前言
研究政治談論網絡（communication network of political discussion）的學者指出，每個人 所處的政治談論網絡的異質性是影響選民政治知識、政治容忍、投票穩定性、乃至政治參與 的重要因素之一。換句話說，網絡中成員政治偏好的異質性，或是每個人感受到自己所處的 談論政治網絡中成員偏好多樣的程度，是個間接測量公民民主參與意願高低的重要指標。然 而，目前關於政治談論網絡的研究還未拓展到造成網絡異質性出現變化的因素或條件，因 此，本研究從媒體使用的角度切入，探討媒體使用如何影響政治談論網絡異質性的情形。具 體的假設是：當總體選民暴露在同質性高的媒體頻率愈低，他們感受到的網絡異質性就會愈 低。這個研究獨特之處，在於使用代理人基模擬（agent-based modeling）途徑來研究人際互 動與媒體使用的動態過程。 本計畫承接上年度的國科會計畫研究成果，採行一系列的電腦模擬實驗。模型的設計考 量到政治知識高、低選民之間取得政治資訊的行為的差異，並透過六組參數的變化，呈現六 種不同情境下媒體使用率降低對政治討論社群異質性提高（或降低）的影響。雖然，上年度 的研究計畫成果指出，本研究所使用的模型（S-RAS）仍然必須持續進行外部有效性的測試， 所以模擬結果並無法直接回饋給經驗世界（empirical world），產生指導作用。然而，外部有 效性不足本是所有實驗研究共同的限制，因此它的高低並不影響模擬實驗作為一個研究途徑 的求知本質。本計畫建基於理論與實務所推導出來的問題意識之上，因此，研究發現多少能 對現今的審議式民主理論、參與式民主理論、社群網絡研究等領域產生一些啟發。本計畫成 果（含前言、研究目的、文獻探討、研究方法、結果與討論等）已以英文寫成，附於自評之 後。
計畫成果自評
本研究與原計畫提案大致相符，唯研究發現與審議式民主的討論不太相關。本研究 突破了政治學者除了使用大型面訪案的資料之外便無法研究群眾行為的限制，得以在這 類資料收集不易的情形下進行理論的探討；當初設定的預期目標（探討媒體使用造成社 群異質性產生變化的條件）達成。本研究計畫的英文稿將在潤稿後嘗試投搞 2009 美國政 治學會年會。本研究進行過程中遇到幾個挑戰：一、以研究者本身能力，尚無法以數學 式寫出模型的設計（僅以文字敘述），因此模型的設計將不易與電腦專業的學者對話。 二、選民對大眾媒體的注意力是否持續降低，或只是移轉到其他型態的媒介仍需持續透 過實證的方式觀察。就第一點而言，需要其他學者的協助；就第二點而言，則需要持續 追蹤文獻的發展以及就現有國內實證資料作出跨年的比較。The Decline of the Politically Informed
Electorate and the Survival of Political
Disagreement within Communication
Networks
One important characteristic of a democracy is the prevalence of political discussion (i.e., talks among individuals about politics and public affairs in both formal and informal settings). Given freedom of speech, voters in a democracy are free to access news media and self-selected contacts for political informa-tion. Recently, an empirical puzzle emerging in democracies shows that the public, especially the younger generation, becomes less interested in accessing electoral news. This phenomena lead scholars to wonder if it hurts democracy (e.g., Bennett, 1998; Mindich, 2004; Robinson and Levy, 1996). For exam-ple, Witnessing the decline of interest in political news, some scholars concerns about its influence on political knowledge. Since the newspaper is the major source of the electorate to become politically informed, when newspaper read-ership drops, "the overall public may be gradually become less well-informed" (Robinson and Levy 1996, 135). “We should [worry] if we believe that democ-racy depends upon an informed citizenry and that being knowledgeable about public affairs is a key ingredient of democratic citizenship” (Bennett, 1998, p. 540).
The declining of the politically aware is one of possible consequences of the decline of propensity to access political information. Nevertheless, the former is not yet the best indicator about how healthy a democracy is. One may ar-gue that democracy functions even when voters become less interested politics, when turnout rate remains low, and when the majority of the electoral body con-tinue to be politically unaware. If the decline of propensity to access political information, as scholars have identified, is a problem threatening the health of a democracy, a better indicator will be whether or not it affects the environment that cultivates democratic citizenship.
Scholars studying the environment that cultivates democratic citizenship, such as political participation and political tolerance, have come to focus on het-erogeneous communication networks, in which an individual perceives diverse political preferences and confronts political disagreement (e.g., Huckfeldt et al., 2002; Mutz, 2002; Scheufele et al., 2006). As a growing literature on the role of heterogeneity of communication networks suggest, the level of network het-erogeneity is an effective indicator for the health of democracy. To see whether or not the decline of propensity to access political information hurt micro-level democracy, we need to redirect our focus from the concept of democracy in
general to network heterogeneity.
The present article uses the agent-based modeling approach to investigate this inquiry: When more and more voters become less interested in acquir-ing political information through political discussion and/or watchacquir-ing campaign news, will there be any change in their perception about the heterogeneity in their networks? Specifically, this question can be divided into three “what-if” questions or scenarios: (1) What if voters become less interested in political discussion but still turn to the mass media for political information? (2) what if voters become less interested in watching, reading, or listening to political news but still habitually discuss politics? (3) What if individuals become indifferent to both communication networks and news media and dive into electronic enter-tainment? The findings of the present article will expand the scope of literature on communication networks to the subject about the antecedent conditions of heterogeneous network.
