©Copyright JASSS

JASSS logo ----

Pawel Sobkowicz (2009)

Modelling Opinion Formation with Physics Tools: Call for Closer Link with Reality

Journal of Artificial Societies and Social Simulation vol. 12, no. 1 11

For information about citing this article, click here

Received: 22-Nov-2008    Accepted: 09-Jan-2009    Published: 31-Jan-2009

PDF version

* Abstract

The growing field of studies of opinion formation using physical formalisms and computer simulation based tools suffers from relative lack of connection to the 'real world' societal behaviour. Such sociophysics research should aim at explaining observations or at proposing new ones. Unfortunately, this is not always the case, as many works concentrate more on the models themselves than on the social phenomena. Moreover, the simplifications proposed in simulations often sacrifice realism on the altar of computability. There are several ways to improve the value of the research, the most important by promoting truly multidisciplinary cooperation between physicists aiming to describe social phenomena and sociologists studying the phenomena in the field. In the specific case of modelling of opinion formation there are a few technical ideas which might bring the computer models much closer to reality, and therefore to improve the predictive value of the sociophysics approach.

Methodology, Agent Based Social Simulation, Qualitative Analysis; Evidence; Conditions of Application

* Opinion modelling - generalities

One of the major problems with 'social physics' or sociophysics literature, especially the exploration and understanding of social processes by means of computer simulation is the lack of connection to real life examples. In an early work in the field of opinion modelling (Nowak 1996) we find the following hope: 'Proponents of simulations view them as a powerful tool that will transform social sciences by bringing precision and rigour, into social theories'. Yet, despite the undoubted advances in simulation techniques and computer power, the transformation has not happened. Partially because modelling has became largely a self-sufficient exercise. In a recent Forum article in Journal of Artificial Societies and Social Simulation (Epstein 2008), the author lists seventeen reasons for using models, the three topmost being prediction, explanation and guiding data collection. All of these goals are related to real data and phenomena. Moss and Edmonds (2005) conclude their paper, titled Towards Good Social Science, with a list of pros and cons of the practice, with the advice for the good works starting with 'the fundamental priority of observation and evidence over models and theory', and the list of malpractices which starts with 'the unthinking and inappropriate use of analytic and statistical techniques'. In the years that have passed since this publication not a lot has changed.

This commentary is prompted by the generalisation of observations made while reviewing publications, which I am asked to do from time to time for several journals. I am privileged to be in a position of an 'interested outsider' to the field, no longer affiliated with any research institution, without the pressure of publication and career. My opinion may then be classified as a voice from the civil society, outside of Academia. Of course I lack the expert knowledge of all the developments in the field, so some of the active authors might consider my remarks unjust. I apologize to all those who do indeed combine, in sociophysics studies, the genuine social data with a resolve to provide an explanation through similarity with physical systems or through computer models. Unfortunately, too many papers written on opinion or consensus modelling (to stick with the branch of sociophysics I am the most familiar with) are totally decoupled from reality. The forthcoming review of the field (Castellano 2007), which presents quite detailed view of the models and theories shows how relatively limited are the attempts to compare the simulations and observations. In my personal experience the reactions to referee calls for closer connection with observed phenomena are also quite disheartening. In one case, the authors have openly stated that they were interested in the results of simulations as such, not with their link to reality. They have even changed the text of the paper, striking out the original token references to social phenomena, and leaving only blatant statement 'we study the model because it is an interesting model'.

Experience from the history of physics teaches us that the interplay between experiment (or observations) and theoretical models is crucial factor for success. To preserve the value that the 'physics' part adds to social studies, a return to this traditional experiment-theory balance is needed. There are a few ways to achieve this, some simple, some not so simple. The foremost is to strive for real multidisciplinary effort. Involvement of both physicists (knowledgeable about statistical mechanics, computer simulations etc.) and sociologists (with the knowledge about real situations where our understanding would benefit from the numerical models) in joint research teams is not always easy, but it seems necessary to keep the discipline alive and in touch with reality. In many cases it would also be fruitful to include biologists or neurologists into the teams. They would provide another perspective on these aspects of phenomena that are related to our biological functions. Such links may be very illuminative and sometimes quite unexpected to researchers outside the field. Just to point out an example: a recent study of Williams (2008) shows that there is direct, neurologically founded link between the physical warmth/coldness and the way people act towards others, the psychological feelings of warmth and coldness. Such studies show that sometimes it is necessary to broaden the outlook and the team as much as possible.

Unfortunately, in many cases the physicists are so fascinated with their tools that they forget to ask if the topic they choose to study—and the details of the model they use—are indeed interesting to the social scientists. As in 'traditional' physics, observations of societies in action, which are here the equivalent of experimental physics, are the ultimate litmus paper for the models and theories.

Of course, it is easier said than done. The cost and difficulty of making real social experiments or large scale observations is enormous, especially compared to computer programming. It is much faster to pick one of the fashionable topics, do some Monte Carlo simulations and write a paper about the results. Which is an important factor, especially in the age of 'publish or perish', and career based on the number of publications.

