deliberative democracy and public discourse: the agent ... · deliberative democracy and public...

Deliberative Democracy and PublicDiscourse: The Agent-BasedArgument Repertoire Model1

Encouraging Reasoning Abilities by Engaging inDirect Conversations about Substantive and Complex Issues

A gent-based modeling is a technique used to study relationships between varia-tion in parameter values or patterns of interaction at the micro-level andoutcomes at the macro-level. By using computer simulation of landscapes

inhabited by cells, or “agents,” the modeler can produce many virtual histories ofthe landscape under different initial conditions (randomized or not) and undervarious experimental conditions. In this article we report the findings of experi-ments run with the Agent-Based Argument Repertoire (ABAR) Model—experimentsdesigned to help answer some of the practical questions that arise in discussions ofthe contribution-enhanced public discourse, that is, more and better deliberation orargumentation among citizens might contribute to the quality of democracy [1].2

POSING QUESTIONS THAT CAN BE ADDRESSED BY AGENT-BASED MODELINGMost discussions of deliberative democracy focus primarily on its potential as avehicle for increasing political participation and the quality of democratic decision-making. They are linked to hoary civic and republican traditions and to research insociology, communications, and political science, identifying small group interac-tion as a fundamental transmission belt, and brake, on the effects of mass media.But the idea of deliberative democracy developed in the 1990s specifically in reac-tion to negative assessments of the quality of political discourse in the United States.With public debate seemingly dominated by huge amounts of money spent onmedia advertisements and by the politics of insult and personal scandal, mecha-nisms are being sought for encouraging citizens to exercise their reasoning abilitiesby engaging in direct conversations about substantive and complex issues. Suchpractices, it is argued, can counterbalance the emotion-laden images and slick slo-gans that dominate the mass media and displace public engagement with real issuesand the implications of collective choices for the public good.

Regardless of the specific mechanisms used (e.g., deliberative opinion polls, in-vigoration of associational life, monitoring and evaluation of campaign advertise-

IAN S. LUSTICK AND DANMIODOWNIK

Ian S. Lustick is Professor of PoliticalScience at the University ofPennsylvania, where he is a Chair ofthe Political Science Department. Heholds the Richard L. Simon Term Chairin the Social Sciences. His areas ofresearch include comparative politics,Middle East politics, and social sciencemethodology. Among his books isUnsettled States, Disputed Lands:Britain and Ireland, France andAlgeria, Israel and the West Bank/Gaza(Cornell University Press, 1993).

Dan Miodownik is a Ph.D. Student inthe Political Science Department at theUniversity of Pennsylvania. He holdsan MA degree in Political Science fromTel Aviv University in Israel.

© 2000 John Wiley & Sons, Inc., Vol. 5, No. 4 C O M P L E X I T Y 13

ments, programs of civic education,town meetings, cybercommunities), de-liberative democracy imagines that byexchanging views with one another,citizens increase their reasoned atten-tion to evidence [2–7]. Instead of rein-forcing existing prejudices and tenden-cies toward selfish regard, short-termthinking, and alienation from politicallife, such conversations are seen asproducing wider access to more rel-evant knowledge and as triggeringagreement on compromises that tap la-tent concerns for the good of otherswhile encouraging increasingly posi-tive-sum solutions to problems thatmight otherwise be treated in zero-sumterms. In its various forms it is widelypromoted as a device for reducing therole of shrill partisanship, ad hominemarguments, and emotional appeals.Citizens, it is hoped, will not onlythereby become better informed, butalso more able, and more inclined, tobecome active in the civic realm. AsRodin and Steinberg [8] have argued,democracy can more successfully copewith a variety of contemporary threatsto responsible and effective govern-ment by capitalizing on the “faculties”of reasoned discussion within “dis-cursive communities”—communitieswhose diverse leaders both representand guide their constituencies. By in-vigorating public discourse, they argue,the quality of public policy can be im-proved, the diversity of political society

respected, the capacity of democracy

for self-correction exploited, and the

boundaries of civil speech and conduct

established [9–14].

But some scholars have been skepti-

cal in response to this kind of argument

and exhortations of any kind toward

improvements in public discourse or

the introduction of various models of

deliberative democracy. For example,

Susan Stokes [15] and Adam Przeworski

[16] warn that deliberation can actually

reduce the quality of decisions and lead

to outcomes significantly more diver-

gent from participants’ interests and

available knowledge than without de-

liberation.3 Others have emphasized

the importance of clear operationaliza-

tion of the concepts so that normative

and empirical claims about public dis-

course and deliberative democracy

can be fairly evaluated. Thus, James

Fearon [17] advocates framing ques-

tions about deliberative democracy

in ways that could allow measurement

of the phenomenon and its effects. He

has suggested why it might be better

to focus on the concrete and measur-

able practice of “discussion” rather

than on the intangible evocations of

“deliberation.” Other analysts approach

the problem from the perspective of

social choice theory, wondering wheth-

er deliberative argument per se can help

solve various indeterminacy problems

in formal models of how preferences

can be aggregated to produce more ra-

tional collective decisions [18, 19]. Po-

litical theorists, such as Joshua Cohen,

Jurgen Habermas, and John Rawls, try

to think carefully about the meaning of

public deliberation for principles of

fairness, equality, and community that

they value, and as a vehicle for the pro-

duction of better citizens [20–22]. Public

policy-oriented analysts consider the

contribution deliberative democracy

might make to enhancing the efficacy of

policy choices by reducing the con-

straints of bounded rationality.

These are separable but overlap-

ping theoretical, philosophical, policy-

oriented, and civic-minded conversa-

tions. Many of the questions posed by

participants in these conversations are

not susceptible to investigation by the

technique of agent-based modeling or at

least not suitable for examination with

the use of our ABAR model. But we be-

lieve that a significant subset of these

questions can be effectively addressed in

this way, and that by thinking about pub-

lic discourse through the lens of agent-

based modeling, fundamentally impor-

tant questions arise that may otherwise

have been ignored. In this article the fol-

lowing questions will be addressed.

1. What are the implications of differ-

ent proportions of opinion leaders

and differentially sophisticated pop-

ulations on patterns of agreement,

The Agent-Based Argument Repertoire (ABAR) Model is used to explore key propositions about key relationshipsposited by students of deliberative democracy. Our experiments support the contention that democratic polities

can maintain stability, civility, and an effective problem-solving capacity by maintaining at least moderate levels ofeducation among citizens. We found that low levels of education, modeled as citizens lacking the ability to

understand more than a small fraction of the arguments being made by other citizens, are associated with veryhigh levels of pervasive disagreement and with the low levels of “agreement clustering”—the absence of the kindpersuasion that occurs among communities able to understand the arguments being made by the mass media and

by their neighbors. We also found that opinion leaders have multiple and important roles. When education levels inthe general population are low, even small proportions of opinion leaders decrease the pervasive amount of

disagreement and incivility in a polity. The presence of opinion leaders also significantly increases agreementclustering, thereby enhancing formation of discursive communities in a society threatened by atomization. When

education levels are very high, a “flip” occurs in the role played by opinion leaders. Very high levels ofsophistication tend to lead to levels of disagreement so low and stable that self-reinforcing communities of

agreement discourage citizens from changing their arguments, even when new conditions render those argumentsinvalid. We found, however, that the presence of opinion leaders improves the ability of well-educated citizens to

exploit their sophistication in response to mass media-provided evidence.

14 C O M P L E X I T Y © 2000 John Wiley & Sons, Inc.

disagreement, diversity, and changewithin a polity?

2. Under what conditions can diversitybe preserved and kept in balancewith levels of disagreement that areinvigorating but not polarizing andopportunities for consensus that makeeffective action possible?

3. What role do opinion leaders andvarying levels of education play inmaking deliberative democracy sus-tainable in volatile, risky, or turbu-lent situations?