The present research uses the agent-based modeling (ABM) approach to ex-plore the relationship between the decline of interest in political information and the survival of disagreement. This approach allows researchers to go be-yond three constraints of the current method. First, it is difficult to answer these questions with empirical data analysis because the unit of analysis for voter research needs to be consistently individual voter across chosen variables. Sec-ond, it is difficult to use statistical analysis to deal with the effect of dynamic changes at the aggregate level (e.g., all voters become less interested in political news) on individual voters. Third, it is costly, time-consuming, and not realistic to conduct society-wide experiment to examine the effects of certain conditions on voters.
Simulations results generated from a laboratory, hence, are informative and suggestive for future empirical research. As simulations of the present research are not based on empirical datasets but on behavioral rules and conditions sug-gested by empirical research, they are not aimed to be applied directly to the empirical world. Even though, the robust findings derived from 100 runs for each condition will provide theoretical implications: a society-wide decline of interest in accessing political news can harm the survival of political disagree-ment within communication networks.
The Declining Interest in Political Information
A growing number of evidence suggest that American voters are less inter-ested in politics and in obtaining political information. Smith (1989) gives little hope of major changes in the electorate’s qualifications for thoughtful voting. He thinks that the American public has not become more politically sophisti-cated since the 1950s and is not likely to become more sophistisophisti-cated in the fu-ture. Moreover, there is little evidence suggesting that the majority of American voters have become more involved in politics and elections or more capable of developing stable and coherent belief systems (Campbell et al. 1960; Converse 1964; Smith 1989, see also, Nie et al. 1976). Even though the use of the Internet has expanded dramatically, the trend of "bowling alone" continues—the over-all percentages of the electoral body that consume political news and political
discussion are declining (Kaid 2003; Patterson 2002; Putnam 2000).
The advocates of deliberative democracy and participatory democracy as-sume that, given enough channels to participate, individuals will actively get involved in their political contexts. However, a number of empirical studies challenges the assumption about the activeness of voters. The most optimistic one is that individuals who are resided in heterogeneous communication net-works (i.e., netnet-works whose members possessing diverse political preferences) are more likely to get involved in political activities (e.g., Anderson et al. 2005). In fact, American voters become less involved in consuming news and be-come less interested in political discussion. A private survey shows that between Memorial Day and Labor Day in 2005, the six broadcast networks, except ABC and Fox, suffered a 13 percent decline in viewership 1. This decline does not
imply that TV news viewers switch to the online news or newspapers for polit-ical information. According to the National Election Survey data for the 2004 presidential election (NES2004), even during the campaign season, 84.7% of the respondents say they read online news less than three days (0, 1, or 2 days) in the week before the survey. Only 7.2% report that they read online news almost every day (6 or 7 days).
A study based on an analysis of Pew data in April 1996 and February 1997, and the 1996 National Election Study shows that young adult voters under 30, who are better exposed to expanded education than their the earlier generations, are less likely to read newspapers, watch TV newscasts, and use the news media. He worries that this would result in citizens ignorant of contemporary Ameri-can politics and incapable to continue the society. “Although young AmeriAmeri-cans are normally less engaged in politics than their elders, today’s youth are more withdrawn from public affairs than earlier birth cohorts were when they were young” (Bennett 1998, 535).
Confirming this pattern, Mindich (2004) suggests that few people who are under 40 years old perceive a need for paying for good news stories. The rea-sons for this decline of media access include the perception of isolation from the political process, the declining of trust in the media, the decrease of social capital for public dialog, and more excuses to stay at home. Although there is a generation gap regarding political knowledge and media use, a bigger picture is that “while most young Americans are tuned out, older Americans are not exactly tuned in” (12). Consistently, those who did not watch presidential de-bates in 2004 are those who are younger, less educated, having lower incomes, and exhibiting lower levels of partisan attachment than those who watched the debate.
This pattern of the dropping of interest in political news also applies to po-litical discussion. Individuals, especially the younger generations, spend more time on electronic entertainment than getting involved in political discussion. Mindich (2004) notices that the younger generation tend to claim their rights of not to think about and talk about politics. According to NES2004, the fre-quency of discussing politics is positively and mildly correlated to one’s level of political knowledge (r=.21, p<.001), the days of watching TV news (r=.11, p<.001) and days of reading newspapers (r=.10, p<.01). This is consistent with the findings that the frequency of political discussion is related to social context
(e.g., close friends and relatives), media use, and personal traits, such as polit-ical interest, strong party attachment, liberal views, and newspaper readership (Anderson et al. 2005; Kim et al. 1999; Straits 1991). Even though we still lack direct evidence showing that individuals do become less interested in discussing politics, the significantly positive correlations suggest that when individuals are less interested in accessing news media, it is likely that they do not turn to dis-cuss politics.
The Decline of Interest and the Survival of Political
Disagreement
Studies of communication research suggest that individuals who live in het-erogeneous communication networks are more likely to be more politically knowledgable, politically tolerant, holding de-polarized and ambivalent pref-erences toward a candidate, and involved in political participation (Huckfeldt et al., 2002, 2004a; Mutz, 2002; Scheufele et al., 2006).2 Hence, the level of heterogeneity of a communication networks, or the degree of disagreement an individual perceives is an indicator of the demoracry at the micro-level.