What do the such works concentrate upon? Quite often it is a derivation of an analytical formula for a particular simplified system, often borrowing some names from human society roles and functions to suggest direct applicability. The (ab)use of mathematical formulae in social sciences is still thought to be of supreme value, despite being rightly ridiculed for more than half century (Parkinson 1958; Andreski 1972). When there are no simple analytical expressions, the topic is often some nice observation of opinion changes, contrasting exponential decay versus fat tail distribution, or some other feature persistent in Monte Carlo simulations. In choosing which results are interesting, sociophysicists are often guided by folk psychology, and select as interesting those that agree with their own general expectations. Without checking if they are bringing any new insights to social sciences or not. Or even if their intuitions (and results) are supported by real observations, which are often at odds with naive expectations. I know, I have been guilty of the sin myself, and I have seen it numerous times in the literature.

When we see our simulations producing nice curves (especially ones that we can approximate by analytical formulas), when we see others publishing similar output of their computer programs, the imperative to publish becomes very strong. The main question ceases to be 'am I describing a real phenomenon correctly and in sufficient detail? ', and turns to preparing the arguments to answer the eventual question of the journal Referee: 'does the paper contain enough new physics to be published in XXX? '

Despite the possible unpopularity of the idea in the sociophysicist community, I would go as far as to postulate that the leading role in such collaborations should go to the sociologists or psychologists. They are at the forefront of the research, studying the real life phenomena. Physicists provide the tools, and, at times, interesting insights and working models, but their role is clearly subservient. After all, nobody (I hope) believes that humans are mindless cellular automata governed by a few simple rules and characterised by a few numeric parameters. Such simplification might be useful in certain circumstances but to put it above the behaviour of real people in real situations is preposterous. Modelling may be very valuable, especially when it provides new understanding, new insights, points out new questions and directions for research. But it should always remain modelling of some real situation or process, and it is this process or situation which is the ultimate goal of activity.

* Technical improvements

In addition to the need for better ties with observational data and promoting truly multidisciplinary way of conducting research, there are as well some technical aspects of the modelling approach that could adapt them better to realistic conditions. A good example of a branch of sociophysics that would greatly benefit from rather simple changes is the study of opinion formation. The field has a long tradition of studies using methods derived from physics. A few of the works, coming from author's rather small collection are listed in the Appendix, where some characteristics are summarized.

Despite this limitation, there are some specific examples where the approaches could be improved. They may be grouped into two classes: Both groups offer relatively straightforward ways of improving current state of research.

Social network modelling

Let's consider network interactions first. The early approaches have used very simplistic connection geometry, usually square lattice with nearest neighbour interactions. Keeping such geometry was fruitful in in the segregation model (Schelling 1971), because of its direct link with physical, 1D or 2D neighbourhood, and computationally justified in early works (e.g. Nowak 1990). With the introduction of small world idea (Milgram 1967) and associated shortcut network topologies (Watts 1998; Watts 1999; Strogatz 2001) a new 'science of the networks' has been born. This has resulted in enormously rich literature (some examples of review papers are Albert 2002; Barabasi 2001; Dorogovtsev 2002; Dorogovtsev 2003, Newman 2000; Newman 2003). The advantage of the new network models (Strogatz-Watts, small world, scale free etc.) is that they represent much better the actual relationships between people and groups.

Many of the new papers on the spread of opinions in populations do take into account these non-trivial connection topologies, and have led to interesting results. For example, the existence of small-world type shortcuts allows information to spread quickly through the society. Local opinions are capable of influencing distant parts immediately. In scale free networks the hubs—highly connected individuals—can influence a lot of other members of society. The ease or difficulty of convincing such highly connected agents is often crucial to achieving consensus. Another aspect brought in by the new generation of models is due to non-uniform nature of connections, which allows more variability in describing local environments. In summary, these simulations may be much closer to the real world.

Yet while the step from simple, geometric approaches to a networked one seems obvious, there are still quite a few papers that use the inflexible, nearest neighbour connection scheme. This is probably due to the simplicity of encoding the 2D geometry into computer models—compute power was an issue twenty years ago, but certainly not today. In my opinion, any research aiming at description of complex human societies should opt for these more advanced interconnection topologies. The use of 'flat earth' can be justified only in cases, where for real reasons, the actual interactions take such form. While rare, such situations do exist and might be worth studying: for example the spread of emotional states and spiteful or cooperative behaviour in traffic jams, where each driver sees and reacts only to the nearest, visible neighbours.

The improvement brought by structurally rich networks is already enormous. The next step is to change the view from static to dynamical. In our societies, connections are continuously created, change their strength, sometimes they are abruptly severed. The models should include such time dependent phenomena. Consider, for example, how often the reaction to an encounter with a person of opposite opinion is to cut off the links to that person, without any of the participants changing their opinions. This process has much deeper consequences than the typical actions of averaging of the different opinions or forced conviction, decided by relative strengths of the agents. The adjustments of the network topology, connectedness between subgroups, flow of information through the network can profoundly change the results. It may turn that some effects would be far easier to explain within dynamic networks than when the model uses dynamic agents in a static network. Of course some links can not be broken (family and workplace relationships, for example). But many others can and are. Supporters of different political parties rarely indulge in an effort to convince each other. Much more often they tend to close the circle of friends and acquaintances, to talk to people who support their views.

One could expect that models with dynamic, reacting networks would show interesting effects in studies of persistence of minority or extreme opinions; preserved not because of particular strength of the believers but because of their tendency to sever as many links with majority as possible. Such systems with dynamically changing connectivity of the network were considered by Gil (2006) and Zanette (2006), however the agent properties and behaviours in their model were very simple.