It is worth emphasizing that the useof the ABAR model to simulate patternsthat may be taken to represent publicdiscourse or deliberative democracy,their requisites, or their hypothesizedconsequences should not be mistakenfor an attempt to replicate the intra-psychic experience or emotional resultof such behavior. Nor is this a techniquefor making point predictions about par-ticular political systems, such as theUnited States, or highly discriminatingjudgments about subtly different theo-ries of public discourse or deliberativedemocracy. Instead, this kind of simu-lation, the hypotheses considered, andthe results obtained should be under-stood as plausibility probes that can berelevant to “real world” events and pat-terns because of the streamlined, con-trolled, and precise results of the inter-action of a small number of crucialassumptions and variables. As the ques-tions listed above suggest, this model ofpublic discourse can help refine andtest expectations of a relationship be-tween enhancing capacities for thepublic interchange of views based onevidence and the character and efficacyof a democratic system.

A BRIEF OUTLINE OF THE MODELThe ABAR model imagines a polity as atwo-dimensional array of citizens. Thesize and shape of the landscape canvary, with boundaries and barriers dis-tributed according to the modeler’spreferences. Agents in the landscaperepresent citizens and are shaped assquares. Each has direct contact withthe eight other in its “Moore neighbor-hood”—those touching its four sides

and its four corners. Each citizen ap-pears as a particular color, correspond-ing to the argument that that citizen iscurrently articulating. For tracking pur-poses a number is assigned to eachcolor, that is, to each argument presentin the polity. At the beginning of a “run”the landscape is “reseeded,” randomiz-ing both the distribution of articulatedarguments and that of subscribed argu-ments. A citizen’s articulated argumentis one among a stipulatable number ofarguments that comprise each citizen’s“argument repertoire” (AR). The argu-ments in each citizen’s repertoire, in-cluding its articulated argument, com-prise a subset of the total number ofarguments present in the argument rep-ertoires of all citizens in the landscape.This number, of arguments presentacross the landscape, is also stipulat-able, though in the current version ofthe model the limit for this number is 20.

In accordance with classic theoriesof social impact, attentive publics, and

the relationship between mass opinionand small group interaction, there aretwo kinds of citizens in the polity: ordi-nary citizens (whom we shall refer tosimply as “citizens”) and opinion lead-ers [23, 24]. The exact proportion ofeach in any landscape is stipulatable forexperimental purposes, as is the size ofthe argument repertoire of opinionleaders. In a work summarizing a well-substantiated body of research, GabrielWeimann [24] characterizes opinionleaders as more influential than ordi-nary citizens, spreading their influencein interactions with intimate associates,somewhat more knowledgeable of theirenvironments, more “cosmopolitan”and “less dogmatic” than ordinary citi-zens, more willing to take risks, andmore apt to take information fromthose with whom they interact (pp. ix,57, 71–77, 84).

However, all opinion leaders haverepertoires that are larger than those ofcitizens. Opinion leaders also differ

FIGURE 1

Screenshot of ABAR model: argument view. Cylinder, 50 × 50, 15 Arguments, Citzens 6 argu-ments, Opinion Leaders 9, Opinion Leader density 5%, Bias Volatility 0.005, Range −2 +1.

© 2000 John Wiley & Sons, Inc. C O M P L E X I T Y 15

from citizens in the rules governingtheir interaction with neighbors. In ev-ery time period each citizen comparesits own articulated argument to thearticulated arguments of its neighbors.(The exact micro-rules are attached tothis article as the Appendix.) These cal-culations are performed synchronously,except that all opinion leaders maketheir calculations and implement theresults before citizens do. By adding thenumber of its neighbors articulated ondifferent arguments, each citizen deter-mines argument weights in its neigh-borhood. If the neighborhood argumentweight of the citizen’s articulated argu-ment is equal to or higher than the ar-gument weight of other available argu-ments, then no change occurs. The citi-zen’s articulated argument remains thesame. If the argument weight of anotherarticulated argument in the citizen’sneighborhood, which is also in its rep-ertoire, is significantly higher than theweight of its articulated argument, thecitizen changes its articulated argumentto conform to that with the greatest ar-gument weight. Its previously articu-lated argument remains in its repertoire.

The calculation of argument weightsis adjusted by two factors in addition tothe number of Moore neighborhoodcitizens articulated on individual argu-ments. In every time period each argu-ment present in the repertoire of anyagent in the landscape is assigned a biasvalue, indicating the amount of evi-dence for the validity of the argumentcurrently being made available to allagents in the polity by the mass media.This value changes randomly withstipulatable frequency (volatility) andrange (extent of possible change innegative or positive direction).

Calculation of argument weights isalso affected by the presence of opinion

leaders in the Moore neighborhood ofany agent. The argument articulated byan opinion leader is counted by agentsin its neighborhood as two, rather thanone. Opinion leaders, in other words,are figured as more influential than citi-zens. The algorithms determining opin-ion leader behavior also differ fromthose producing basic agent calcula-tions in that an opinion leader changesits articulated argument to one withinits repertoire after noticing even a slightargument weight difference. Likewise,although citizens can substitute new ar-guments for old ones in their repertoiresif the argument weight superiority of thenew argument is overwhelming com-pared to their articulated argument,opinion leaders are more sensitive to,and more ready to discard, argumentswithin their repertoires for new but at-tractive arguments. In sum, opinion lead-ers are modeled as having larger reper-toires, more influence, more sensitiv-ity to shifts in their incentive structure,

and more openness to changing theirrepertoires as well as their articulatedarguments.

Statistical monitors are used to trackchanges in the distribution of ar-ticulated and subscribed (within

repertoire) arguments over time and todescribe changing “tension” or “dis-agreement” levels. Tension levels modelaggregate amounts of disagreement inthe polity and are determined, at eachtime period, by counting the number ofencounters between agents articulatingdifferent arguments. The model can beviewed dynamically in two displays. Thearticulated argument display shows ex-pansion and contraction in the articu-lated presence of different arguments inthe polity, with each argument assigneda distinct color. In the tension display,the evolving history of the landscapecan be viewed in terms of locally pre-sent encounters between agents articu-lating similar or dissimilar arguments.

FIGURE 2

Screenshot of ABAR model: tension view. Cylinder, 50 × 50, 15 Arguments, Citzens 6 argu-ments, Opinion Leaders 9, Opinion Leader density 5%, Bias Volatility 0.005, Range −2, +1.

These meliorative results for thepresence of opinion leaders were

considerably more dramaticunder conditions of more rapidand sizeable changes in massmedia reports concerning the

validity of different arguments.


In this display, yellow represents highlevels of difference, red represents me-dium levels, dark green low levels, andlight green no difference. Figures 1 and 2show typical ABAR screenshots of an ar-gument and a tension level display of onerun of the model at 500 time periods.

EXPERIMENTAL RESULTSIn this article we report three groups ofexperiments—experiments focusing onthree measures of the character of ademocratic society. These three attrib-utes are of interest:

● The amount of publicly expresseddisagreement in a polity.

● The amount of clustered agreement,or consensus, in a polity.

● The amount of diversity in a polity.

I n the discussion section we explorethe implications of our findings forthe achievement of the kind of bal-

ance that supporters of deliberative de-mocracy seek in public discourse be-tween disagreement, coherence, andvariety.