One reason that political disagreement can survives in communication works without being eradicated through conformity is that communication net-works “serve as transmitters and intermediaries that connect individuals to the events and circumstances of democratic politics" (Huckfeldt et al., 2002, p. 19). Individuals tolerate and embrace their friends’ opposite preference because they are friends rather than strangers. Disagreement hence is allowed to survive in one’s communication network. Besides affection between the dyads that help survive political disagreement, the decline of partisanship and the way infor-mation obtained through communication networks is distributed also matter. “whether or not the first individual is influenced by the opinion of of a second individual within the network depends on the distribution of opinions across all the other individuals within the network who are also connect to the first indi-vidual," “individuals are less likely to be persuaded by opinions that win only limited support among the participants within their communication networks” (Huckfeldt et al., 2004b, p.20). In other words, when an individual will stand against the opposite preference within their networks as long as they have sup-portive members.
The current list of conditions in which political disagreement survives or the heterogeneous networks increases in number is far from complete. Beyond the above factors at the intra-personal and interperonal levels, there is plenty of room to explore aggregate-level conditions that lead to the survival of dis-agreement within communication networks. The present research one of the conditions is the society-wide decline of interest in political information.
2Note that there is on going debate about whether the heterogenous networks increases
po-litical participation or not. Mutz (2006) suggests the negative relationship while Scheufele et al. (2006) suggests the opposite.
Assumptions and Hypotheses
The analysis of the present reserach is based on two core assumptions about individual voters. First, individuals tend to discuss politics with political experts (even those hoding opposite preferences) (Huckfeldt et al., 2002; Beck, 1991; Mutz and Martin, 2001; ?; Scherer and Cho, 2003; Wyatt et al., 2000) and like-minded others within their self-selected communication networks (Carmines and Huckfeldt, 1996; Huckfeldt, 2001; Miller and Krosnick, 2000). Second, compared to accessing communication networks, individuals are more likely to acquire political information from self-selected news sources more (Mutz and Martin, 2001).
When individuals as a whole become indifferent to news reports during a campaign season but do not change their propensity to discussing politics, infor-mation will be expected to flow come from well-informed network members. If communication become highly homogenized, it is likely that the individual will conform to the majority preference. Hence, the first hypothesis will be, when the
electoral body becomes less interested in accessing news sources during a cam-paign season, the proportion of individuals who perceive network heterogeneity will decrease (H1).
Following the second scenario, if individuals as a whole feel less interested in discussing politics, but keep their propensity to access the news media, the homogenizing effect of social network will decrease due to the decline of in-terpersonal contact. The second hypothesis, hence, will be, when the electoral
body becomes less interested in discussing politics during a campaign season, the proportion of voters who perceive network heterogeneity will increase (H2).
The third condition is the worse scenario. If the second assumption is cor-rect, then the homogenizing effect of communication networks is expected to be smaller. The third hypothesis will be, when the electoral body becomes less
interested in both political discussion and accessing political news during a campaign season, the proportion of voters who perceive network heterogeneity will decrease (H3).
Method
The S-RAS model is a proper agent-based simulation program for the in-quiry of this project. It is constructed with the Swarm toolkit (http://www.swarm.org) that empowers researchers to customize characteristics of agents, particularly agents’ propensities to use the media and to discuss politics. The term agents in the S-RAS model refers specifically to voters. ABM researchers design and validate a model and specify initial settings of the environment. During a simu-lation, researchers do not interfere in the process of the automatic actions among agents; instead, researchers wait and see if there is any meaningful patterns emerging from the simulations.
The design of S-RAS model draws on John Zaller’s (1992) Receive-Accept-Sample (RAS) model and Huckfeldt, Johnson, and Sprague’s (2004a) autore-gressive model. Agents of S-RAS self-select their networks members with higher level of political expertise and/or those who agree with them in voter
preference or party identification. As a result, one should expect to see that networks in S-RAS will homogenize network members’ preferences.
This present research considers individual differences by varying individual agents with respect to the probabilities to access news and discussing politics, and capacity to memorize preferences obtained from past interactions with oth-ers and the media. Ordinary citizens (C1 agents) are in general less politically attentive than political experts (C2 agents). All models randomly generate the same number of C2 agents, which accounts for about 2% of the 16,000 agents. Because of the settings, C2 agents account a very small proportion of the popu-lation, it is expected that the influence of C2 agents on C1 agents to be mild.
Agents of S-RAS access self-selected news media and perform selective per-ception on the information. The two news media "objects" in the S-RAS model refer to two sources that broadcast preferences opposing each other (e.g., favor-ing John F. Kerry vs. favorfavor-ing George W. Bush in the 2004 U.S. presidential election) and can be accessible by any agent at any time. For individual agents, these news objects are better understood as their favorite news programs. One aspect on which agents differ from each other is the probability of performing selective perception when accessing news media. By design, C1 agents are less likely to perform selective perception than C2 agents. This means that C1 agents are more likely to randomly receive preferences whenever they access the news media and that C2 agents are more likely to perceive preferences consistent with their current preferences.
One important initial setting and constraint of S-RAS is that all agents are polarized with regard to their vote preferences. Half of C1 agents hold "YES" and the other half holding "NO," while half of C2 agents holding "YES" and the other half hold "NO". There can be numerous initial settings, such as some voters are independent or undecided, but this setting of "a polarized society" will be the simplest one that characterizes a very competitive presidential cam-paign season. This setting helps simplify the initial environment for cross-model comparison. Future applications of the S-RAS model can relax this polarization setting for other types of societies.