Time-dependent topologies provide quite interesting way to describe medium-sized groups, where statistical methods are less applicable. At the same time, sociological observations and experiments is more feasible for such small and mid-sized groups. It might be interesting to see if the predictions of agent-based models would describe better the stability of minority groups and provide suggestions as to the methods and outcomes of their re-integration into the main society, e.g. by enhancing the links that are, by their nature, less affected by opinion differences. Such approaches are historically known to be quite effective, in almost all human societies (for example in forms of pacifying bonds forged by intergroup marriages). Yet recent history provides examples that such links might be insufficient if other factors are simply too strong. The ethnic conflicts that erupted after the disintegration of Yugoslavia, despite many years of extensive interlinking between communities during Tito's years can be a very interesting ground for modelling studies. Another field where suitable and interesting data could be used might be in time-flow communities, such as schools, where easily recognized and monitored imposed links (such as division into classes and age groups) intertwine with temporary coalitions. Experiments with consensus formation and information acceptance could provide a good comparison for various theoretical models and suggestions for further development.

There is an interesting parallel between such dynamic social networks and consensus studies and certain models of neural correlates of psychic phenomena. In some approaches, the time dependent, variable 'coalitions' of neurons, excited together and corresponding to mental states are vying for the dominance in overall brain activity (see, e.g. Chialvo 1999; Chialvo 2004; Sporns 2000; John 2001). This similarity may, perhaps, be quite stimulating in terms of concepts and methods.

Interaction modelling

Second group of proposed improvements relates to the description of the agents, their characteristics and interactions. In modelling human societies we are faced with ostensibly insurmountable task of compressing the multifaceted human nature, driven by individual histories and almost infinitely rich in motivations and action choices by a simple mathematical or programming entity, with a limited repertoire of actions and very basic description. The justification for such simplification comes, fortunately, not from limitation of sociophysics, but from empirical evidence from psychology and sociology. Yes, in certain circumstances one can expect people to react in predictable, rather simple ways. Thus, it is possible that a study using computer based agents would yield sensible and valuable results. The crucial questions become: in which situations can we use such simplifications and how to choose which parameters should describe the agents and their interactions? In particular, what should be the set of descriptors needed in studies of opinion changes?

As we see, the choice of models is already quite rich. Moreover, in addition to agent characteristics, we should also consider variations of the interactions process. What changes during an encounter—is it only agent's opinion? For example, an encounter may lead to increase or decrease of its strength, susceptibility or persuasiveness. How many steps are in the process leading to individual opinion change? Should the process be broken into steps of information exchange, evaluation and positioning? Are there any thresholds (similar to neuron excitation threshold) that trigger any change? How should the Monte Carlo simulations represent human interactions, what is the relation between 'computer time' and reality? Do we really meet each other thousands of times that seem to be needed to achieve stability in computer experiments? Or should the real processes of opinion formation, where we meet just a few people just a few times, be described using transient, nonequilibrium states? The last question is particularly important, as these transients are the very parts of simulations that nobody wants to talk about, because they depend on particular initial conditions. But this is exactly the issue that makes social sciences interesting and difficult—how to distinguish between universal and historical view. In reality a lot depends on the initial conditions and historical accidents.

It seems that one of the interesting and almost untouched ideas is to take into account the economics of changing the opinions of others. Simulations of cooperation and aggression in societies have used successfully the pay off and cost of attitude detection or cost of retribution for a long time. Thus the step is rather natural and obvious for consensus formation—yet it is seldom used. Including such cost calculation in simulations is rather straightforward, and easily incorporated in MC procedures. Such considerations would be especially important in modelling directed changes, i.e. changes brought by individual leaders and organizations. Measuring the effect of different strategies for using available resources to achieve the desired level of consensus would make sociophysical studies much more interesting and close to reality of commercial advertisement or political propaganda.

It should be noted that above certain limit of complexity the models cease to be useful, because it is just as difficult to understand them as it is in real social situations. The balance between the number of parameters that can be varied in a model on one hand, and, on the other, its resemblance to reality and predictive power must be preserved. And the decisive factor would not be the preference for this or that model, for example continuous vs. discrete variables, but comparison with observations. Making the models more and more complex just for the sake of adding one more parameter and making the paper fit for publication does not seem in the spirit of physics any more. On the other hand, it should be noted that even complex models can be analysed in more depth than real social situations. For example, by meticulous checking of the role of parameters through repeated simulations—something that is not possible in real life. However, from the point of view of an outside observer I would much prefer simpler models, but ones aimed at explaining a behaviour of a real system, where the meaning of parameters can be understood. A perfect—and very funny—example is provided by the analysis of scientific publications, with the observation that, quite probably, a lot of scientists do not read the papers they cite. This real world phenomenon has allowed to propose a very simple computer model of 'randomly citing scientist', which has reproduced, remarkably closely the actual distribution of frequency of citings (Simkin 2003; Simkin 2005a; Simkin 2005b; Simkin 2006). The combination is an almost perfect example of what sociophysics should be: direct link to real world, a simple model where the role of parameters is understood, close resemblance between simulation results and observations. Thus, it is possible to go beyond the ghetto of formulae and Monte Carlo runs. Simkin's model, simple as it is, has given a valuable insight into the observed behaviour. This proves that social modeling is most useful in analyses of repeatable phenomena and their statistical properties, not in individual events.