In all our experiments we arrangethe model with two categories of set-tings. In the first category are those set-tings that we keep the same for all theexperiments reported in this article. Inthe second category are those settingswe adjust so their effects can be evalu-ated. The first category includes the sizeand shape of the landscape, or polityitself, the number of runs of the modelconducted for each experimental con-dition, and the number of time periodscomprising each run of the model.These settings were used for all experi-ments and will not be repeated in eachsection of the article. They are as follows:

Size and shape: The polity is arrayedas an unrolled cylinder whose verticaland horizontal dimensions are 50 citi-zens by 50 citizens. One hundred of thecitizens—50 along the top and 50 alongthe bottom—are replaced with imper-meable and inactive squares that serveas boundaries. The “sides,” however,“wrap around” and touch one another,as in a cylinder. This shape was chosenbecause it incorporates the notion of aborder separating the polity from theoutside world while also providing the

potential for continuous processes ofsequential interactions to spread acrossthe entire polity from almost everypoint in almost every direction.

Number of runs: Each condition ofeach experiment was run 20 times.

Length of runs: Each run of eachcondition in each experiment was runto 500 time periods. Extending thelength of these periods is likely to yield

somewhat different results in some ex-periments and under some conditions.For most experiments, however, experi-ence with the model suggests that rela-tively stable patterns have emerged bythe completion of 500 time periods.

DisagreementWhen citizens are said to “talk past eachother,” what is meant is that the argu-

FIGURE 3

Effect of repertoire size on disagreement level.

FIGURE 4

Effect of argument repertoire size and proportion of opinion leaders on disagreement levels.


ments made by citizens in the samecommunicative field, however coherentand important or wrongheaded andbaseless, are not being listened to orunderstood by other citizens makingother arguments. Deliberative demo-crats deplore such patterns, hoping toreplace them, not with a uniformity ofargument, but with a public discourseof disagreement and agreement basedon mutual intelligibility and evidence-based persuasion.

We use the ABAR model to studyvariation in aggregate disagreement lev-els by counting the number of encoun-ters between proximate citizens articu-lating the same or different arguments.Each citizen can articulate one argu-ment at a time. Each argument is drawnfrom the set of subscribed argumentsthat that particular citizen understandsat that time. His/her repertoire of un-derstood or subscribed arguments is it-self a subset of the whole set of argu-ments understood by at least one citi-zen in the population. Extremely highlevels of disagreement indicate a kind ofcacophonic public discourse in whichfew citizens are agreeing (articulatingthe same argument simultaneously).Extremely low levels of disagreementindicate a kind of totalitarian public dis-course in which one and only one argu-ment is being made by virtually allcitizens.

We began by examining how muchdisagreement would be exhibited in apolity as a function of the sophisticationof that polity’s citizens. We thereforecompared the amount of disagreementobserved in the polity after 500 time pe-riods at various levels of sophistication.By sophistication we mean the numberof arguments in the repertoire of a citi-zen in the polity.4 For most of our ex-periments we envisioned the polity ashaving a total of 15 arguments distrib-uted randomly across the population ofcitizens, although under no circum-stances could any one citizen have inhis/her repertoire all 15 of these argu-ments. We then observed levels of dis-agreement at each of 12 levels of citizensophistication—citizens “understand-ing” (having in their argument reper-toire) 1, 2, 3, . . . , 12 arguments. This

baseline experiment was run with noopinion leaders present. The bias set-tings were designed to be relativelystable. The volatility of evidentiary bi-ases supplied by the mass media wasset at 0.005 (meaning that for each ar-gument at each time period there is a0.005 probability that its bias value willbe eligible for reassignment). The rangeof possible variation in these bias levelswas also set fairly narrowly. No biascould be lower than 12 or higher than+1. Together these settings were used toproduce a “stable” (though not stag-nant) environment.

The results of this experiment aredisplayed in Figure 3. We see that atvery small repertoire sizes disagreementlevels are very high. In polities imaginedas containing citizens with more so-phistication (i.e., having larger indi-

vidual argument repertoires) disagree-ment levels drop. This decline is rapidat first and then moderates in its slopeas argument repertoires become larger.With high levels of sophistication (argu-ment repertoires that include the largemajority of arguments present in thepopulation), the decline in amounts ofdisagreement continues, but at a smalland diminishing rate. Overall levels ofdisagreement in the high sophisticationconditions are between one half andone third as low as those in the low so-phistication conditions.

These results are consistent with theview that even small amounts of educa-tion to marginally increase the flexibil-ity of a citizenry’s thinking could verysignificantly decrease the amount ofnonunderstood “shouting” of argu-ments that could be expected in a polity

FIGURE 5

Effect of argument repertoire size and proportion of opinion leaders (x, x+2) on disagreement ina volatile environment (bias, 0.05).

In general, our study of the relationship among general disagreementlevels, the extent of the clustering of agreement in the polity, and theamount of diversity of points of view available to the polity shows that

there are complicated trade-offs among these three values, that adesirable balance among them is more likely to be achieved in stable

conditions than in unstable conditions, and that in all conditions the roleof opinion leaders is positive, important, and roughly proportionate

(within low levels) to their proportion of the population.


unguided by a strongly biased mass me-

dia and deprived of a significant stra-

tum of opinion leaders. Investments in

such programs will continue to produce

results, but less robustly as the sophis-

tication of the population rises.

But most societies do have signifi-

cant proportions of opinion leaders—

individuals who are more sophisticated,

more influential, more attuned to the

media’s signals, and quicker to be per-

suaded by evidence. What are the im-

plications of the presence and propor-

tion of opinion leaders for the relation-

ship between citizen sophistication and

disagreement levels? Figure 4 compares

the results of making 5 or 10% of the

polity’s citizens opinion leaders.5 We

can see from this figure that the pres-

ence of opinion leaders does have an

effect—an effect that is proportionately

stronger in the 10% condition com-

pared to the 5% condition. Opinion

leaders accelerate the decline in dis-

agreement levels in relatively unsophis-

ticated populations. This effect declines

in robustness until it disappears alto-

gether between AR sizes 7 and 9. In the

higher ranges of citizen sophistication

the presence of 5 or 10% opinion lead-

ers produces slight increases in the

amount of disagreement.

In other words, there is a kind of

“flip” in which opinion leaders have

the effect or, as it were, play the role,

of reducing unproductive shouting

when narrow-mindedness produces a

public discourse with too much dis-

agreement and not enough persuasion.

However, when moderately high levels

of sophistication lead to relatively low

levels of disagreement (because citizens

can quickly reach agreement with one

another and then support each other in

maintaining their views), opinion lead-

ers serve as a kind of “yeast” that pre-

serves a measure of deliberative fer-

mentation in the public arena.

In this stable environment these ef-fects are of limited magnitude. In par-ticular, the amount of disagreement inpolities populated by very sophisticatedcitizens (AR more than two thirds of to-tal arguments in the polity) is only mar-ginally increased by the “fermenting”

presence of opinion leaders at eitherthe 5 or 10% level. In these conditions, itappears that the added influence, in-quisitiveness, and receptivity to evi-dence opinion leaders manifest can dolittle to prevent or reverse the tenden-cies of sophisticated citizens to formtight circles of reinforcing agreementwith their neighbors. This kind of “early

lock-in” appears to reduce opportuni-ties for citizens, however sophisticated,to hear different views from their neigh-bors, as opposed to simply receiving in-formation from the mass media aboutarguments they understand.

The question then arises as to theeffects of opinion leaders under condi-tions of instability in the environment.

FIGURE 6

Effect of argument repertoire size and proportion of opinion leaders (x, x+2) on disagreement ina risky environment (range: −2, +2).

FIGURE 7

Effect of argument repertoire size and proportion of opinion leaders (x, x+2) on disagreement ina turbulent environment.


Specifically, what effect does instabilityof different kinds have on the relation-ship between citizen sophistication anddisagreement or on the role and signifi-cance of opinion leaders for maintain-ing a healthy public discourse?