Modeling Strategy
The above points out the most important settings for the simulation; these settings were held constant throughout the simulations. This section specifies how to vary two parameter values, the maximum probability of accessing news media and the maximum probability of political discussion, to explore the ques-tions this article addresses. The variance of values in the two parameters will result in six models. Each of the six models will run for 100 times with different random seeds and yield summary statistics. The first series of models, Models A, B, and C, will together demonstrate the effect of the decline of interest in dis-cussing politics on the proportion of gents perceiving preference heterogeneity in their communication networks. Specifically, the maximum probability of po-litical discussion is held constant across the three models, while the maximum probability of C1 agents to access the news media in Models A, B, and C are set to 0.5, 0.25, to 0.1, respectively. As a result, ordinal citizen agents in Model
A have greater odds to access to their favorite news media than their counter-parts in Models B and C. The second set of models resemble the first set, except the changes in the value of the second parameter, the propensity to access the news media. Models D and E holds constant the probabilities of accessing the news media constant and vary the second one. The maximum probability of C1 agents to discuss politics are 0.25 and 0.1 in Models D and E, respectively. The comparison between Model A, D, and E will show how the decline of in-terpersonal political discussion (from 0.5, 0.25 to 0.1) affects the proportion of agents perceiving political disagreement within their communication networks. Model F is the one where C1 agents are least likely to access the news media and discuss politics. The maximum probabilities of C1 agents to both accessing self-selected news source and discussing politics are set to 0.1. Hence, contrast-ing the results of Model F with those of the first five models will help inspect the effect of the decline of interest in acquiring political information.
Simulation Results
The first half of the article specifies how to use the S-RAS model to answer the research question; this section reports the simulation results in three per-spectives. First, the summary statistics in Table 1 show general differences of the results across the models. Second, the time-series graphs in Figure 3 show the dynamics of preference change at the aggregated level.
Third, the opinion grids in Figure 4 visualize the distribution of opinion clusters and the diversity perceived in agents’ communication networks. The initial condition is represented in the top panel of Figure 4. In this grid, the black cells represent agents who say “NO” and the white cells represent the agents who say “YES.” The opinion distributions for each of the models at the 2001st time step are illustrated in the lower part of Figure 4. The same initial conditions are fed into each model, and while the models run I observed the evolution of individual opinions as well as the time paths of the summary indicators (these are all displayed in Figure 3).
If the deliberative democracy theory is correct, individuals are more likely to be open-minded to evaluate different preferences when they perceive a high level of heterogeneity in voter preference. This is to say, on an opinion grid, an communication network mixed with white and black cells cultivates political deliberation (i.e., exchanging preferences and seeking consensus through deeper conversation). Contrarily, if a communication network becomes homogeneous, which is either totally white (or light grey) or completely black (or dark grey), no preference change is likely to occur within the network.
The curves and opinion grids shown are representative of a broad set of simulations. The curves and the grid pictures are obtained from the models with means of the two variables (Opinion “YES” and Diversity) falling within 2 standard deviations from the mean of the 100 simulation results. Specifically, I pick Run 014 for Model 1, Run 033 for Model 2, Run 045 for Mode E, Run 063 for Model 4, and Run 087 for Model 5.
Comparing the statistics, time-series graphs, and opinion grids of Models A, B, and C outline the consequences of the decline of (C1 agents’) propensity to
access the news media. Comparing these types results of Models A, D, and E sketches out the outcomes when C1 agents become less likely to discuss politics. Because the maximum probabilities of both variables in Mode E are 0.1, I treat it as a special case and compare it against the two sets of models.
Table 1 shows the summary statistics of the 100 runs of each model. The means of the variable Proportion of Agents Holding “YES” is about .50 across the board, which suggests that by the time when the simulation ends (at the 2001st time step) no dominant preference forms in any model. As described on page 6, this consequence results from the setting of polarized agents. The major two clear patterns Table 1 show are that (1) the means of perceived diversity decrease in the sequence of Models A, B, and C, and that (2) the means of perceived diversity increase in the sequence of Models A, D, and E. These two sets of models suggest that the perceived diversity increases when C1 agents become less likely to access their favorite media source, and it decreases when the agents become less likely to discuss politics.
Figures 2 on the following page and 3 on page 11 show time-series graphs of the simulations. The X-axis is the time step, while the Y-axis is the average pro-portion. The curves in Figures 2(a) and 2(b) show the dynamical changes of vote preference of all agents. They suggest the continuation of preference polariza-tion. These two pictures also suggest that when C1 agents become indifferent in accessing their favorite news sources (but remain interested in discussing poli-tics occasionally), change of public preference can become intensive. Compared to all other models, Model C has the greatest variance.3
Figures 3(a) and 3(b) show the dynamics of the perceived preference het-erogeneity . The patterns of decreasing proportion in Figure 3(a) and increasing proportion in 3(b) are consistent with those shown in Table on this page. The curves in these figures help compare Mode E against the two sets of the models. First, compare Model C and Mode E in Figure 3(a). Given that C1 agents are indifferent in accessing their favorite news programs, when they become less likely to discuss politics, the proportion of perceived diversity increases. Sec-ond, compare Model E and Mode E in Figure 3(b). Given that C1 agents are in different in political discussion, when they become indifferent in their favorite
3Figures 2 (a) and (b) do not show the curve of Mode E because, when added, it overlaps
other curves and makes it difficult to identify the four curves.
