3
Voting Choice Computer Simulation Model

The Computer Simulation Model of Voting Choice (McPhee, 1961, 1963) was inspired by the research underlying the Columbia electoral studies of the 1940 US presidential campaign published in the famous book The People’s Choice (Lazarsfeld et al., 1944). On the basis of the empirical findings reported and analysed in the volume authored together with Bernard Berelson and Paul Lazarsfeld (Berelson et al., 1954), William McPhee developed a computer simulation modelling approach and elaborated its theoretical and methodological aspects. Starting in 1958, more experimental research work with this model was developed together with his research assistant Robert B. Smith, who designed and performed the simulation runs on an IBM 650 computer, and Jack Ferguson, who joined their team a year later. ¹ The model was tested in the 1960 primary elections in Wisconsin (USA), and the theoretical and experimental results and conclusions were published starting in 1961, though much of this research remained unpublished or unknown. When retrieved much later, it was re‐evaluated for its precious contribution to the simulation modelling of political attitudes based on their content and on the culture of community political participation (Huckfeldt et al., 2004: 13).

The electoral studies developed by the Columbia group are mainly concerned with individual citizens’ electoral behaviour in presidential campaigns. As theory, they are characterized by a fundamental tenet: the social context is essential for the formation and change of the individual’s voting preferences and choices.

McPhee’s Voting Choice Computer Simulation Model includes theoretical and experimental elements concerning the formation and change of political attitudes toward candidates and parties during and across electoral campaigns. The model proves the main hypotheses of the Columbia electoral studies: the electoral media and campaigns have only ‘minimal effects’ on individuals’ voting behaviour. Instead, individual voting choices are shaped by the social context represented by the interpersonal networks of relations (dyadic networks) within which individuals interact in face‐to‐face discussions about political issues. Though dependent on the social and individual attributes, the voting choice is modelled as a behavioural response to social interaction stimuli coming mainly from the interpersonal networks of relations.

However, McPhee goes beyond the classic limits of the behaviourist paradigm. He works on a learning assumption which endows each individual with the ability to learn from the political influence exercised by peers in interactive discussions within the dyadic networks. The model thus achieves specific dynamics: individual voting dispositions, preferences and choices change over repeated persuasive political campaign stimulation such that perceived electoral stimuli (political messages) are reinforced (or, on the contrary, forgotten) by means of a learning mechanism. Thus, in the aggregate, overall voting choice patterns emerge from the dynamics of the individual political disposition change. Variability of the individual political predisposition, in turn, depends on convergent effects of perceived campaign stimuli and the political influence experienced from individual interactions.

Conceptual Model

McPhee’s theoretical and experimental research represented pioneering work in the computer modelling and simulation of electoral behaviour. At the conceptual level, the Voting Choice Computer Simulation Model combines principles from psychology, sociology and social psychology with voting behaviour and mass communication theories. The model combines the psychology of attitude formation under social and political pressure with an instrumental utility‐based modelling approach of the voting choice as dynamically balancing the costs and benefits of individual interactions within the ambit of interpersonal networks.

Columbia electoral studies offer empirical support to the behaviourist ‘stimulus–response’ paradigm in modelling citizens’ voting behaviour. In this paradigm, formation and change of individual political attitude toward the electoral issue as well as individual voting choice are viewed as being determined by a combination of external social stimuli and internal predispositions. Regarding this simulation model, the first in the short history of political attitude computational modelling research, there are two essential aspects of major interest for the purpose of our study: the modelling principles and the dynamics of the political attitudes as generated by the simulation model.

The computational modelling of political influence and the simulation of political attitude change are based on the fundamental idea that political preferences are shaped by the social characteristics of the voters and by their social interaction experience.

The main assumption of McPhee’s model is that the political attitudes of the citizens (i.e. their issue positions with respect to the electoral candidates and parties in the presidential elections) are not determined by individual attributes and predispositions only. Their political attitudes are influenced by the social environment and the individual interactions with peers, which allow for social pressures to propagate through networks of interpersonal relations. Citizens get the news from the media, but have discussions about it with their acquaintances, close ties or influential others in their networks of personal contacts. These dyadic interactions are considered to be essential for the formation and change of political attitudes within the interpersonal networks, especially during electoral campaigns, when political interest is highly stimulated (Berelson et al., 1954; Huckfeldt et al., 2004: 39).