T o study the effects of instability inthe environment, or at least in themessages about the environment

transmitted to citizens by the mass me-dia, we distinguish between volatilityand range. Volatility refers to the paceof change, the likelihood at any time pe-riod that new evidence will appear withrespect to the validity of an argument.Range is about the size of the swings inthe biases assigned to specific argu-ments that can occur when new evi-dence appears. Thus, range pertains tothe weightiness of the evidence, not thelikelihood it will change. By increasingthe rate of change without increasingthe range of variation of assigned bi-ases, environments can be made to ex-hibit instability due to added volatility.By increasing the range of variationwithout increasing the pace of changesin assigned biases, environments can bemade to exhibit instability due, as itwere, to the increased possibility ofdrastic mistakes and rich rewards. Wedescribe this kind of less stable environ-ment as one characterized by increasedriskiness. When both volatility and riski-ness are increased, we describe the en-vironment as “turbulent” (though not inthe technical meaning of the term).

F igure 5 shows the results of experi-ments designed to test the effect ofthe presence of different propor-

tions of opinion leaders in populationsof varied sophistication under relativelyvolatile conditions. In these experi-ments the volatility of evidentiary biasessupplied by the mass media was set at0.05 (rather than 0.005 for stable envi-ronments), meaning that for each argu-ment at each time period there is a 0.05probability that its bias value will be eli-gible for reassignment. By comparingFigure 5 with Figure 4 we can see, per-haps surprisingly, that levels of dis-agreement at every sophistication set-ting are significantly lower under con-ditions of volatility than in stableenvironments. This occurs as a result of

an accelerated stream of opportunitiesfor citizens to discover that argumentsmade by some of their neighbors corre-spond to newly available validating evi-dence. In both figures, that is, understable and volatile conditions, we ob-serve that the overall shape of the dis-agreement curve, moving from less togreater citizen sophistication, is similar,but we see that the effects noticed un-der stable conditions are amplified un-der volatile conditions. That is, the ef-fect of very small increases in sophisti-cation in unsophisticated polities iseven more pronounced, and differencesin disagreement across moderately andvery sophisticated conditions are evensmaller. We also see that the “flip” inthe role of opinion leaders remains, al-though it takes place “earlier” (at AR 7,rather than AR 9). In other words, underconditions of volatility, and given amoderately sophisticated citizenry,opinion leaders act to foster greateramounts of public discourse rather than

to replace public disagreement withgreater consensus.

The next set of experiments consid-ered the effect, not of volatility, but ofan increased range in the absolute sizeof evidentiary biases assigned to par-ticular arguments. The range of eviden-tiary bias variation was increased from12,+1 to 12,+2. The effects of therebymaking the environment more risky,without making it more volatile, are evi-dent in Figure 6. We can see that theresults of this form of instability areroughly similar to that of volatility whencompared to the levels of disagreementunder different sophistication levels ina stable environment. But not only aredisagreement levels in the risky condi-tion lower than in the stable condition,they are also lower than disagreementlevels in the volatile condition. Indeedwhen citizens confront an environmentthat appears to be changing in relativelymore consequential ways with respectto the arguments they are making, we

FIGURE 8

Low agreement clustering. Cylinder, 50 × 50, 15 Arguments, Citizens 6 arguments, OpinionLeaders 9, Opinion Leader density 5%, Bias Volatility 0.005, Range −2, +1.


see that small increases in sophistica-tion from very unsophisticated levels(AR = 1 and 2) have even more dramaticeffects on the reduction of cacopho-nous disagreement than when instabil-ity is expressed as volatility. We also ob-serve that disagreement levels arrivecloser to their minimum at lower levelsof sophistication than in the volatilecondition.

Of particular interest is that in therisky condition, where the range ofvariation in evidentiary biases is wider,the flip in the effect of opinion leadersoccurs at a lower sophistication level(AR 4), compared to the volatility con-dition (AR 7). This would seem quiteconsistent with the interpretation thatopinion leaders help citizens find com-mon ground when that is the strategicproblem for the polity and help animateor sustain deliberation in response tonew evidence when the strategic prob-lem is that self-reinforcing groups havebecome so set in their opinions that

they resist the implications of evidencethat their arguments are incorrect. Wealso observe that with respect to theirrole as discourse encouragers, the scaleof the effects of the presence of opinionleaders in the risky condition is signifi-cantly greater than in the volatile con-dition. It is also worth noting that be-cause arguments can be relatively moreor less “wrong” or “right” in a risky en-vironment, the polity as a whole gainsmore from the presence of opinionleaders under risky conditions than inmore stable times. Another way of put-ting this is that in risky environmentsthe absence of opinion leaders is morelikely to lead to disturbingly high levelsof cacophony (when citizens are unso-phisticated) and disturbingly lower lev-els of public deliberation (when sophis-tication levels are medium or high) thanin stable or even volatile conditions.

When evidentiary biases reported bythe mass media are volatile and chang-ing within a wider range, that is, when

the environment is experienced by citi-zens as both volatile and risky, we mayconsider it as “turbulent.” Figure 7 dis-plays the results of experiments com-paring the effect of the presence of 0, 5,and 10% opinion leaders at different so-phistication levels under turbulent con-ditions (volatility set at 0.05; bias rangeset at 12,+2). Most striking is that al-though the overall shape of the curve ismore similar to the steadier rate of re-duction in disagreement characteristicof our study of stable environments, thecurve shows disagreement levels thatare 60 to 70% lower at every level of so-phistication. Thus, when volatility andrisk are combined, the polity experi-ences a rather drastic decline in delib-erative public discourse. Under turbu-lent conditions, and with no opinionleaders present, disagreement levelswhen the size of argument repertoires issmall (AR 2 or 3) are lower than thoseregistered in stable conditions evenwhen the AR is at its maximum (12).Again we see that opinion leaders flip“appropriately” and at low levels of citi-zen sophistication, from seeking to re-duce disagreement when citizens aretoo unsophisticated to understandmuch of what is being said to them bytheir neighbors, to promoting open-minded discourse when, after AR = 3,citizens appear locked-in to opinionsthey previously discovered they sharewith their neighbors.

Agreement Clustering

O ur measure of disagreement in apolity is quite general. As an in-dex, it registers gross amounts of

disagreement among citizens withoutindicating anything directly about itsspatial distribution or organization. Thereciprocal of a disagreement level wouldrepresent just as gross an index of “agree-ment.” However, we wanted to investi-gate determinants of the organization ofagreement. To accomplish this objectivewe use a statistical monitor built intothe program. The output of this monitoris a description, at each time period, ofthe number of citizens articulating eachargument being made by anyone in thepolity. This information is then pro-cessed into a report of the proportion of

FIGURE 9

Medium agreement clustering. Cyclinder, 50 × 50, Citizens 6 Arguments, Opinion Leaders 9,Opinion Leader Density 5%, Bias Volatility 0.05, Range −2 +1, 2 Arguments between 10%–20%.


the polity’s citizens currently articulat-ing particular arguments. The final out-put of the monitor is a tally of howmany arguments, at each time period,are being articulated by 0–10%, 10–20%,20–30%, 30–40%, and so on of citizens.Each run of each condition can then bedescribed by using the Herfindahl In-dex—traditionally employed to describethe extent of concentration in an indus-try or market. This aggregate index iscalculated as equal to the sum of thesquares of the market shares. If, for ex-ample, in a particular run, one argu-ment is being articulated by between 10and 20% of citizens and two other argu-ments are being articulated by between20 and 30%, then the Herfindahl indexfor that run is calculated by treating thefirst argument as articulated by 15% ofthe polity, and each of the other twoarguments as articulated by 25%. Thesquare of 0.15 is 0.0225. The square of0.25 is 0.0625. Doubling 0.0625 (because

there were two arguments made by be-tween 20 and 30% of citizens) and add-ing it to 0.0225, a Herfindahl index ofconcentration is calculated as 0.1475.6

We read the average Herfindahl in-dex scores for an experimental condi-tion as a report of the coherence andorganization of public discourse—adescription of what may be termed“agreement clustering.” Thus relativelylow values indicate, in general, a disor-ganized array of atomized individualsmaking large numbers of arguments ondifferent topics in faint or indistinguish-able regions. Medium values corre-spond to a small number of distinct andsizable groups surrounded by disorga-nized regions or a relatively large num-ber of smaller but coherent groups ar-guing directly with one another inhighly focused ways. High values indi-cate polarization of the polity’s citizensalong one or two cleavage lines. At theextreme, as values of the Herfindahl in-

dex approach 1.0, the polity is domi-

nated by one and only one opinion. Fig-

ures 8 to 12 show sample screen shots

of runs at 500 time periods typical of

these conditions.

