. In contrast, the value of political information taken through social interaction is problematic. Even if you acquire information from a generally trustworthy individual who suggests that George W. Bush would make a terrible president, you might want to evaluate the worth of that information. The important point is that communicated information does not necessarily translate into influence, and in this sense the influence of even effectively communicated information is quite problematic. How do people evaluate the worth and credibility of political information? What makes for political information on the part of a communicated opinion or preference? A range of factors could be considered: the clarity with which individuals communicate, the imputed expertise of political discussants, and more. In this instance, we build on our earlier discussion (see pp. 4, ff.) to focus on the incidence of opinions within communication networks (Huckfeldt, Johnson, and Sprague, 2000). If you think that George W. Bush is high quality presidential material, how might you respond to someone who presents you with information to the contrary? Suppose that people have a "mental checklist" which summarizes the opinions they have encountered. What if a person will adopt a new opinion only if their mental checklist indicates that a credibly high level of support exists for that point of view (McPhee 1963)? The political influence of any single interaction ceases to be determinate, and the agent becomes an evaluator of information received through a serial process of social interaction. Any single information source is seen within the context of all the information that is available. In our network mediated model of dyadic influence, we evaluate the consequences of this new assumption. Discussants are selected in the same manner as the baseline model, but agents keep records on the contacts they have experienced and use those records when formulating their responses to new points of view. When an interaction occurs, the agent polls the others that it agrees with on more than one-half of the issues, and if more than one-half of them agree with the new point of view, it is adopted. 11

Thus, new ideas or novel preferences should take longer to catch on, and individual agents should be less susceptible to persuasion. What are the results? As Figure 2(B) shows, when the influence is moderated in this way, diversity is maintained both within the larger population and within networks of political communication. First, the level of acquaintanceship is lower than in the previous models, reflecting the fact that the opinions of the agents are more diverse because people are regularly put in contact with others with whom they disagree. Second, only a relatively small proportion of networks are composed of dyads with identical preferences. Finally, the average proportional agreement with any discussion partner (harmony) is only slightly above.6. (We hasten to add that this figure is highly representative of the 100 runs we performed with these settings.) What do these results suggest? Much more work clearly remains to be done, and we have only begun to address the complex political processes that yield sustained disagreement and diverse preferences in democratic politics. But these first results point to the importance of separating the communication of information from the persuasiveness of information. Even effectively communicated messages may lack influence, and this analysis points to the importance of interdependent citizens as discriminating consumers of political information. Just as important, this example serves to illustrate the interdependent components of a research strategy based on both empirical data analysis and the specification and evaluation of simulation models. We propose to employ this strategy in analyzing a series of explanations regarding the survival and consequence of political disagreement. Several more of these are outlined briefly below. THE IMPACT OF GEOGRAPHIC DISPERSAL Our earlier analyses suggest that the survival of disagreement and the communication of divergent political viewpoints depend on low density networks networks in which everyone does not know everyone else. Several factors might enhance the likelihood of such networks. In particular, to the extent that individuals have larger, more spatially dispersed networks, we would expect lower density networks, with more weak ties (Granovetter 1973), and more individuals serving as bridges across the structural holes that separate otherwise disconnected networks (Burt 1992). The single-grid models we have considered thus far do not encourage that variety of experience. None of the agents have the ability to form associations with individuals who are located beyond the four cells that are contiguous to their own. This is, of course, an imperfect and perhaps misleading abstraction. Indeed, a popular argument among many social scientists is that the organization of modern life has created a world in which space, distance, and location are increasingly irrelevant to patterns of social interaction. Urban residents are just as likely to associate with someone who lives across the city as they are to associate with someone who lives across the street (Eulau and Rothenberg 1986). Moreover, many people are located and involved in a variety of organizations and institutions that produce multiple layers of spatial complexity, and these organizational forms create networks of interaction and communication that are spatially dispersed (see Fuchs 1955). What are the factors that give rise