Social Process Model

The original model is strongly influenced by Paul Lazarsfeld, who in the early 1930s developed research on consumer behaviour, in particular on the psychological aspects of deliberation and action concerning purchase behaviour (Lazarsfeld, 1934). Lazarsfeld defined consumer behavioural tendencies. To formally describe them, he used variables representing the consumer’s internal predispositions and socio‐economic status, the characteristics of the products to be purchased and the influences from the external social environment and interpersonal networks of contacts. He further described consumer behaviour with a three‐process scheme including stimulation, consultation (i.e. influence) and deliberation. Reconsidered later in the Columbia group, headed by Lazarsfeld, this scheme has been inspiring for the modelling of the voting behaviour of the individual citizen as subject to social context influence.

The computer simulation voting model is based on three causally chained processes: stimulation, influence and learning. The processes are linked in a loop which feeds back the political influence onto the predispositions and the perception of new campaign stimuli. The system thus achieves a dynamic behaviour. In this system, the learning process provides for a self‐regulatory function, making possible the adjustment of individual agents’ predispositions and perceptions of the social stimuli as a consequence of social network influence.

The model describes presidential elections. In the elections, a community of n citizens is involved. The real community of citizens is described as a set of n individual (artificial) actors with three relevant characteristics: political interest (utility), political preferences and political predispositions. The actors interact among them within networks of interpersonal ties. The actors are the target of the campaign messages, which are defined and described as political stimuli. The outcome of a simulation run is a ‘map’ of distributed voting choices, which provides for a dynamic picture of political choices (votes).

The computer simulation run of the model has three steps.

The first simulation step consists of the political stimulation process. An individual agent is characterized by previously determined predispositions with attributes of direction and intensity. These predispositions are expressed as probabilities or levels of preparation for social action (i.e. voting) and levels of political interest.

The second simulation step consists of the political influence process: the individual actor i encounters another actor j from a set of previously determined actors (sociometric method). The actor i compares its political interest value with the value of the contact actor; if their utility value is the same, they both get rewarded by an increase in their intention to vote. If their intention to vote is thus reinforced, they will make the final vote decision. Otherwise, their intention level will decrease, which will force the actors to wait for another stimulation stage in order to become more interested in voting.

The third step consists of the learning process. The learning mechanism updates the political predispositions and feeds them back to the inputs of the stimulation process. Thus, the updated predispositions are subjected again to the influence of the campaign media. The aggregated variation results in a dynamic process in which variation is generated by the interactions within the network of personal connections and reinforced by the persuasive campaign messages.

McPhee adopts this social process model in his voting choice simulation model and makes it operationally effective by including several computational simulation techniques. ²

Dyadic Networks

With regard to the political persuasion issue, the Columbia School introduced a new conceptual paradigm on the formation and change of political attitudes: the network of interpersonal relations, which is a concept that defines the individual’s connections with close ties, acquaintances and influential persons (i.e. opinion leaders) with whom they discuss the political issues of the day and whom they trust for their authority and expertise.

Columbia empirical studies showed that exposure to electoral campaign messages and communication does not have a major impact on changing the individuals’ voting preferences unless mediated by personal relations (opinion leaders, influencers and close ties) in face‐to‐face personal discussions. Even if change‐effective, the mass media messages (and the overall campaign media news coverage) result in the reinforcement of existing political attitudes, since individuals selectively seek and accept information, discussants and media news which would rather confirm and/or sustain their own preferences, trying to avoid cognitive dissonance and sometimes the discomfort of conflicting situations based on disagreement (Festinger, 1950, 1957).

Two‐Step Flow of Communication

In their book on interpersonal influence, Elihu Katz and Paul Lazarsfeld (1955) address the issue of political attitude formation and change in the context of a presidential election campaign. During elections, public and private mass media channels compete for influencing citizens’ opinions, attitudes and behaviours via campaign messages. The authors show that the networks of interpersonal relations actually operate as intermediate filters for the mass media messages: individual citizens’ predispositions are influenced by their networks of close ties in evaluating and understanding the campaign messages, and in shaping their final voting choices.

The concept of dyadic networks is employed in the theory of the two‐step flow of communication (Katz and Lazarsfeld, 1955) about opinion leaders and their impact on individual political preference formation and change. It is also employed in the political persuasion theoretical approaches concerning both communication and political participation and partisanship issues (Berelson et al., 1954).