Returning to opinion formation studies and the growing number of models: only empirical verification can allow to decide which model is the most appropriate for which situation. The brief review, assembled in the Appendix, shows that only a small fraction of the works even attempt to look at the reality. So much more praise should go for the authors who take this important and difficult step. But for the others, I think that a change in overall atmosphere is needed, even though it might be considered hurtful for existing research projects. The journal editors and referees should stress the need to go beyond studying models for model's sake each time the consider publishing a manuscript; funding agencies should promote truly multidisciplinary cooperation between different groups of researchers. The success of such cooperation in other applications of physical models to social phenomena (for example to understanding traffic patterns) shows that it can be done, and that physics can be valuable tool to expand our understanding of humanity. At the same time the recent financial crisis has shown that coupling of physics-derived financial models with human greed and other 'un-mathematical' behaviours does not lead to predictable results, teaching us more caution in blind belief in statistical tools and models. The same caution should be used for opinion studies.

* Appendix

In the Appendix, I have aimed at synthetic presentation of the characteristics of a small selected subset of the literature on the physical modelling of opinion formation. The list of papers used for this analysis is not meant to be exhaustive, as the present author is only an amateur with limited view of the field, and inclusion or omission of particular paper does not represent any judgement on the value.

There are several 'mainstream' models used in modelling consensus formation. Among the most widely used are Sznajd model, Deffuant model, Krause-Hegelsmann and other models (Sznajd 2000; Deffuant 2000; Hegelsmann 2002). Below is a very short summary of the main models.
Sznajd model
considers a population of agents with discrete (+1/–1) opinions oi . The dynamics of opinions in the model is based on an influence of pair of neighbouring agents, with two basic rules: 'ferromagnetic' (if the opinions of agents oi and oi+1 are the same then opinions of neighbouring agents oi-1 and oi+2 will be changed to oi ) and 'antiferromagnetic' (when the opinions oi and oi+1 are different then an antiferromagnetic pattern forms, oi-1 = - oi = oi+1 = - oi+2 ). The model has been originally proposed for 1D geometry, later extended to 2D and more complex network topologies. Some variants use only the 'ferromagnetic' part of the interaction, assuming that if the opinions of the starting pair of agents differ, they have no influence on their neighbours. In Sznajd model, the opinion flows out from a group of agreeing agents, rather than in, from the environment to an agent, but in some cases the results are similar.
Deffuant model
uses a population of agents with continuous, bounded range of opinions. At each time step any two randomly chosen agents interact (global connectivity). They re-adjust their opinions when their difference of opinion is smaller in magnitude than a threshold. When the two agents have opinions oi and oj then if the difference of opinions is small enough, | oi -oj |< d, after the encounter their opinions would be, respectively o'i = oi + μ (oj -oi ) and o'j = oj + μ (oi -oj ) , where μ is the convergence parameter taken between 0 and 0.5 during the simulations.
The strength of interactions between agents is given by an array of parameters sij , so that at every interaction step the agent's opinion is modified to o'i = gi oi + (1 - gi )∑ j (sij oj ), where gi is the agent's susceptibility. The interaction here may be considered global, or, if proper limitations of the strength of influence matrix are imposed, it may represent any type of the network. In the original papers a stochastic form of this matrix has been used, with normalisation condition .
There are quite a few variants and combinations of the main models mentioned above. The opinions might be multidimensional, and the processes of individual opinion change due to an encounter or influence of other agents might be based on the interplay between various 'interest points' (dimensions of the opinion vector). In addition to the strength of the influence some other characteristics of an agent have been proposed, for example 'susceptibility' and 'persuasiveness'. Some papers consider the role of special agents, for example extremists (agents with unchangeable opinions, in the continuum models usually located at the far ends of the spectrum of opinions), or leaders (agents 'equipped' with particularly large values of strength or otherwise capable of playing a crucial role). Some models add general biases (advertising influence) to explain the reaction of the society to external influences.

The network aspect of the models is also very differentiated. Starting from simple Euclidean geometry (usually discretized) in one, two or more dimensions, through globally connected meshes to very advanced network topologies. Then the links may have special attributes, influencing the interactions between agents. In some models all encounters, when they are allowed are in principle uniform. In other models, the 'social distance', modelled by network properties changes the outcome of encounters.

The variety of the models shows how active is the research field, but, at the same time there is no consensus as to which model is 'the best'. It is quite natural, as most likely various models would offer advantages for different social phenomena. Only comparison with social, empirical data would enable us to judge if adding further refinements is necessary or not. But such comparison seems to be quite rare. The following Table attempts to characterise the a small subset of publications related to the issues discussed in the main part of the paper. Perhaps such summary would be useful in showing a bird's eye view of the techniques used and possibilities to improve the methods. For a recent review and richer literature subset see Castellano (2007).