Table 1: Snapshot at the 2001st Time Step—Average for 100 Simulations
Model A B C D E F Probability Media Access [0, .50] [0, .25] [0, .10] [0, .50] [0, .50] [0, .10] Discussion [0, .50] [0, .50] [0, .50] [0, .25] [0, .10] [0, .10] Outcome Average Proportion “YES” .503 .501 .499 .502 .502 .502 (.025) (.039) (.039) (.021) (.016) (.034) Perceived Diversity .396 .306 .234 .438 .468 .346 (.011) (.013) (.016) (.010) (.008) (.011) Note: Each simulation runs for 2000 time steps.
Figure 2: The Time-Series Graphs of the Proportion of Agents Holding “YES”
(a) Decline in the Probability of Media Access
Figure 3: The Time-Series Graphs of the Perception of Network Heterogeneity
(a) Decline in the Probability of Media Access
news programs, the proportion of perceived diversity drops dramatically. These results together imply that accessing self-selected news sources helps individu-als preserve their political disagreement and increases preference heterogeneity within networks.
Figure 4 on the next page gives the snapshots of the 16,000 agents’ opinions when the simulations stop at the 2001st time step. The picture of the initial set-ting show the random distributions of both C1 and C2 agents. First, compare the pictures of Models A, B, and C. Distinct opinion clusters form and become more homogeneous when agents spend less time accessing their favorite news sources, spend more time spent on interacting with network fellows, or even stop interacting with their political context. This pattern suggests that little deliber-ation occurs in Models B and C. As white clusters becomes whiter and black ones become blacker, deliberation can occur in border networks where agents perceive diverse preferences.
Second, compare the pictures of Models A, D, and E in Figure 4. The line between white and black becomes blurry and opinion clusters diminish. This pattern, correspondent to the increase of perceived diversity shown in Figure 3(b), suggests that, when C1 agents become less involved in political discussion, the homogenizing effect of communication networks decreases. This results in a greater number of agents holding political disagreement and greater chances for political deliberation. Interestingly, this trend does not result in more bright white areas and more deep dark areas on the grids. Instead, grey areas expand. As a grey cell indicates that the agent’s opinion is close to 0.5 (away from 0.0 or 1.0), the emergence of the pattern like the picture of Model E suggests that, when the influence of communication networks declines, an increasing number of agents are likely to become independent.
Last, contrast the picture of Mode E against the above two sets of grids (Models A, B, and C and Models A, D, and E). The comparison between Model C and Mode E suggests that, when C1 agents become less likely to discuss politics, the homogenizing influence of social networks decreases. Addition-ally, the borders between black and white clusters blur and more grey areas emerge. Moreover, even though opinion clusters remain, the strength of opin-ion decreases, a pattern characterized by less bright white and less dark areas. Comparing Model E and Mode E leads to the same conclusion. When C1 agents become less likely to access their favorite news programs, which means that they can choose to interact with their network fellows or simply stop interacting in all political contexts, grey areas diminish and clusters form.
In sum, the patterns emerging from the series of simulation support the first and third hypotheses: When the electoral body becomes less interested in ac-cessing self-selected news sources during a campaign season, the number of voters who perceive preference heterogeneity declines. This implies that ac-cessing self-selected news sources plays a critical role in sustaining political disagreement. The results do not support the second hypothesis: When the electoral body becomes less interested in discussing politics during a campaign season, the number of voters who perceive preference heterogeneity declines. Instead, the number of individuals perceiving heterogeneity in voter preference increases. This implies that political discussion plays a critical role in homog-enizing the society. The findings above further suggest two conditions where
Figure 4: Opinion Grids of Preferences
Initial Setting
Model A Model B Model C
individual perceive less preference heterogeneity: continuous accessing of self-selected news sources, and involvement in political discussion with self-self-selected network members. The next section discusses the implications of the findings for deliberative democracy.
Conclusion and Discussion
This article uses agent-based modeling to simulate a polarized society where individuals become less involved in accessing political news and political dis-cussion. The findings about the consequence of news access correspond to scholars’ concern about diminishing political deliberation: Individuals are more likely to perceive less preference heterogeneity and conform to the majority preference. The findings, however, relieve a bit of the worry about the conse-quence of decreasing involvement in political discussion. As long as individuals access alternative information sources, even though such alternatives are self-selected, the number of individuals perceiving preference heterogeneity does not decrease.
An explanation about how such results emerge requires an inspection of the dynamics of the simulation. First, when individuals become less likely to ac-cess their favorite news sources, large-scale and homogeneous opinion clusters form. Within such clusters, serious reflections on different perspectives become less possible. Moreover, when individuals become indifferent to news media sources but remain interested in discussing politics occasionally, one can ex-pect a dramatic fluctuation in public preference change. If it is true that indi-viduals like to access people by self-selection, then self-selected news sources can be the counter-forces of the homogenizing effect of communication net-works. When individuals’ preferences are consistent with the majority of their network fellows, their accessing of self-selected news programs will strengthen their preferences over time. For individuals who hold minority preference in their networks, however, accessing self-selected news sources helps preserve their own political disagreement and increases others’ awareness of the hetero-geneity within networks.
Second, the increasing number of people participating in political discus-sion does not lead to the perception of greater preference diversity; instead, the average proportion perceiving diversity decreases across the board.