Homophily Principle: ‘Likes Attract’

The experiments based on the Voting Choice Computer Simulation Model use sociometric measurements. The empirical studies in the Erie County study showed that the selection process is dependent on mainly two factors: social status (socio‐economic level, religion and residence) and similarity.

Studying the similarity factor, Paul Lazarsfeld and Robert Merton introduced the concept of homophily (Merton, 1948; Lazarsfeld and Merton, 1954) for describing a principle of connecting people in local social networks: ‘likes attract’. The explanation for the similarity factor is the general tendency of individuals to avoid conflicting situations, which makes them choose persons on the basis of shared beliefs, similarity of preferences and attitude agreement. These concepts and theories provided the basis for Lazarsfeld and his associates to introduce the concept of the network of relationships for replacing the traditional sociological concept of ‘group’. The concept allowed the study of interdependent patterns of communication and influence in interpersonal networks (Huckfeldt et al., 2004: 26).

Individuals have a tendency to select like‐minded people to discuss political issues on which they share attitudes or preferences. This tendency is self‐reinforcing, as an individual would always select like‐minded people to share issue positions. One advantage for the dyadic networks is that agreement is reached by consensus, easily achieved when individuals select each other for interaction on the basis of the homophily principle. One disadvantage is that such networks would invariably end up in a universal agreement if the increase in agreement were to depend linearly on an increase in the number of peers sharing the same political attitude toward an electoral issue. In this case, the study of diversity survival appears as a challenging research problem. McPhee was the first to approach it in a simulation model of political attitude change (McPhee, 1963: 81). His approach represents, in many ways, a fundamental reference for the one developed much later by Huckfeldt and his collaborators (2004).

Operational Model

From a simulation experimental perspective, McPhee’s Voting Choice Computer Simulation Model is basically working within the classic theories of social dynamics. What makes the model so interesting is the design of the political influence mechanisms and processes. With a strong algorithmic flavour owing much to the programming framework of classic computer science of the early 1940s, the simulation model achieves a particularly interesting process dynamics from a strictly deterministic programming paradigm (see Figure 3.1).

Figure 3.1 The individual agent, its internal representations and the flow of electoral stimulation.

Much of the operational view underlying the voting choice simulation model is based on the classic theories of dynamics. Such theories use variables and processes to account for a causal chain explanation of change in the model system: change of each variable is causally connected to changes in other variable(s). Processes of change iteratively update the variables, providing for the variability of the system as a whole. System overall dynamics emerge as a global pattern as the system variables are updated by these processes.

The model is based on the idea that the social environment during electoral campaigns can be described as a stochastic environment. Along with a rational view of the individual citizen, and with the stochastic characteristics of the social environmental stimuli, the model employs the new ideas following from the empirical (survey and panel) studies underlying the Columbia theories of political behaviour. Two ideas will become influential for all the subsequent research in simulation modelling of social and political phenomena. One concerns the introduction of influence processes which involve, for each individual actor, repeated discussions with the relations within interpersonal networks until consensus is reached and a voting choice is made. The second idea is based on the homophily concept (‘likes attract’), introduced by Lazarsfeld and Merton (1954), that people selectively accept both media messages and others’ opinions if they reinforce and/or confirm the individual’s own beliefs and issue positions (attitudes).

At the operational level, McPhee’s Voting Choice Simulation Model introduces several computational and simulation modelling techniques.

The model consists of a computational system representing a community with n individual citizens involved in a presidential election campaign. System dynamics model the development in time of the aggregate political preferences of n individual actors preparing to make their final voting choices. It is a process‐based model which addresses issues of social dynamics theory and uses classic computational modelling concepts. Given a set of n individual actors, the model simulates the aggregation of citizens’ voting preferences as their predispositions change during repeated persuasive campaign stimulations and political influences exercised by peers in the dyadic networks.

First, McPhee adopts a technique of computational simulation based on probability distributions of perceived stimuli (McPhee, 1963: 170–179; McPhee, Smith and Ferguson, 1963: 77–78). This technique involves several concepts used for the first time in political attitude simulation modelling. One concerns the scaling of the perceived stimuli, a concept inspired from Lazarsfeld’s (1950a,b, 1959) two‐class model and extended so as to include several classes like in the Guttman scale measurement method: the technique of scaling the individual perceptions of social stimuli, rating them into several classes from ‘weak’ to ‘strong’ and using this in the stimuli sampling. The other one is the Monte Carlo method of random sampling, a method which has been used in numerical simulations for achieving probability distributions of a certain item, in this case the probability distributions of the political attitudes during and across electoral campaigns.