As in our experiments regarding gen-

eral amounts of disagreement in the

polity, we were also interested in the ef-

fect citizen sophistication, the presence

of opinion leaders, and unstable envi-

ronments would have on variation in

agreement clustering. Figure 13 com-

pares the results of experiments that

trace the average Herfindahl index for

differing sophistication levels when

there are different proportions of opin-

ion leaders. We observe that in all con-

ditions a mild curvilinearity is observed

within Herfindahl index scores that are

low to medium, at most. Within this

range, the amount of agreement clus-

tering rises as low levels of sophistica-

tion increase from the random distribu-

tion of argument articulation from

which all runs originate. In each condi-

tion the amount of argument clustering

reaches a peak between AR = 3, when 10

or 15% of the citizens are opinion lead-

ers, and AR = 9, when there are no opin-

ion leaders. These peaks are then fol-

lowed by declines that tend to be more

precipitous when the proportion of

opinion leaders is lower.

A pparent here, again, is the change

in the role of opinion leaders. No-

tice how in the low sophistication

conditions the lines tracing changes in

amount of agreement clustering when

opinion leaders are present show values

that are higher than those in the ab-

sence of opinion leaders. Between AR =

4 and AR = 5, however, the opinion

leaders lines cross and begin registering

agreement clustering values that are

lower than those produced in the me-

dium and higher AR ranges when opin-

ion leaders are absent. Opinion leaders,

in effect, are fostering concentration or

aggregation of agreement—buildingcommunities of like-minded citizens—when citizens are unsophisticated andhave difficulty realizing any commonapproach to problems with their fel-lows. Then, as levels of aggregatedagreement increase, opinion leaders

FIGURE 10

Medium agreement clustering: low tension. Cylinder, 50 × 50, 15 Arguments, Citizens 6 Argu-ments, Opinion Leaders 9, Opinion Leader, Density 0%, Bias Volatility 0.05, Range −2 +1, 5Arguments Between 10%–20%.


switch to play the role of argument sus-tainers—encouraging citizens, as itwere, to avoid overcommitment to par-ticular arguments in the face of chang-ing evidentiary biases.

The steadying role of opinion leadersin a polity is even more striking underconditions of instability. Figures 14 and15 show, respectively, how heightenedvolatility and increased risk in the envi-ronment as presented to citizens by themass media significantly increase thelevels of agreement clustering at everycitizen sophistication level and in everyopinion leader condition. Again theriskiness factor has a more robust im-pact than volatility. But also note that inboth figures the change in the role ofopinion leaders is again evident. It maybe observed that in all of these condi-tions, and again in Figure 16 indicatingthe effects of turbulence (featuring bothvolatility and riskiness), that the rangeof variation in agreement clustering

across sophistication levels is verymuch reduced when opinion leadersare present, and smaller. If very highlevels of agreement clustering can betaken to represent an overweening con-centration of opinion, a kind of over-simplification of a polity’s response to acomplex and changing world, thenopinion leaders would seem to play acrucial role in holding these “panicky”tendencies in check. We may furtherobserve that although high levels of citi-zen sophistication do seem to reducethe tendency of unstable environmentsto encourage tendencies toward unifor-mity of opinion, this faculty loses its po-tency under turbulent conditions. Thus,in Figure 16 we see agreement cluster-ing reaching extremely high valueswhen opinion leaders are absent. Whenthey are present, however, they do pro-vide that steadying influence that, inthis condition, seems to prevent thepolity from freezing into an inflexible

and uniform stance toward a “frighten-ing” world.7

DiversityPrevailing treatments of deliberative de-mocracy value an enriched public dis-course not only because it can reducetensions and unproductive or vitupera-tive interaction among citizens and notonly because it can help create the basisfor coherent collective understandingsand joint problem-solving, but becausediversity of opinion and cultural style isheld to be a good in and of itself. Butfrom an evolutionary and complexitytheory perspective, this accent on diver-sity has its own powerful logic andrationale.

Worlds change in fundamentally un-predictable ways. The burning of Ama-zon and Southeast Asian rain forests area human tragedy, not because we knowhow plant and animal species there canhelp us now, but because they contain avast reservoir of species whose valuemay only be discoverable with futuretechnologies and only may be apparentin relation to problems that have not yetarisen.

The same principle can be applied toarguments. Some arguments may seem,and even be, chronically and foolishlywrong, but unless we deem them to ex-ist beyond some absolute moral pale,the community can be seen to benefitby their existence. The benefit is asso-ciated with the historical fact, and logi-cal possibility, that arguments and waysof thinking that are wrong today may beright, under vastly different conditions,tomorrow. Therefore political systemsand designs for enhancing public dis-course should reasonably establishmaintenance of diversity as one of threeobjectives—along with reduction in un-intelligent disagreement and promotionof opportunities for consensus—to bebrought into balance.

We have therefore used the ABARmodel to study the consequences of in-creasing sophistication levels, opinionleaders, and environmental instabilityfor maintaining diversity in the polity.To compare the effects of different con-ditions on the diversity of the polity wecreated a “diversity index.” The diver-

FIGURE 11

High agreement clustering: polarization. Cylinder, 50 × 50, 15 Arguments, Citizens 5 Arguments,Opinion Leaders 7, Opinion Leader, Density 0%, Bias Volatility 0.05, Range −2 +2, 2 ArgumentsBetween 30%–40%.


sity index is computed by counting thenumber of arguments being articulatedin a population at the end of a run (500time periods in these experiments) byno more than 1% of citizens. We thensubtract this number of effectively “ex-tinguished” arguments from the totalnumber of arguments that were beingarticulated at the beginning of the run.8

This number represents our diversityindex. A diversity index score for eachcondition is produced by averaging thediversity indexes resulting from each ofthe 20 runs of that condition.

The experiments we ran show thatunder stable conditions diversity re-mains high across all sophistication lev-els. When the citizenry of the polity isvery unsophisticated (AR = 1 to 4), thereis hardly opportunity for citizens tocome to understandings that changetheir articulated arguments from theones each articulated at the outset. Buteven in the medium and high AR ranges,

when citizens are able to communicateeffectively and respond to evidence of va-lidity for their arguments by collective ad-justments in their positions, relativelyfew arguments are reduced to such ob-scurity that the proportion of citizensarticulating them is less than 1%. As Fig-ure 17 shows, even with no opinionleaders the diversity index never dropsbelow 14.3 (with a maximum of 15).

In a relatively volatile environment,diversity levels do begin declining sig-nificantly as the sophistication level ofthe population increases above AR = 5.It is apparent from Figure 18 that thisdecline in diversity occurs with sophis-tication levels that are otherwise rela-tively desirable in that, as we have seen,disagreement levels are lower andagreement clustering levels higher inthis AR range. But we can also see againthat the presence of opinion leadersserves the polity well by protecting itagainst the drop in diversity at AR = 5and beyond that otherwise occurs.What we may say is happening is thatthe opinion leaders help insure the con-tinued availability of minority opinionsdespite chronically low evidentiary biasvalues. They can perform this role be-cause a few of them are able to use theirgreater persuasive powers to protectsmall groups of citizens articulating the

FIGURE 12

High agreement clustering: totalitarian. Cylinder, 50 × 50, 15 Arguments, Citzens 6 Arguments,Opinion Leaders 9, Opinion Leader Density 5%, Bias Volatility 0.05, Range −2 +2, 1 ArgumentBetween 70%–80%.