Advocates of deliberative democracy expect that citizens actively participat-ing in political discussion, especially with randomly, rather than subjectively, selected people. Through such active, face-to-face interactions, individuals ide-ally will learn more about public problems, consider fairly when making voting judgements, and feel politically empowered (see, Fishkin 1995; Cohen 1989). Unfortunately, this ideal is not very possible when voters have become less in-formed of campaigns and less active in political participation. The idea of prob-ing the true preference of the public through deliberative poll also encounters the fact that individuals lack strong motivations to attend a town hall meeting and compete with strangers holding opposing preferences.
This article sheds light on this issue by examining the consequences of polit-ical discussion within a more realistic settings—individuals access self-selected
mass media and self-selected discussion partners. If it is true that heteroge-neous political contexts facilitate political deliberation, then the findings con-tradicts what advocates of deliberative democracy expects: Active participation in political discussion may not increase circumstances under which individu-als perceives preference diversity. Decreasing individuindividu-als’ propensity to discuss politics within their communication networks, ironically, is the way to maximize their perception of political disagreement.
This conclusion also implies that current TV news programs that aim to arouse political discussion in the living room, online message boards, and vir-tual forums, do not help individuals to perceive wider political views in daily life. Instead, when they are emotionally motivated, they are more likely to in-teract with their self-selected discussion partners and become similar-minded. The deliberation may then become, in James Madison’s term, “passion.” Per-ceiving a certain degree of heterogeneity in their daily interacting with political context seems to be good enough for the electorate to initiate comparisons be-tween different vote preferences or even to make a fair judgement of candidates. We should worry less about individuals becoming less involved in discussing politics than about consuming political information. Even if the information is biased, this article confirms the important role of multiple news sources in sus-taining political disagreement at the individual level and at the aggregated level. Biased media may not directly result in biased voters as Bennett (1998) argues; this article suggests that voters’ interacting with their homogeneous networks is the major cause of voter bias.
Three issues deserve exploration in future research. First, when do voters that perceive heterogeneity initiate meaningful exchanges of political prefer-ences? Second, what patterns can we find if agents have more than two pref-erences and access more than one favorite news source? The ABM approach in general, or the S-RAS model specifically, can be a new tool for researchers to explore the dynamics of the perception of polarization during a campaign season.
Appendix: The Design of the S-RAS Model
The Features of Agents
Two types of citizen agents compose the S-RAS model—the less aware cit-izen agents, referring to ordinary citcit-izens who pay little attention to politics and elections, and the politically aware citizen agents, referring to the citizens with greater political knowledge. The following two subsections describes the common features shared by the two types of agents and their differences, re-spectively. These descriptions may seem trivial, but they help to discern how the model actually works.
The general features of agents
In the S-RAS model, all citizen agents follow four specified behavioral rules. The first two are about accessing communication networks, the third about me-dia use, and the fourth about opinion change.
The first rule is that agents tend to discuss politics with the politically aware, i.e., agents of greater political expertise (Huckfeldt et al., 2002; Beck, 1991; Mutz and Martin, 2001; ?; Scherer and Cho, 2003; Wyatt et al., 2000). As a result, agents are more likely to do so when they judge the discussants as trustworthy and more politically knowledgeable than themselves (Carmines and Huckfeldt 1996; Huckfeldt 2001; Miller and Krosnick 2000).
Second, every agent subjectively chooses eight contacts as communication network members. Some of the members are more frequently contacted than the others based upon the following rules: (1) An agent makes a contact list of favorable discussants according to the similarity in political predisposition (i.e., party identification) and the level of political awareness (i.e., political ex-pertise); (2) Agents will seek those with the same party identification before turning to friends who have high political expertise but different political party identification; and (3) The least favorable discussants in one’s network are those of different political party identification and those with lowest political exper-tise. Ultimately, when an agent finds an available discussant, both agents will become unavailable to the other agents (Huckfeldt and Sprague 1995).4
The third behavioral rule is that agents will think about accessing the news media (i.e., check random probabilities against their propensities) before dis-cussing politics with their network members. The term “news media” used in
4Granovetter (1973) suggests two types of social networks: strong-tie social networks—
family and close friends—and weak-tie social networks—coworkers or other friends. This categorization is straightforward but limited. In the new millennium, people can form their communication networks beyond the geographic boundary of family and neighborhood. In-fluential political discussants may come from weak-tie networks, such as friends met on the Internet. Although the agents in this model are described as residing in a grid, I do not mean to say that social interaction is strictly based on geographical distance. Rather, these eight possible contacts might be located in a diverse set of geographical positions. Moreover, although it is possible to vary the number of contacts, I do not let this parameter vary. That agents have eight contacts does not mean they have a fixed number of contacts. In stead, according to the design of individualized preferences, the number of an agent’s favorite contacts will be less than eight in the initial, random setting.
this article coincides with what Zaller calls “elite discourse” because most infor-mation about an election that people watch, read, and listen to are candidates’ speeches and debates, politicians’ campaign talks, and invited experts’ analyses, etc, in the news media. This idea comes from the fact that people tend to check the news media before interacting with their communication network members (Mutz and Martin, 2001).5
The fourth rule is about opinion change. Agents’ current vote preference (0 or 1) is obtained from their opinions (varying between 0.0 to 1.0 as simulation runs). As Axiom 4 suggests, an agent’s current opinion =C+DD , where D denotes a dominant message (i.e., message that is more intense during the period of attitude change) and C denotes a countervalent message (i.e., the less intense message). C and D are measured by the numbers of considerations.