Second, McPhee accommodates in his model the idea of social influence forces and the concept of social reality as defined by Lewin (1947a, b). McPhee’s idea that the social empirical evidence has to remain a reference level in testing the validity of simulation experiments is rooted in the Lewinian concept of social reality. As invoked by McPhee, and as one of the long‐debated and systematically questioned elements of simulation experiments’ validation theories, it still represents a key dimension of the validity requirements, which actually are the most sensitive and challenging aspects in social and political agent‐based simulation modelling research.

Third, he introduces the idea that repeated individual interactions within dyadic networks allow individuals to exercise political influence on each other by means of face‐to‐face discussions. This political persuasion can be remembered. As persuasion effects accumulate, learning allows for remembering them, or at least part of them. Modified predispositions are then subjected again to perceived stimuli. The individual actor can thus achieve different impressions from different stimulations and from modified predispositions. This process works on the basis of a cumulative learning mechanism which provides for a dynamic change process and allows the study of aggregated dynamics.

Electoral Campaign Appeals

A stimulus situation is modelled as a set of inputs denoted as ‘appeals’. The ‘appeals’ are relevant stimuli defined as political messages sampled by the individual actors themselves on a rational choice basis (McPhee and Ferguson, 1962; McPhee and Smith, 1962; McPhee, 1963). The stimulus situation is modelled as a probability distribution and becomes a vector of political influence. It takes as inputs the media news and coverage as well as messages in the interpersonal networks (McPhee, 1963: 173).

The model formalizes the empirical findings in the election studies and explains the macro‐level aggregate patterns of voting preferences. The model uses these macro‐level patterns to predict political attitude change phenomena across elections (Smith, 1985: 63).

What people usually do is selectively ‘sample’ the news themselves: individual preferences, and tendencies make each person understand the news from different perspectives, with different backgrounds and with different cognitive abilities. This difference is due to both internal and external factors, but what interests here is the external factor, particularly the role played by the social network as a ‘filter’ for the individual perception. Real stimuli, therefore, become perceived stimuli; that is, representations of sampled biased information from a stochastic communication environment. For a large number of individuals used to computationally represent and simulate a community of citizens in an election scenario, modelling the social stimuli and the stimulus situation is not an easy task. The complexity of social stimuli is modelled so as to retain the operational aspects manageable in a computer simulation.

Probability Distributions of Appeals

The model defines and represents the electoral stimuli (McPhee, 1963: 170) as a distribution of probabilities. The concept is inspired from a survey technique of converting a proportion (i.e. the proportion of people perceiving various degrees of strength of electoral appeals) into a probability measure. A stimulus distribution thus appears as the formal representation of the distribution of people’s ‘impressions’ acquired from sampling the real stimuli provided by the electoral campaign environment. Aggregated, these patterns of perceptions of sampled stimuli distributions describe the variability of people’s political (party) preferences. If repeatedly exposed to electoral campaign stimuli, this variability specifies a functional dependence between perceived strength degrees of social stimuli and changes in the individuals’ predispositions. This function can be instrumentally employed in controlling political preferences by varying the stimulus in order to change the behavioural response (McPhee, 1963: 173). The conditional response is then analysed to evaluate the influence of the social networks in the formation and change of individual political behaviours and attitudes. The model, therefore, provides a conceptual and practical view on how political influence actually works in electoral campaigns and what are the practical means to simulate it with a computer program.

These initial distributions of probabilities expressing stimuli and attitudes towards election (party or candidate) are meant to express the dependence of the individual’s voting intention formation and voting decision‐making on the social context. The initial distributions express a certain relationship between social structures (like social status, social role), predispositions, which are subjected to influences from personal interconnections, mass media messages and so on, and social action (voting).

Variables

The model uses variables to represent the individual citizen (voter) and its relevant attributes: predispositions (i.e. political attitudes toward candidates and parties), political interest (utility) and political preference. Variables representing the individual’s predispositions can have three values: ‘Democratic’, ‘Republican’, ‘No Voting’. There are two directions (Democrat and Republican), and seven degrees of intensity defined in ‘The index of political predisposition in the 1948 election’ (Janowitz and Miller, 1952: 715).