Table 1: Selected examples of opinion modelling works

ReferenceMain topicTopologyOpinion modelAgent characteristicsLink characteristicsSpecial featuresComparison with real data
Behera 2003comparison of Sznajd and Voter models1DSznajd model, Voter modelopinion (discrete)NN, NNN, uniform strengthinfluence of general bias; synchronous vs. asynchronous updatenone
Bernardes 2001scaling of the cluster growth2DSznajd modelopinion (discrete)uniformmodified Sznajd model used to compare with empirical data; analysis applied in transient regimeyes, Brazilian elections
Bernardes 2002explanation of election results and the Sznajd model on Barabasi network3D cubic; scale freeSznajd modelopinion (discrete)uniformcomparing tempo of equilibrium in 3D and network topologiesyes, Brazilian elections
Caruso 2005Multidimensional analysis of contexts for decisions with local agent interactions but also with effects of global factorsglobal connectivitybipolar, with number of agents of fixed opinion (RC,LC) and undecided (CG)payoff based on multidimensional context vectorglobal, with strengths of interaction dependent on type of agentsome agents form coalitions that do not change their opinion; general bias influenceyes, Italian and German elections
Conradt 2003decision making in animalsglobal connectivitybinarybased on costs of synchronization of actions, also with incomplete informationglobal, uniform strengthcontrast between `despotic' and `democratic' groupsyes, many animal groups, some references to human societies
Conradt 2005review of consensus decision making in animalsmany models mentioned in a reviewmany examples of animal behaviour
Deffuant 2000presentation of consensus model with continuous opinion spectrumglobal connectivity; 2D squarecontinuous, adjusted if opinions are closer than certain thresholdopinion, threshold for agreementuniform strength; global connectivity or NN compareddiscussion of multidimensional opinions (vector)none
Deffuant 2002consensus model with analysis of uncertaintyglobalcontinuous, adjusted if opinions are closer than certain thresholdopinion, threshold for agreement, uncertaintyuniform strength analysis of extreme opinions survivalnone
Deffuant 2004modelling opinion shift to extreme valuesglobalcontinuous, adjusted if opinions are closer than certain thresholdopinion, threshold for agreement, uncertaintyuniform strengthspecial agents (extremists) with opinions located at edges of the spectrum and low uncertainty are used some reference to psychological data, no comparison
Dyer 2008consensus in human crowdsno simulation model introducedexperiments on crowd behaviour
Elgazzar 2002opinion formation on small-world networksmall worldSznajd model opinion NN and shortcuts, uniform strength none
Fortunato 2004bnetwork universality of confidence parameter dependenceglobal, 2D lattice, random network, scale free networkDeffuant model, continuousopinion, threshold for agreementuniformcomparison of network topologiesnone
Fortunato 2004acombination of Sznajd and Deffuant models2D contrasted with scale-freemodified Sznajd modelopinions, continuousNNweak and strong versions of opinion change are discussesnone
Fortunato 2005comparison of Krause-Hegselmann, Sznajd and Deffuant modelsscale freevarious versions of characteristics and dynamics, corresponding to the models being comparedreactions to extreme events: external modification of single agent's opinion and subsequent dynamicsnone
Fortunato 2007simple `word-of-mouth' influence model used to explain election results universalityhierarchical networkunidirectional influence candidate preferencehierarchical, unidirectionaldiscovery of universal features in actual election results and simple fit with `word-of-mouth' modelyes, many examples of electoral results
Galam 1997achieving consensus through minimization of individual conflictsIsing-like model; agents interact with a fixed fraction of all agentsstatistical analytic considerations, no simulation; agents characterised by opinion, discrete; social pressure; internal preferencenone
Gil 2006opinion formation on evolving networksinitially global; as results of interactions links are cut, topology changes opinions adjusted with probability p1, links cut with probability p2opinion (binary)uniformformation of separated clustersnone
Hegelsmann 2002introduction of Krause-Hegelsmann model, globalKrause-Hegelsmannopinion, time dependent influence matrix, susceptibilityweighted interactions with other agents none
Holyst 2001social impact of strong leaders2D Nowak-Latane model, with strong leadersopinion, strength of influence, external influence, social `temperature'weighted interactions with other agents, strength decreasing with physical distancecomparison between simulations and analytical `mean-field' approximationsnone
Kacperski 1999social impact of strong leaders2D Nowak-Latane model, with strong leadersopinion, strength of influence, external influence, social `temperature'weighted interactions with other agents, strength decreasing with physical distancestudy of effects of flips of leader's opinionsnone
Kacperski 2000phase transitions of global opinions for groups with leadersglobalNowak-Latane model, with strong leadersopinion, strength of influence, external influence, social `temperature'weighted interactions with other agents, strength determined by arbitrary `adjacency matrix'presence of universal `phase transitions'none
Lewenstein 1992mean field theory of social impactcomparison of global interactions, sparse network, hierarchical network, 2D latticemodified Nowak-Latane modelopinion (binary); persuasiveness; supportivenessinteraction strength weighted by `social distance'comparison of various geometries; analysis of effect of noisenone
Lorenz 2007survey of results under Krause-Hegelsmann and Deffuant modelsglobal, although discussion of social network dependence is includedKrause-Hegelsmann and Deffuantopinion (continuous, multidimensional)uniform for defiant model, weights in the Krause-wheelsman modelcomparison between agent based models and density models (which can be interpreted as limit case for infinitely many agents)none