FIGURE 13

Effect of argument repertoire size and percentage of opinion leaders on agreement clusters.


minority view from overweening pres-sures to conform to the wider array ofarguments that, at one time or anotherin the volatile condition, are being por-trayed by the media as more valid thantheir iconoclastic views.

As we have seen earlier, the effects ofinstability are amplified in the risky

condition. Figure 19 shows how diver-sity levels dropped to between 11 and9.5, even in low sophistication levels. Inthe absence of opinion leaders, thediversity index continued dropping un-til bottoming out at 8 at AR = 6. Butwith 5 or 10% of the citizenry actingas opinion leaders, diversity indexes

began rising at the AR = 3 to 4 level,maintaining diversity scores of 11 orabove from AR = 5. Interestingly, highdiversity appears to have been main-tained, even in the absence of opinionleaders, when sophistication levels ofthe citizenry were extremely high (AR =11 to 12). This suggests that in a riskyenvironment the likelihood is high thatmost arguments will, before being ex-tinguished, enjoy at least a brief periodduring which their validity will bestrongly supported by signals comingfrom the mass media. If citizens areflexible enough in their thinking, theycan respond to these opportunities withenough facility to establish neighbor-hoods of reassurance capable of pre-serving their availability. It is possible,however, that further research mayshow that this effect could disappear ifthe time horizon is lengthened.

E vidence that this may in fact be thecase is available in Figure 20, re-porting results of experiments test-

ing the impact of turbulence on therelationships among sophistication,opinion leaders, and diversity. In theabsence of opinion leaders, diversity in-dex scores, after dropping in low andmedium AR levels to below 8, do notthen rise under conditions of citizen so-phistication. In fact they go even lower,dropping to between 7 and 4 in the AR =8 to 12 range. Under these both riskyand volatile conditions, the contribu-tion of opinion leaders to the preserva-tion of diversity is particularly striking.When 5% of citizens are opinion lead-ers, diversity levels are sustained above8 for all sophistication levels. When 10%of citizens are opinion leaders, diversitylevels stay between 9 and 11. It wouldappear that the neighborhoods of reas-surance that sophisticated citizensbuild to protect minority views in arisky environment are unable to survivewhen those neighborhoods are alsosubjected to high rates of change in theamount of evidence available to them.That is, when citizens are sophisticatedenough, high evidentiary biases will,with great probability, eventually triggerenough positive responses in the disfa-vored and minority opinion neighbor-hoods that the overall presence of that

FIGURE 14

Effect of argument repertoire size and percentage of opinion leaders (gap of 2) on agreementclusters in a volatile environment (bias, 0.05)

FIGURE 15

Effect of argument repertoire size and percentage of opinion leaders (gap of 2) on agreementclusters in a risky environment (range, −2,+2)


argument in the arena of public dis-course falls below the line of extinctionfor that argument (1%).

CONCLUSIONSCertain patterns appear in the results ofour experiments with the ABAR modelthat help address some of the questionsabout the potency with which civic edu-cation, reasoned discourse, and active

leadership can enhance the political fac-

ulties of deliberative democracy. We

found solid evidence that overall levels of

noncommunicative disagreement in

polities as modeled by ABAR tend to fall

sharply as the education level (meaning

increased sophistication in the under-

standing of more publicly heard argu-

ments) of the population increases. We

found that agreement clustering, indicat-

ing convergence among discursive com-

munities in response to evidence about

the validity of differing points of view,

rises when education increases from low

levels. We saw relatively clearly that there

is a trade-off in populations with low

education levels. Associated with high

levels of disagreement, and even atomi-

zation, which we may generally view as

undesirable, are high levels of diversity—

generally seen as desirable. But without

the capacity to communicate coherently

about changing aspects of the polity’s en-

vironment, undereducated citizens can-

not build the kind of substantial but not

universal consensus that communities

need to make use of the good ideas that a

diverse polity generates.

The challenge that emerges is to

achieve a balance among:

1. Disagreement levels that are low

enough to avoid cacophony but high

enough to insure vitality and flexibil-

ity in the polity as a whole; agree-

ment clustering levels that represent

robust communities of agreement

2. Coherent collective responses to per-

ceived challenges, and sustained at-

tention to particular points of view,

without effectively reducing the di-

versity of opinions available across

the polity

3. Diversity levels that are high, so as to

preserve options for an uncertain

future, while not sacrificing coher-

ence and community to an insis-

tence on maximizing the availability

of difference.

Significantly, we found powerful and

repeated suggestions that the presence of

even small percentages of opinion lead-

ers—citizens relatively more educated

than their fellows, relatively more at-

tuned to changes in their environment as

reported by the mass media, relatively

more responsive to rational argument,

and relatively more influential in delib-

erative encounters—contribute to the

attainment of each of these balances.That is, we found that for each dimen-sion of interest, namely disagreement,agreement clustering, and diversity,that opinion leaders militated against

FIGURE 16

Effect of argument repertoire size and percentage of opinion leaders (x, x+2) on agreementclusters in a turbulent environment.

FIGURE 17

Effect of proportion of opinion leaders (x, x+2) on diversity in a stable environment.


extreme values and did so under high orlow education levels. When too muchdisagreement, too little agreement clus-tering, and/or too much diversity waspresent, the addition of higher propor-tions of opinion leaders reduced the firstand third measures and increased thesecond. When the relative absence of dis-agreement, the emergence of high levels

of agreement clustering, or the disap-pearance of diversity signaled what canbe thought of as the “freezing” of the pol-ity, that is, higher proportions of opinionleaders tended to increase the first andthird measures and reduce the second.

Of particular importance is that whenthe modeled polities were stressed—whenthe volatility, riskiness, or turbulence of the

environment were increased—opinionleaders could make an even more dramaticcontribution toward maintaining the polityin the kind of balance that would preservepublic discourse and deliberation as aproblem-solving faculty of the democraticcommunity.

T here were also some more surpris-ing, but perhaps equally interest-ing and provocative, findings.

Chief among these is that although de-sirable levels of disagreement, agree-ment clustering, and diversity arehigher in very highly educated popula-tions than they are in populations withvery low education levels, generallyspeaking the most desirable balancesseem to be exhibited in the mid-rangeof education levels. There is some sig-nificant variation here. For example,when the environment is turbulent, di-versity levels fall dramatically in thehigh education conditions, whereasagreement clustering levels become ex-tremely, even dangerously, high (in theabsence of opinion leaders). But it isperhaps surprising that there is no regu-lar or linear relationship between in-creasing the education level of thepopulation as a whole and conditionsfavorable to, or reflective of, good pub-lic discourse.

T his may have to do with particularaspects of the model. Or it maysignal fascinating and nonintu-

itively obvious relationships betweenthe outcomes of interest and the preva-lence of apathy, fanaticism, exclusiv-ism, and the free flow of information ina society. These are all factors that canbe reasonably hypothesized to play a keyrole in the workings of deliberative de-mocracy. Although they can be pro-grammed as features of the ABAR model,they were not included in the experi-ments reported here.

For example, apathy and fanaticismexist in every polity. That is, there issome proportion of the population thatabsorbs information and experiencespersuasion without seeking to influencethose around them, and there is someproportion that aggressively seeks to ex-ert influence while refusing or stronglyresisting being influenced. If apathy andfanaticism act as breaks on the speed

FIGURE 18

Effect of repertoire size and proportion of opinion leaders (x,x+2) on diversity in a volatileenvironment.