Individual Actor

The individual citizen is modelled with the help of a set of variables which extract the most relevant attributes of a real citizen (see Figure 3.2): predispositions (political attitudes), political interest and political action (i.e. voting choice). Besides these attributes, the model includes data concerning the individuals’ socio‐economic status, religion and residence attributes, collected from empirical studies. ³

Figure 3.2 Individual actor.

Individual actors modify their predispositions under the influence of social networks: they interact with each other within interpersonal networks of relations. For each individual actor, the collection of personal relations is identified in preliminary empirical studies by sociometric measurements.

The view of an individual citizen whose political attitudes (predispositions), political interest and political participation are modified by social networks (group membership) is characteristic for the Columbia School. It is the view of small worlds in which individual actors depend on each other in networks of close relations (family, friends, co‐workers). The influential others (opinion leaders) in their networks help them adjust their understanding of the media news and form or change their political attitudes.

Figure 3.3 Variables, mechanisms and processes involved in the modelling of political influence. Variables are used to represent (a) individual attributes and (b) stimuli from the social environment. Mechanisms controlling the political influence processes: (a) a threshold mechanism controlling the value of the political interest (i.e. subjective utility); and (b) a cumulative mechanism for predispositions update during the learning process. Processes: (a) grey arrows indicate stimulation processes; (b) black arrows indicate influence processes.

Political Attitude

In the early literature on attitude’s definition and measurement, it is associated with a behavioural predisposition to the stimulus (object or situation) which elicits attitude activation (Bogardus, 1931: 62; Allport, 1935: 810; Campbell, 1963).

From an operational perspective, in the voting choice simulation model the predispositions are defined as issue positions of the citizen with respect to the election campaign issues. Such issue positions (i.e., political attitudes) have an internal structure which is socially determined (Smith, 1985: 61). ⁴

Predispositions are characterized by two attributes: direction and intensity. The attribute of direction takes two values (for example, ‘left’ and ‘right’), while the attribute of intensity could take several values (usually, a real number between 0 and 1) describing the level of activation/interest in a stimulus situation. For the 1948 presidential elections modelling, two directions were defined (‘Democrat’ and ‘Republican’) and seven degrees of intensity in ‘The index of political predisposition in the 1948 election’ (Janowitz and Miller, 1952: 715).

Political action (voting choice) is determined by both internal and external forces. On the one hand, the social structural attributes (status, religion and residence) play a relevant role as well as the individual’s predispositions. On the other hand, the perceived social stimuli are as relevant as the internal factors. The political stimuli are represented by the campaign messages and interpersonal influences.

Mechanisms

Several mechanisms are involved in making effective the political influence from either stimuli arriving from the campaign media or from the network of interpersonal relations.

The model uses a social psychological mechanism of political persuasion which combines external sources of influence (social structure, political messages, interpersonal relations) and internal predispositions in modifying the individual preference (Smith, 1985: 62).

Stimulus and predisposition make a threshold mechanism which controls the level of political interest (i.e. individual utility). It determines the participation and also the direction of participation: left or right, Democrat or Republican Party.

Each citizen is supposed to interact with others within networks of personal relations, acquaintances and close ties. Though interaction is necessary for political influence to operate effectively, it is allowed only under strictly defined circumstances: any two individual agents start discussing only if their average utility value is over a specific threshold. Since utility values are compared against a predetermined threshold, only higher values of political interest will result in political predisposition (attitude) change in favour of voting. This process is self‐reinforcing, so that the strength of a previously settled predisposition might increase.

If the comparison of two individual actors shows that their political interests are similar, then political attitude is reinforced by an increase in the level of political interest of each actor, which can trigger an increase in their partisanship and, consequently, in their political participation.