to spatially dispersed networks of political communication? What are the consequences for political communication and influence, as well as for the survival of disagreement? We are able to address these issues in the context of the Indianapolis-St. Louis study by geocoding the residential locations of main respondents and their discussion partners. Thus, we are able to locate individuals within relevant census geography and to calculate distance measures which provide a measure of spatial dispersion within the main respondent networks (Baybeck and Huckfeldt 2000). In the context of Swarm and agent-based simulations, we accommodate geographic dispersal by incorporating the possibility of movement between multiple grids. Each grid can be designed with unique features to accommodate differences in the structure of opportunities for discussion. For ease of discussion we have considered agent movement between a "home grid" (the original Axelrod environment) and a number of "work grids", but they might be church grids, or softball grids, or even bowling grids. Different agents might have variable numbers of grids within which they travel. 12

Interaction in these work grids is completely independent of geographic location in the home grid. The allocation of time during the day across the grids makes use of the scheduling scheme described above. And we are also able to control the initial distributions of opinions across the grids (Huckfeldt, Johnson, Sprague and Craw 2000). THE ROLE OF POLITICAL EXPERTISE There may be situations in which people choose to interact with others on the basis of their perceived expertise, rather than their similarity with one's own opinions. In such situations, citizens are more likely to come into contact with politically divergent preferences. While some analyses employ a cognitive dissonance interpretation (Festinger 1957) to emphasize the importance of agreement and shared political outlooks for political communication, others point to the presence or absence of political expertise among potential discussion partners. In particular, Downs (1957) argued that political discussion is an efficient way to minimize the information costs of political engagement. He argues (p.229) that sensible people search out well informed associates who possess compatible political orientations, with the consequence that citizens become efficiently informed both individually and collectively. In contrast, Calvert (1985) also focuses on the political utility of socially communicated information, but he argues that information can be more useful if it is acquired from someone with whom the recipient disagrees. These issues become more complex because, even if citizens do communicate with others based on perceived levels of political expertise, they may respond to the discomfort of disagreement by over-estimating the political expertise of those with whom they agree and underestimating it among those with whom they disagree (Lord, Ross, and Lepper 1979; Lodge, Taber, and Galonsky 1999). Analyses based on the 1996 Indianapolis-St. Louis study suggest that the distribution of expertise within the electorate plays an important role in affecting patterns of political communication within networks of social relations (Huckfeldt 2001), and three results are particularly important here. First, citizen judgments regarding the political expertise of others are based in reality, driven primarily by actual levels of expertise, with only minor effects due to actual and perceived agreement. Second, citizens communicate more frequently with those whom they judge to be politically expert, quite independently of agreement or disagreement. Third, this asymmetrical quality of communication, in which people rely more heavily on locally defined experts, increases the effectiveness of communication on the part of politically expert citizens. What are the implications of the analysis for democratic politics? Perhaps most important, the capacities of individuals to render meaningful judgments regarding the expertise of alternative information sources is quite striking. People are not lost in a cloud of misperception when they engage in social communication about politics, and neither is the information they obtain simply a mirror of their own preferences. Hence, one of the reasons that "democracy works" might be that citizens rely on "horizontal networks of relations" for meaningful political engagement (Putnam 1993; Mondak and Gearing 1998). This means, in turn, that that the civic capacity of democratic electorates is a complex product of communication among and between populations with different levels of political expertise, and we will pursue both the aggregate and individual level consequences in the context of our project. DOES DISAGREEMENT PRODUCE POLITICALLY DISABLING CONSEQUENCES? The individuals who occupy pivotal roles in the transmission of information across diverse networks may suffer from role-related stresses. In their prescient analysis of social communication in electoral politics, Berelson et al. (1954) draw attention to the two sides of such "cross pressures". At the same time that cross pressured citizens help to provide the political system with dynamic flexibility, they may also be more likely to suffer from indecisiveness, ambivalence, and political withdrawal as a consequence of the conflict that arises due to independent and conflicting streams of incoming information regarding politics. We will pursue these implications, both on the basis of the Indianapolis- St. Louis study, and on the basis of Swarm modeling. 13