FIGURE 19

Effect of proportion of opinion leaders (x, x+2) on diversity in a risky environment.


with which deliberative discourse canspread convincing arguments through apolity, their absence might tend to ex-aggerate the results of increases in theeducational level of the population oradjustments in other variables.

One additional promising avenue ofcontinued research worth mentioningis the question of finding the mix ofleadership, education, and heterogene-ity that does the best job of maintainingthe overall rationality of the polity. Byconsidering the bias values of argu-ments as indications of the changingcharacter of problems faced by the pol-ity, the polity’s “rationality” can be as-sessed, as well as its overall adaptive-ness. By summing the bias value of eachargument being made at a particulartime period or range of time periods,and by comparing this value to thatproduced by multiplying the number ofcitizens in the polity by the highest biasavailable in that time period or range oftime periods, an overall measure of ra-tionality, or efficacy, can be produced.This approach could help clarify think-ing about the efficacy of democratic de-liberation, its limitations, and its par-ticular strengths. These and other simi-lar experiments, including explorationof the implications of contradictory ar-

guments existing in the same citizen,will be possible as the new and morepowerful version of the ABAR model ismade operational.

Finally, it is worth noting that ouroverall approach to the study of publicdiscourse reflects and encourages anonstandard point of view. Most stu-dents of democratic deliberation focuson what citizens should be, what theydo or do not do, and how they respondto discursive opportunities or messages.We also ask questions about citizensand their attributes, but in the politiesas modeled in ABAR landscapes and inthe dynamic visual displays producedas virtual histories of the polities we fo-cus just as much if not more on argu-ments rather than individuals [25–27].Do arguments prevail, disappear, re-treat, or return? Are they available whenconditions suggest their validity? Whatdifference does the presence of more orfewer arguments within repertoires orwithin a polity make? Under what con-ditions do better arguments spreadmore quickly than poor arguments?9

Because polities are communities intime and not only in space, this per-spective may be particularly appropri-ate. Individual citizens are born, de-velop into maturity, and die. But the po-

litical community that joins them is

larger and lives on. What in fact lives on

are arguments—arguments comprising

a public discourse linking citizens to

one another in space, but joining them

as well to past and future generations in

a greater conversation that is the life of

the polity itself.

FIGURE 20

Effect of argument repertoire size and proportion of opinion leaders in a turbulent environment(both risky and volatile).

REFERENCES1. Lustick, I.S. J Artif Soc Soc Simul

2000, 3, http://jasss.soc.surrey.ac.uk/3/1/1.html/.

2. Fishkin, J. Democracy and Deliberation:New Directions in Democratic Reform.Yale University Press: New Haven, CT,1991.

3. Hall Jamieson, K. Dirty Politics: Decep-tion, Distraction and Democracy. Ox-ford University Press: New York, 1992,p 203–266.

4. Bessette, J. The Mild Voice of Reason:Deliberative Democracy and AmericanNational Government. The University ofChicago Press, Chicago: 1994, p 212–246.

5. Rosen, J. Getting the Connections Right:Public Journalism and the Troublesin the Press. Twentieth Century Fund:New York, 1996, p 1–6, 49–83.

6. Charity, A. Doing Public Journalism.Guilford Press: New York, 1995.

7. Black, J. (ed.) Mixed News: The Public/Civic/Communitarian Journalism De-bate. Lawrence Erlbaum: Mahwah, NJ,1997.

8. Rodin, J.; Steinberg, S.P. Public Dis-course and the Work of the Penn Na-tional Commission. Memo to Membersof the Penn National Commission, De-cember 14, 1998.

9. Gutmann, A.; Thompson, D. Democ-racy and Disagreement. The BelknapPress of New York: Cambridge, 1996.

10. Etzioni, A. The New Golden Rule: Com-munity and Morality in a DemocraticSociety. Basic Books: New York, 1996.

11. Miller, D. Political Studies 1995, 43,432–450.

12. Elster, J. In Deliberative Democracy. El-ster, J., Ed. Cambridge UniversityPress, Cambridge: MA, 1998; p 1–18.

13. Farrell, T. Norms of Rhetorical Culture.Yale University Press: New Haven,1993; p 187–275.

14. Nino, C. S. The Constitution of Delib-erative Democracy. Yale UniversityPress, New Haven 1996; p 1–14, 144–186.

15. Stokes, S. C. In Deliberative Democ-racy. Elster, J., Ed. Cambridge Univer-sity Press: Cambridge, MA, 1998; p123–139.


16. Przeworski, A. In Deliberative Democracy. Elster, J., Ed. Cambridge University Press, Cambridge: MA, 1998; p 140–160.17. Fearon, J. D. In Deliberative Democracy. Elster, J., Ed. Cambridge University Press: Cambridge: MA, 1998; p 63.18. Elster, J. In Deliberative Democracy: Essays on Reason and Politics. Bohman; Rehg, W., Eds.; The Mist Press: Cambridge, MA, 1997; p 3–34.19. Knight, J.; Johnson, J. Political Theory 1994, 22, 277–296.20. Cohen, J. In Deliberative Democracy; Elster, J., Ed. Cambridge University Press: Cambridge: MA, 1998; p 185–231.21. Calhoun, C., Ed. Habermas and the Public Sphere. MIT Press: Cambridge: MA, 1992.22. Rawls, J. Political Liberalism. Columbia University Press: New York, 1993.23. Katz, E.; Lazarsfeld, P. F. Personal Influence: The Part Played by People in the Flow of Mass Communications. The Free Press: New York, 1955.24. Weimann, G. The Influentials: People Who Influence People. SUNY Press: Albany: NY, 1994.25. Dawkins, R. The Selfish Gene. Oxford University Press: Oxford, UK, 1989; p 189–201.26. Dennett, D. Darwin’s Dangerous Idea. Simon & Schuster: New York, 1995; p 341–370.27. Blackmore, S. The Meme Machine. Oxford University Press: Oxford: UK, 1999.

APPENDIX

Rules Governing the Agent-BasedArgument Repertoire Model10

The Landscape: A population of square-

shaped agents (citizens) in a two-

dimensional array comprises a land-

scape (polity). The shape and size of the

landscape is stipulatable.

The Environment: The environment

of the polity has biases toward each ar-

gument present in the repertoires of

citizens in the population, that is, to-

ward all subscribed arguments. The set

of bias values toward argument x, B(x),

is stipulatable. Bias values for each ar-

gument change randomly and are re-

ported to all citizens by the “mass me-

dia” at a rate determined by an environ-

mental volatility setting.

Citizens: There are two types of citi-

zens who inhabit the environment: ba-

sic citizens and opinion leaders. Opin-

ion leaders (OL) are both more sensitive

to change in their environments and

more influential in their impact on citi-

zens in their neighborhood. Each citi-

zen (ordinary or opinion leader) has a

repertoire of C arguments. The ele-

ments of C will be chosen from a series

of D arguments (D # 20) such that no

two elements in C may be identical. Forthe ordinary citizen, the length of C willbe L (de facto 6); for the opinion leader,the length of C will be H (de facto 9),such that H > L. The first element ofrepertoire C is deemed the articulatedargument. The articulated argument isthe argument being expressed by thatparticular citizen to its neighbors. Eachcitizen’s arguments in C other than the

articulated argument are unknown to

other citizens.

During each turn, each agent inter-

acts with its Moore neighborhood of

citizens.

All opinion leaders act first and in

parallel. Each opinion leader looks tohis Moore neighbors and goes throughthe following process.