Nowak 1990simulations for the social impact theory of Latane2DNowak-Latane modelopinion (binary); persuasiveness; supportivenessinteraction strength weighted by `social distance' some remarks on qualitative social opinion phenomena, no direct comparison with simulations
Nowak 1996general introduction into social modelling; dynamic social impact2D, some discussion on 1D caseNowak-Latane modelopinion (binary); persuasiveness; supportiveness; self-supportivenessvariable strength of interaction, decreasing with distance none
Pluchino 2005Opinion Changing Rate model introductionglobalOCR model, stressing dynamic aspects over equilibriumopinion (continuous, unbounded); natural opinion rate change; variable strength of interaction, depending on opinion differencefocus on differences in tendencies to change opinionsnone
Roehner 2005supplementing computer opinion models with real world data presentation of `experimental data' with some discussion of the ways of information spread through formal mass media channels as well as through informal channels (rumours, hearsay, gossip) and some macro-constraints data on consensus formation in various countries for using cell phones while driving
Sabatelli 2003aaverage opinion dependence in Sznajd model with noise2DSznajd model with noiseopinion (binary)uniform, NN none
Sabatelli 2003brole of update mechanism (synchronous vs. async) and memory in Sznajd model2Dmodified Sznajd modelopinion (binary)uniform, NNsynchronous updating role discussesnone
Schulze 2004results for Sznajd model with global and local interactions2Dmodified Sznajd modelopinion (discrete, with Q available options)uniform, global for comparison of opinions, the local for conversionsAdditional analysis of advertising bias for one of opinionsnone
Schweitzer 2000modelling collective opinion formation by means of active Brownian particles2D, with moving agentsNowak-Latane modelopinion (binary); social temperature; strength of influence; self-support; global bias interaction strength given by social distanceagents are not fixed at locations, they move within the 2D geometry, to minimize the pressure on their opinionnone
Slanina 2003analytical results for the Sznajd model of opinion formationnetwork with local and global interactionsSznajd modelopinion (discrete, with Q available options)uniformanalytical results for certain simplificationsnone
Sobkowicz 2003bcomparison of results for Nowak-Latane model in various geometries and with inclusion of strong leaderscomparison of 2D, random network, scale-free network, hierarchical networkmodified Kacperski-Holyst modelopinion (binary); strength; leader strength; external biasinteraction strength given by social distancedependence of behaviours on position of leader in scale free networksnone
Sobkowicz 2003aeffects of leader's strategy on opinion formationcomparison of, scale-free network and short range networkmodified Kacperski-Holyst model with introduction of costs of convincing other agentsopinion (binary); strength; leader strength; external bias, costs and available resources, strategyinteraction strength given by social distance and individual strategies in assigning weights to linksstrategies define how each agent uses finite resources to convert others to his opinion, leaders have finite resourcesnone
Sousa 2004bconsensus formation on a triad scale-free networkspecial case of scale free networkmodification of Sznajd modelopinion (binary and discrete spectrum)uniformdiscussion of opinion changes when conviction comes from uniform opinion of 1, 2, 3 neighbours or 3 neighbours forming a trianglenone
Sousa 2004abounded confidence model on a still growing scale-free networkscale free networkDeffuant model, continuousopinion (continuous)uniformnetwork grows at the same time as the opinions are adjusted, but no special effects were discoverednone
Stauffer 2001review of Sznajd model2DSznajd modelopinion (binary)uniform generalization to many possible states is used to explain the distribution of votes among candidates in Brazilian local elections
Stauffer 2002persistence of opinion in the Sznajd consensus modelD-dimensional geometric network; D=1,2,3,4Sznajd modelopinion (binary)uniformstudy of the number of agents that did not change their opinionnone
Stauffer 2003Simulation of Consensus Model of Deffuant et al on a Barabasi-Albert Networkscale free networkDeffuant modelopinion (continuous); uniformbehaviour of scale free network found to be similar to random onenone
Stauffer 2004scaling law for defiant model on scale free networkscale free networkdiscretized Deffuant modelopinion (Q discrete opinions); uniformmulti-layer model representing various age levels; advertising effectsnone
Sznajd 2000introduction of Sznajd model1D chainSznajd modelopinion (binary)uniform none
Tessone 2004neighbourhood models of minority opinion spreading1D, 2DGalam model, global and near neighbourhood interactionsopinion (binary)uniform within local cellseffects of synchronous and asynchronous updates discussednone
Weisbuch 2002comparison of simulation results for continuous and binary opinions and for local and global interactionsglobal; 2D local interactionsmodified Deffuant model;opinion (multidimensional, continuous, bounded)uniform none
Weisbuch 2004 discussion of inuence of possible social networks topologies on the dynamics of Deffuant modelscale free networkDeffuant modelopinion (continuous)uniform none
Weisbuch 2005discussion of the role or extremists in Deffuant modelglobal; 2D local interactions; scale freeDeffuant modelopinion (continuous)uniform none
Wu 2004dynamical theory of opinion formation in social network random graphforced change of opinion on disagreementopinion (binary)uniformsimple analytical treatment and simulations; discussion of effects of some agents with fixed opinionsnone