As emphasized in his earlier work with Berelson and Lazarsfeld (Berelson et al. 1954: 25–27), McPhee’s idea behind the mechanisms of controlling the political interest variable is that, for any two individual citizens, political participation (voting) is enhanced by the social structure (high social status and relevant social role) and by the level of political interest. Each discussion in the dyadic network results in rewarding both actors if they have similar voting intentions (political preferences for the same party), or punishing them if their preferences differ: similarity of political interest (i.e. the same utilities) increases their participation, and each individual gets rewarded by an increase of its utility value, whereas dissimilarity in political interest results in weakening partisanship of each, such that both agents are punished by a decrease in their utility value. Reward and punishment help in implementing the homophily principle of selective interactions in the interpersonal networks; they are instrumentally employed in achieving either homogeneity (convergence to agreement by an actor’s compliance in his interpersonal network of relations) or polarization phenomena. The idea of using reward and punishment in achieving convergence is a strong argument in favour of homophily as the principle for describing political influence propagation in social networks. McPhee’s Voting Choice Simulation Model is the only one to use a control framework of the convergence to homogeneity phenomena. All the other subsequent approaches to the universal agreement phenomena are concerned with the problem of diminishing the convergence tendencies in social simulation and achieving methods which conserve disagreement and diversity without affecting the democratic type of the model society (Johnson, 1999; Huckfeldt et al., 2004).

Finally, the learning mechanism actually completes the operation of the political influence process. It allows for the final levels of political interest to be used for updating the predisposition of each individual agent: the modified predispositions are backpropagated and subjected once more to campaign stimulation. If exposed again to political messages, the updated predisposition, inducing the individual’s voting behaviour, becomes associated with particular stimuli. The idea, inspired by early behaviourist theories of stimulus–response learning (Estes, 1950) and one‐trial learning (Guthrie, 1946), explains how stimulus constitutive elements and predisposition structural attribute values (e.g. direction) are associated with specific behavioural responses.

Though considered a classical example of ‘stimulus–response’ behavioural learning and often criticized for too much social determinism (Visser, 1996), McPhee’s model goes beyond these paradigmatic limits. He describes the political persuasion process within the dyadic network as a Polya process (Polya, [1957]1973) and suggests that particular values of the predisposition’s attributes (i.e. direction) appear repeatedly as the individual agent shows preference for discussing the issue with like‐minded people (McPhee et al., 1963: 79). These appearances, shared by like‐minded persons in the dyadic network, will actually reinforce a particular voting preference. The process describing this tendency is a self‐developing process. As a particular voting preference appears more often, this facilitates the tendency toward achieving agreement, making the individuals discuss and increase the level of their political participation. In the aggregate, the effect of these self‐developing, self‐reinforcing processes becomes evident in the emergence of clusters of voting preferences. The macro‐level phenomena are emphasized in the Wisconsin Primaries case study. ⁵

The 1960 Wisconsin Primary election computer simulation experiment was meant to show how mass media messages and the networks of interpersonal relations influence the individuals’ initial predispositions and shape their final voting choices. The case study of the presidential elections in Wisconsin, USA, 1960, with two presidential candidates, Nixon and Humphrey, shows how the simulated changes in the campaign ‘appeal’ result in changes in the vote distributions. The case study proves how the voting preferences depend on the social context and on the campaign stimulation. The variations in the voting preferences of two types of target population, namely farmers and town middle‐class in northwestern Wisconsin, are studied as being dependent on the dynamics of their predispositions. The predispositions change process plays an important role in the dynamics of the relationship between the dispersion of the ‘appeals’ (stimuli) and the form of response (vote choice): depending on the subjective ‘interpretation’ of the appeal, the individual’s behaviour depends on the social‐status characteristics (Protestants/Catholics, rural/urban), on the level of individual’s disposition and on the appealing power of the stimuli (the so‐called ‘farm’ appeal). The dynamics of change in the individual predispositions provide for the emergence of voting preference polarization phenomena.