1. Argument weights for all articulatedarguments in the neighborhood, in-cluding the articulated argument ofagent OL in the center of the neigh-borhood, are calculated. The argu-ment weight (AW) of an argument ina neighborhood is equal to the num-ber of citizens in the neighborhood,including OL, articulated on that ar-gument, plus that argument’s envi-ronmental favorability bias, plus onepoint for each opinion leader in theneighborhood (including the agentin the center) articulated on thatargument. For example, if argumentx is articulated in 5 citizens, two ofwhom (including the citizen in thecenter of the Moore neighborhood)are opinion leaders, and if the envi-ronmental bias for that argument is11, then the AW for that argument is5 + 2 + (11) = 6.

2. If the argument weight of agent OL’sarticulated argument is equal to orgreater than the AW of any other ar-ticulated argument in its neighbor-hood, then OL’s articulated argu-ment remains unchanged, OL’s rep-ertoire C does not change, and theprocess ends.

3. If an argument, x, in C other thanOL’s articulated argument, has an

AW greater than that of OL’s articu-

lated argument, then x becomes OL’s

articulated argument. The formerly

articulated argument becomes a

nonarticulated argument in C. If the

AWs of more than one argument in

the neighborhood and within OL’s

repertoire are greater than the AW of

OL’s articulated argument, then the

argument with the biggest AW be-

comes the articulated argument for

E. If these AWs bigger than the AW of

OL’s articulated argument are equal,

then one of these arguments be-

comes OL’s articulated argument.

Which one? Answer—the one with

the largest subscription in the neigh-

borhood, then in the population, and

then the lowest digit between the

two argument labels.

4. If an argument, x, not in C, is articu-

lated with an AW at least 3 points

greater than OL’s articulated argu-

ment, then x is added to OL’s reper-

toire (but does not activate on that

turn) and the argument in OL’s rep-

ertoire with the lowest AW in the

neighborhood is removed from the

repertoire. If multiple arguments are

thereby candidates for removal, the

argument listed to the extreme right

in the cache is removed.

5. If an argument, x, not in OL’s reper-

toire, is articulated with an AW at

least 6 points greater than OL’s ar-

ticulated argument, then x becomes

an articulated argument of agent E,

the formerly articulated argument

becomes a nonarticulated argument

within OL’s repertoire, and an argu-

ment in OL’s repertoire is removed,


using the same procedure as in step

4 above.

After opinion leaders act, all basic

citizens act in parallel. Each ordinary

citizen A looks to his 8 neighbors and

goes through the following process.

1. Argument weights for all articulated

arguments in the neighborhood, in-

cluding the articulated argument of

agent A in the center of the neighbor-

hood, are calculated. The argument

weight (AW) of an argument in a

neighborhood is equal to the num-

ber of citizens in the neighborhood,

including A, articulated on that argu-

ment, plus that argument’s environ-

mental favorability bias, plus one

point for each opinion leader in the

neighborhood (including the agent

in the center) articulated on that

argument.

2. If the argument weight of agent A’s

articulated argument is equal to or

greater than the AW of any other ar-

ticulated argument in its neighbor-

hood, then A’s articulated argument

remains unchanged, A’s repertoire C

does not change, and the process

ends.

3. If an argument x in C, other than A’s

articulated argument, has an AW two

or more points greater than that of

A’s articulated argument, then x be-

comes A’s articulated argument. The

formerly articulated argument be-

comes a nonarticulated argument in

C. If the AWs of more than one argu-

ment in the neighborhood and

within C are two or more points

greater than the AW of A’s articulated

argument, then the argument with

the biggest AW becomes the articu-

lated argument for A. If these AWs

two points bigger than the AW of A’s

articulated argument are equal, then

one of these arguments, using the

procedure described in step 3, for

opinion leader citizens, becomes A’s

articulated argument.4. If an argument x not in agent A’s rep-

ertoire is articulated with an AW atleast five points greater than A’s ar-ticulated argument, then x is added

to A’s repertoire and the argument inA’s repertoire with the lowest AW inthe neighborhood is removed fromthe repertoire. If multiple argumentsare thereby candidates for removal,the removed argument is the argu-ment listed at the extreme right ofthe cache.

5. If an argument x not in A’s repertoireis articulated with an AW at leastseven points greater than A’s articu-lated argument, then x becomes anarticulated argument of agent A, theformerly articulated argument be-comes a nonarticulated argumentwithin A’s repertoire, and an argu-ment in A’s repertoire is removed,using the same procedure as in step4 above.

Initial Conditions: Citizens’ initial rep-ertoires are given with a uniform distri-bution and randomly. The percentageof opinion leaders in the population isset at the beginning of each run.

NOTES1. An earlier draft of this article was prepared

for the Penn National Commission on Soci-ety, Culture, and Community. The softwarefor the ABAR model was developed byVladimir Dergachev, Department of Math-ematics, University of Pennsylvania, in coop-eration with Ian S. Lustick. All experiments inthis paper can be replicated by downloadingthe model and using the manual available atthe sites listed in the article. Please sendcomments to [email protected].

2. This model was originally developed as theAgent-Based Identity Repertoire (ABIR)Model in order to test propositions aboutcollective identity and identity change thatcan be inferred from the constructivist theo-retical position. See Ian S. Lustick [1]; http://jasss.soc.surrey.ac.uk/JASSS.html/. Withonly minor changes in terminology, it hasbeen possible to translate the basic dynam-ics of this model into a device for studyingagents who understand and articulate over-lapping sets of arguments to learn abouttheir world and adapt themselves accordingly.

3. This can occur mainly because false beliefscan, via many forms of deliberation and dis-cussion, come to supplant true beliefs. SeeSusan C. Stokes [15] and in the same vol-ume Adam Przeworski [16].

4. It is important to remember that this notionof “sophistication” does not include someof the aspects of sophistication that com-mon usages of the term might suggest, in-cluding enhanced ability to distinguish be-

tween the validity of arguments that can beunderstood.

5. In these experiments opinion leaders haverepertoires that are two arguments largerthan ordinary citizens. If “x” is the size of theargument repertoire of an ordinary citizen,then “x+2” is the size of the AR of an opinionleader in this set-up of the model. “x+6”would indicate an AR for an opinion leaderthat is equal to the AR of an ordinary citizen,plus 6.

6. This is actually an adjusted version of theHerfindahl index since the identities occupy-ing between 0 and 10% of the landscapepopulation are not included in the sum ofsquares.

7. This steadying role of opinion leaders is es-pecially interesting when it is rememberedthat it arises, at the macro, or polity level,from behaviors at the micro (individual opin-ion leader) level, which are in fact more fluid,more “mercurial,” than is the behavior ofordinary citizens.

8. In the experiments reported here, 15 argu-ments were present at the beginning of eachrun. That means, with randomization pro-ducing the initial conditions, that each runbegan with each of the 15 arguments beingarticulated by between 6 and 7% of citizens.

9. This approach can accurately be interpretedas “memetic,” consistent with work by Rich-ard Dawkins [25], Daniel Dennett [26], andSusan Blackmore [27].

10. For a manual describing in detail how to op-erate the model, described in terms ofagents possessing repertoires of “identities”rather than “arguments” and “activated iden-tities” rather than “articulated arguments” goto MACROBUTTON HtmlResAnchor http://www.polisci.upenn.edu/profileil.html. TheABIR model itself can also be downloaded atthis site.

WEBSITEShttp://www.la.utexas.edu/course-materials/

government/32570/fishkin/fishkin.html

Professor James Fishkin’s project on

deliberative democracy polling.

http://jasss.soc.surrey.ac.uk/3/1/contents.

html

An application of the Agent-Based

Identity Repertoire Model to problems

of collective identity mobilization and

change. See the January 2000, Vol. 3,

No. 1, for the Lustick article.

http://www.upenn.edu/pnc/

Penn National Commission on Soci-

ety, Culture, and Community—a proj-

ect fostering improvement in the qual-

ity of public discourse.


deliberative democracy and public discourse: the agent ... · deliberative democracy and public...

Documents