* References

ALBERT R and Barabási A L (2002) Statistical Mechanics of Complex Networks. Review of Modern Physics, 74:67-97.

ANDRESKI S (1972) Social Sciences as Sorcery. André Deutsch.

BARABSI A L, Ravasza E and Vicsek T (2001) Deterministic Scale-Free Networks. Physica A, 64:559,.

BEHERA L and Schweitzer F (2003) On Spatial Consensus Formation: Is the Sznajd Model Different from a Voter Model? International Journal of Modern Physics C, 14 (10):1331-1354.

BERNARDES A T, Costa U M S, Araujo A D and Stauffer D (2001) Damage Spreading, Coarsening Dynamics and Distribution of Political Votes in Sznajd Model on Square Lattice. International Journal of Modern Physics C, 12 (2):159-168,.

BERNARDES A T, Stauffer D and Kertész J (2002) Election results and the Sznajd model on Barabasi network. The European Physical Journal B-Condensed Matter and Complex Systems, 25(1):123-127.

CARUSO F and Castorina P (2005) Opinion dynamics and decision of vote in bipolar political systems, URL http://arxiv.org/pdf/physics/0503199.

CASTELLANO C, Fortunato S and Loreto V (2007) Statistical physics of social dynamics, URL http://arxiv.org/pdf/0710.3256. Accepted by Reviews of Modern Physics.

CHIALVO D R (2004) Critical brain networks. Physica A Statistical Mechanics and its Applications, 340:756-765.

CHIALVO D R and Bak P (1999) Learning from mistakes. Neuroscience, 90(4):1137-1148.

CONRADT L and Roper T J (2003) Group decision-making in animals. Nature, 421(6919):155-8.

CONRADT L and Roper T J (2005) Consensus decision making in animals. Trends in Ecology & Evolution, 20(8): 449-456.

DEFFUANT G, Amblard F and Weisbuch G (2004) Modelling Group Opinion Shift to Extreme: the Smooth Bounded Confidence Model, URL http://arxiv.org/pdf/cond-mat/0410199.

DEFFUANT G, Amblard F, Weisbuch G and Faure T (2002) How can extremism prevail? A study based on the relative agreement interaction model. Journal of Artificial Societies and Social Simulation, 5(4):1 URL https://www.jasss.org/5/4/1.html.

DEFFUANT G, Neau D, Amblard F and Weisbuch G (2000) Mixing beliefs among interacting agents. Advances in Complex Systems, 3:87-98.

DOROGOVTSEV S N and Mendes J F F (2003) Evolution of Networks From Biological Nets to the Internet and WWW. Oxford University Press, 2003.

DOROGOVTSEV S N and Mendes J F F (2002) Evolution of networks. Advances in Physics, 51:1079-1087.

DYER J R, Ioannou C C, Morrell L J, Croft D P, Couzin I D, Waters D A and Krause J (2008) Consensus decision making in human crowds. Animal Behaviour, 75(2):461-470.

ELGAZZAR A S (2002) Applications of Small-World Networks to some Socio-economic Systems, URL http://arxiv.org/pdf/cond-mat/0212071.

EPSTEIN J M (2008) Why Model? Journal of Artificial Societies and Social Simulation, 11(4):12 https://www.jasss.org/11/4/12.html.

FORTUNATO S and Castellano C (2007) Scaling and Universality in Proportional Elections. Physical Review Letters, 99(13):138701.

FORTUNATO S (2004a) The Sznajd Consensus Model with Continuous Opinions, URL http://arxiv.org/abs/cond-mat/040735.

FORTUNATO S (2004b) Universality of the Threshold for Complete Consensus or the Opinion Dynamics of Deffuant et al., URL http://arxiv.org/pdf/cond-mat/0406054v1.

FORTUNATO S and Stauffer D (2005) Computer Simulations of Opinions URL http://arxiv.org/abs/cond-mat/0501730.

GALAM S (1997) Rational Group Decision Making. A random field Ising model at T=0. Physica A, 238:66-80.

GIL S and Zanette D H (2006) Coevolution of agents and networks: Opinion spreading and community disconnection. Physics Letters A, 356(2):89-94.

HEGSELMANN R and Krause U (2002) Opinion dynamics and bounded confidence models, analysis, and simulation. Journal of Artificial Societies and Social Simulation (JASSS) 5(3)2 https://www.jasss.org/5/3/2.html.

HOLYST J A, Kacperski K and Schweitzer F (2001) Social impact models of opinion dynamics. Annual Review of Comput. Phys., 20:531-535.

JOHN E R (2001) A field theory of consciousness. Conscious Cogn, 10(2):184-213.

KACPERSKI K and Holyst J A (1999) Opinion formation model with strong leader and external impact: a mean field approach. Physica A, 269:511-526.

KACPERSKI K and Holyst J A (2000) Phase transitions as a persistent feature of groups with leaders in models of opinion formation. Physica A, 287:631-643.

LEWENSTEIN M, Nowak A, and Latané B. (1992) Statistical mechanics of social impact. Phys. Rev. A, 45:763-776.

LORENZ J (2007) Continuous Opinion Dynamics Under Bounded Confidence: A Survey. International Journal of Modern Physics C, 18 (12):1819.

MILGRAM S (1967) The Small World Problem. Psychology Today, 2:62-67, 1967.

MOSS S and Edmonds B (2005) Towards Good Social Science. Journal of Artificial Societies and Social Simulation, 8(4):13 https://www.jasss.org/8/4/13.html.

NEWMAN M E J (2000) Models of the small world. J. Stat. Phys., 101:819-841.

NEWMAN M E J and Park J (2003) Why social networks are different from other types of networks. Physical Review E, 68(3):36122.

NOWAK A and Lewenstein M (1996) Modeling Social Change with Cellular Automata. In Rainer Hegselmann, Ulrich Mueller, and Klaus G. Troitzsch, editors, Modelling and Simulation in the Social Sciences From A Philosophy of Science Point of View, pages 249-285. Kluver, Dordrecht.