If an agent perceives more than six dominant messages and four counterva-lent messages of a dominant message, his or her opinion is 0.6. Because this figure is greater than 0.5, his or her current vote preference is 1. Similarly, if another agent receives three “1”s and seven “0”s in the past ten activities (ei-ther accessing the media, talking to o(ei-ther agents, or doing nothing but resuming current opinion), his or her current opinion is 0.4 and his or her current vote preference is 0.
Note that the four axioms in RAS deal with how opinions originate, but not with how and when individuals change their vote preferences. Although Zaller proposes a probability function of preference change based on Axioms 1 and 2, this formulation “omits any reference to ’considerations,’ which cannot be easily measured in most attitude change situations, and refers instead to the probability of change in a person’s summary attitude report.” As he continues, “it also omits any reference to countervalent messages, even though they will often be present in attitude change situations” (Zaller 1992, 123). Indeed, the probability formulation is good for the fitting empirical data, but it omits the details about how agents proceed perceived information.
The theory of autoregressive influence provides an alternative solution. The term “autoregressive influence” means the influence of perceived external pres-sure, including peer pressure. According to the theory, people tend to change their vote preferences when they perceive that they are the minority in their so-cial context (Huckfeldt et al. 2004a; Neuman et al. 1992). Such soso-cial influence “depends on the distribution of opinion across all other individuals within the network who are also connected to the first individual” (Huckfeldt et al. 2004b, 20); therefore, when individuals perceive that messages from their social con-text turn to oppose their current preferences, they are likely to conform to the majority. Hence, in the S-RAS model agents will change their vote preferences from 1 to 0 (or 0 to 1) when they perceive at least 50 percent of the messages from their context (including friends and the news media) opposing their current vote preferences.
Figure 5 on page 19 is a summary of the above behavioral rules. It serves as the guidance of modeling (or programming). This chart shows how an agent
5Even so, flexibility is allowed. Agents vary in the probabilities of accessing the media and
the probabilities of discussing politics. If they have low probability of accessing the media but high probability of discussing politics (by initial setting), they will follow the probabilities rather than the rule.
accumulates information received from dyadic conversation and the mass me-dia, and forms vote preferences. Note that before the simulation ends, agents will keep choosing between discussing politics, accessing the media, or doing nothing, at every run time (or iteration).
The differences between the politically aware and the less aware
This subsection details individual differences between the politically aware
and the less aware. This categorization is important for two reasons. First, it
reflects two important observation about the electorate. The first observation is that partisan voters, including political elites, have “large storage of political lore” and will be more stable in their vote preferences than ordinary voters that are less politically interested (Converse 1964, 1990). Converse suggests that elites are different from the mass public in two respects. First, because of inter-est in politics, elites process political information more efficiently than general voters do; these pieces of accumulated information in turn mobilizes themselves to obtain more political information, or at least to maintain the level of interest. Second and consequently, he observes that few ordinary voters with lower ed-ucation level participate in political process. The second observation about the electorate is that political elites are more more polarized than ordinal voters (Fiorina et al. 2005). This suggests that the political aware are more likely to be biased, partisan, and selective in consuming news messages (see also, Oliver 2002).
The second reason for classifying citizen types is that this categorization helps associate the attributes of voters. As Zaller suggests, the politically aware are different from the less aware with respect to four features: they are politically knowledgeable, likely to access political information, tend to resist political in-formation that is inconsistent to their political predispositions, and have greater ability to store preexisting considerations.6
Table on page 20 details how these differences are operationalized in the S-RAS model. First, the politically aware have higher levels of political expertise (from 6 to 10 on a 10-point scale) than the less aware (from 1 to 5). Second, the aware are more likely to access the news media (with probabilities of the variable Propensity to Access the News Media varying from 0.6 to 0.9) than the less aware citizens (with probabilities varying from 0.1 to 0.5). Third, the aware are more likely to discuss politics (with probabilities of the variable Propensity
to Discuss Politics varying from 0.6 to 0.9) than the less aware (with
proba-bilities varying from 0.1 to 0.5). Fourth, the aware tend to selectively perceive preferences (with probabilities of the variable Propensity to Selective Perception varying from 0.6 to 0.9) than the less aware citizens (with probabilities varying from 0.1 to 0.5). This means that the politically aware who hold the prefer-ence of “1” will likely to perceive “1” no matter what news media source they
6I assume that an agent’s probability of accessing the media is independent of the
probabil-ity of discussing politics, because there is not yet a conclusion about relationship between these two variables. Some scholars suggest a negative relationship. They think that individuals having weak party identification, higher education level, or feeling comfortable with discussing politics with other people are less likely to access news media (e.g., Mutz and Martin, 2001). The oth-ers suggest a positive relationship: talking to like-minded people is associated with perceiving media bias (see, Eveland Jr. 2003).
Table 2: The Differences between the Aware and the Less Aware in the S-RAS model
The Less Aware The Politically Aware Political Expertise (10-point scale) [1, 5] [6, 10]
Propensity to Access the News Media [.1, .5] [.6, .9] Propensity to Discuss Politics [.1, .5] [.6, .9] Propensity to Selective Perception [.1, .5] [.6, .9] Capacity to Store Messages 10 20
access. Finally, the aware store most recent 20 pieces of perceived vote prefer-ences (with the value of Capacity to Store Messages set to 20), while the less
References
Anderson, C. J., Blais, A., Bowler, S., and Donovan, T. (2005). Losers’ Consent:
Elections and Democratic Legitimacy. New York: Oxford University Press.
Beck, P. A. (1991). Voters’ Intermediation Environments in the 1988 Presiden-tial Contest. Public Opinion Quarterly, 55(3):371–394.
Bennett, S. E. (1998). Young americans’ indifference to media coverage of public affairs. PS: Political Science & Politics, 31(3):535–541.
Campbell, A., Converse, P., Miller, W., and Stock, D. (1960). The American
Voter. New York: Wiley.
Carmines, E. G. and Huckfeldt, R. (1996). Political behavior: An overview. In Goodin, R. E. and Klingemann, H.-D., editors, A New Handbook of Political
Science, pages 223–254. New York: Oxford University Press.
Cohen, J. (1989). The good polity: Normative analysis of the State, pages 17– 19. Oxford: Basil Blackwell.
Converse, P. (1964). The nature of belief systems in mass publics, pages 206– 261. New York: Free Press, New York.
Converse, P. (1990). Popular representation and the distribution of informa-tion. In Information and Democratic Processes, pages 369–90. Urbana, IL: University of Illinois Press.
Eveland Jr., W. P. (2003). The effect of political discussion in producing in-formed citizens: Roles of informtion, motivation, and elaboration. San Diego, CA. International Communication Association Annual Conference. OSU. Fiorina, M. P., Abrams, S. J., and Pope, J. (2005). Culture War? The Myth of a
Polarized America. New York: Pearson Longman.
Fishkin, J. S. (1995). The Voice of the People: Public Opinion and Democracy. New Haven: Yale University Press.
Granovetter, M. S. (1973). The strength of weak ties. American Journal of
Sociology, 78(6):1360–1380.
Huckfeldt, R. (2001). The social communication of political expertise.
Ameri-can Journal of Political Science, 45(2):425–438.
Huckfeldt, R., Johnson, P. E., and Sprague, J. D. (2002). Political environments, political dynamics, and the survival of disagreement. Journal of Politics, 64(1):1–21.
Huckfeldt, R., Johnson, P. E., and Sprague, J. D. (2004a). Political
Disagree-ment: The Survival of Diverse Opinions within Communication Networks.
Huckfeldt, R. and Sprague, J. D. (1995). Citizens, Politics, and Social
Commu-nication: Information and Influence in an Election Campaign. Cambridge,
England; New York: Cambridge University Press.
Huckfeldt, R. R., Johnson, P. E., and Sprague, J. D. (2004b). Political
Disagree-ment: The Survival of Diverse Opinions within Communication Networks.
New York: Cambridge University Press.
Kaid, L. L. (2003). Effects of political information in the 2000 presidential campaign: Comparing traditional television and internet exposure. American
Behavioral Scientist, 46(5):677–691.
Kim, J., Wyatt, R. O., and Katz, E. (1999). News, talk, opinion, participation: The part played by conversation in deliberative democracy. Political
Commu-nication, 16:361–385.
Miller, J. M. and Krosnick, J. A. (2000). News media impact on the ingredients of presidential evaluations: Politically knowledgeable citizens are guided by a trusted source. American Journal of Political Science, 44(2):301–315. Mindich, D. T. Z. (2004). Tuned Out: Why Americans under 40 Don’t Follow
the News. New York: Oxford University Press.
Mutz, D. (2002). Cross-cutting social networks: Testing democratic theory in practice. American Political Science Review, 96(1):111–126.
Mutz, D. C. (2006). Hearing the Other Side: Deliberative Versus Participatory
Democracy. New York: Cambridge University Press.
Mutz, D. C. and Martin, P. S. (2001). Facilitating communication across lines of political difference: The role of mass media. American Political Science
Review, 95(1):97–114.
Neuman, W. R., Just, M. R., and Crigler, A. N. (1992). Common knowledge:
News and the construction of political meaning. Chicago: University of
Chicago Press.
Nie, N. H., Verba, S., and Petrocik, J. R. (1976). The Changing American Voter. Cambridge, Mass.: Harvard University Press.
Oliver, M. B. (2002). Individual differences in media effects. In Bryant, J. and Zillmann, D., editors, Media Effects: Advances in Theory and Research, pages 507–524. Mahwah, NJ: L. Elbaum Associates, 2nd edition.
Patterson, T. E. (2002). The Vanishing Voter: Pblic Involvement In an Age of
Uncertainty. Alfred A. Knopf, New York: Distributed by Random House.
Putnam, R. (2000). Bowling Alone: The Collapse and Revival of American
Community. New York: Simon & Schuster.
Robinson, J. P. and Levy, M. R. (1996). News media use and the informed public: A 1990s update. Journal of Communication, 46(2):129–135.
Scherer, C. W. and Cho, H. (2003). A social network contagion theory of risk perception. Risk Analysis, 23(2):261–267.
Scheufele, D. A., Hardy, B. W., Brossard, D., Waismel-Manor, I. S., and Nisbet, E. (2006). Democracy based on difference: Examining the links between structural heterogeneity, heterogeneity of discussion networks, and demo-cratic citizenship. Journal of Communication, 56(4):728–753.
Smith, E. R. (1989). The Unchanging American Voter. Berkeley : University of California Press.
Straits, B. C. (1991). Bringing strong ties back in interpersonal gateways to political information and influence. Public Opinion Quarterly, 55(3):432– 448.
Wyatt, R. O., Katz, E., and Kim, J. (2000). Bridging the spheres: Political and personal conversation in public and private spaces. Journal of
Communica-tion, 50(1):71–92.
Zaller, J. (1992). The nature and origins of mass opinion. Cambridge University Press, Cambridge England ; New York, NY.