NOWAK A, Szamrej J and Latané B (1990) From Private Attitude to Public Opinion: A Dynamic Theory of Social Impact. Psychological Review, 97(3):362-376.

PARKINSON C N (1958) Parkinson's Law: The Pursuit of Progress. John Murray.

PLUCHINO A, Latora V, and Rapisarda A (2005) Changing Opinions in a Changing World: a New Perspective in Sociophysics, Int. J. Mod. Phys. C 1694):515--531.

ROEHNER B M (2005) Consensus formation: The case of using cell phones while driving, URL http://arxiv.org/abs/physics/0502046.

SABATELLI L and Richmond P (2003a) Non-monotonic spontaneous magnetization in a Sznajd-like Consensus Model. URL http://xxx.lanl.gov/pdf/cond-mat/0309375.

SABATELLI L and Richmond P (2003b) Phase transitions, memory and frustration in a Sznajd-like model with synchronous updating. URL http://xxx.lanl.gov/pdf/cond-mat/0305015.

SCHELLING T S (1971) Dynamic Models of Segregation. Journal of Mathematical Sociology, 1:143-186.

SCHULZE C (2004) Sznajd opinion dynamics with global and local neighbourhood, URL http://arxiv.org/pdf/cond-mat/0402397.

SCHWEITZER F and Holyst J A (2000) . Modelling collective opinion formation by means of active Brownian particles. European Physical Journal B, 15:723-732.

SIMKIN M V and Roychowdhury V P (2003) Read before you cite. Complex Systems, 14:269-274.

SIMKIN M V and Roychowdhury V P (2005a) Copied Citations Create Renowned papers? Annals of Improbable Research,, January-February 2005, pages 24-27 .

SIMKIN M V and Roychowdhury V P (2005b) Stochastic modeling of citation slips. Scientometrics, 62(3):367-384.

SIMKIN M V and Roychowdhury V P (2006) An introduction to the theory of citing. Significance, 3:179.

SLANINA F and Lavicka H (2003) Analytical results for the Sznajd model of opinion formation. European Physical Journal B - Condensed Matter, 35(2):279-288.

SOBKOWICZ P (2003a) Effect of leader's strategy on opinion formation in networked societies. URL http://arxiv.org/pdf/cond-mat/0311566.

SOBKOWICZ P (2003b) Opinion formation in networked societies with strong leaders, URL http://arxiv.org/pdf/cond-mat/0311521.

SOUSA A O (2004a) Bounded confidence model on a still growing scale-free network, URL http://arxiv.org/pdf/cond-mat/0406766.

SOUSA A O (2004b) Consensus formation on a triad scale-free network, URL http://arxiv.org/pdf/cond-mat/0406390.

SPORNS O, Tononi G and Edelman G M (2000) Connectivity and complexity: the relationship between neuroanatomy and brain dynamics. Neural Networks, 13(8-9):909-922.

STAUFFER D and de Oliveira P M C (2002) Persistence of opinion in the Sznajd consensus model: computer simulation. The European Physical Journal B-Condensed Matter, 30 (4):587-592.

STAUFFER D and Meyer-Ortmanns H (2003) Simulation of Consensus Model of Deffuant et al on a Barabasi-Albert Network, URL http://arxiv.org/pdf/arXiv:cond-mat/0308231.

STAUFFER D, Sousa A and Schulze C (2004) Discretized Opinion Dynamics of The Deffuant Model on Scale-Free Networks. Journal of Artificial Societies and Social Simulation, 7(3)7 https://www.jasss.org/7/3/7.html.

STAUFFER D (2001) Monte Carlo simulations of Sznajd models. Journal of Artificial Societies and Social Simulation, 5(1)4 https://www.jasss.org/5/1/4.html.

STROGATZ S H (2001) Exploring complex networks. Nature, 410:268-276.

SZNAJD-WERON K and Sznajd J (2000) Opinion Evolution in Closed Community. Int. J. Mod. Phys. C, 11:1157-1166.

TESSONE C J, Toral R, Amengual P, Wio H S and San Miguel M (2004) Neighborhood models of minority opinion spreading. The European Physical Journal B-Condensed Matter, 39 (4):535-544.

WATTS D J (1999). Small Worlds. Princeton University Press, Princeton.

WATTS D J and Strogatz S H (1998) Collective dynamics of 'small-world' networks. Nature, 393:440-442.

WEISBUCH G (2004) Bounded confidence and social networks. The European Physical Journal B-Condensed Matter and Complex Systems, 38(2):339-343.

WEISBUCH G, Deffuant G, and Amblard F (2005) Persuasion dynamics. Physica A: Statistical Mechanics and its Applications, 353:555-575.

WEISBUCH G, Deffuant G, Amblard F and Nadal J P (2002) Meet, Discuss, and Segregate! Complexity, 7(3):55-63.

WILLIAMS L E and Bargh J A (2008) Experiencing Physical Warmth Promotes Interpersonal Warmth. Science, 322(5901):606.

WU F and Huberman B A (2004) Social Structure and Opinion Formation, URL http://arxiv.org/pdf/cond-mat/0407252.

ZANETTE D H and Gil S (2006) Opinion spreading and agent segregation on evolving networks. Physica D: Nonlinear Phenomena, 224(1-2): 156-165, 2006.


ButtonReturn to Contents of this issue

© Copyright Journal of Artificial Societies and Social Simulation, [2009]