Model’s Relevance

McPhee’s Voting Choice Simulation Model has been forgotten for a long time, most probably because of its explicit background in behaviourist theory. The model is based on the stimulus–response paradigm, and this made its critics see it as working on social determinism principles which could be accepted as a rough explanation of the associations between stimuli and voting preferences during the electoral campaigns, but which could not explain more than that. For example, it could not explain the voting choices and the polarization of voting preferences unless a free‐choice dimension was introduced in the process of individual predisposition change. McPhee was aware of these critics and addressed the subject in one of his papers (McPhee, 1961), trying to explain how, how much and to what purpose was his approach going beyond the principles and constraints imposed by the classic stimulus–response model of behaviour. His modelling approach includes a dynamics of voting preferences aggregation which could not have been achieved by the stimulus–response paradigm only. Moreover, his model exceeds the limits of the social process model by introducing some mechanisms – learning, in particular – which made the processes acquire a certain dynamics. In explaining this design choice and its meaning, McPhee included at least two aspects which are essential for our understanding of the particular dimensions on which computational and simulation modelling of political attitudes have developed ever since. McPhee’s model puts a special emphasis on the dynamics of individual interactions within the networks of personal ties. It is these dynamics which, as observed empirically in his experiments, had inspired him to develop research on two dimensions. One belongs to political methodology and addresses an issue which somehow was in vogue at the time: ‘plausible scenarios’. During the 1950s and at the beginning of the 1960s, the idea of social forecasting, originally developed in the area of military studies by Herman Kahn (1962), was gaining support in social studies. McPhee was the first to introduce a new paradigmatic approach in political attitude simulation modelling: he speaks about his simulation experiments as ‘what could be’ and ‘what might have been’ types of experiments, thus inducing the idea that a computer simulation works on possible scenarios and develops each of them in the ultimate detail, so that aggregate phenomena could be virtually observed and virtually experimented. In the political science of the 1950s, this means to break up the tabu of empirical research. Such a speculative exercise demands a reference system in which the truth requirement could be anchored, so that he calls on Lewin’s concept of social reality (Lewin, 1947a, b). This view of plausible scenarios which can be virtually generated by computer simulation runs and studied as an alternative reality addresses a notion which will be established and recognized much later as the ‘would‐be worlds’ concept introduced by John Casti (1997) in his theory on computer simulation modelling. Making reference to a ‘thought experiment’, a concept introduced earlier by Daniel Dennet (1994), Robert Axelrod (1997) suggested a culture dissemination model which is based on a thought exercise, understood as a ‘what if…’ scenario generated with agent‐based modelling techniques.

McPhee used this concept to address the fundamental characteristic of a computer simulation model to provide a view of the potential developments of a process or set of processes which depend on individual interactions. It is precisely this aspect of the potential developments which McPhee addressed in describing the aggregate patterns and dynamics of political attitude change in his simulation model. McPhee foreshadowed a concept established much later in computational and simulation modelling, which is specific to the generative approaches of political and social phenomena (Axelrod, 1997; Cederman, 2005). The ability to experimentally investigate so early (perhaps too early for the limited capacity of the computers of that time) the concept of ‘thought exercise’ makes Bill McPhee and his co‐workers, Bob Smith and Jack Ferguson, the forerunners of a simulation modelling concept which will later on be employed by Robert Axelrod (1997) in his culture dissemination model.

The second essential aspect in McPhee’s justification addresses one of the fundamental goals of social science theory: explaining the link between the micro and macro levels of a social system in computational and simulation modelling terms. McPhee was the first to develop a political attitude computer simulation model which describes virtual experiments able to provide believable clues on how the micro‐level individual interactions might evolve in the aggregate.

By the end of the 1950s, the experience McPhee had already gained from the electoral studies and computer simulation modelling, turned into a new research project. The issue proved to be so highly relevant as to foster McPhee’s engagement in a new fundamental research programme on computer simulation modelling of mass dynamics defined and explained together with James Coleman (McPhee and Coleman, 1958). The programme aimed to study the electoral phenomena in large populations of individual voters (McPhee and Coleman, 1958: 6). The issue of interest was the aggregate phenomena as the outcome of the individual interactions developing at the micro level of a social system (McPhee and Coleman, 1958: 7). The approach addressed the problem of studying the aggregate electoral dynamics with the help of computers and specific computational instruments able to translate the discrete observations describing the individuals and their interactions into artificial complex social systems describing the emergence of macro‐level phenomena (McPhee and Coleman, 1958: 9). Though never fully accomplished by its authors, this programme nevertheless laid the theoretical foundations of a new approach on social action in the study of social systems: the emergence of macro‐level social and political phenomena from the individual interactions at the micro level.

As a matter of fact, this subject has become a research issue as the ‘small world’ research has slowly grown since the early 1950s. It emphasized new directions of research in social systems involving computational support for modelling and simulations. Two famous research groups have systematically pursued this orientation of research: while William McPhee was developing his voting choice computer simulation model, another important research group was developing a separate project on a computer simulation model of the presidential elections of 1960 and 1964. Though developed separately (McPhee, 1963: 169), the approach of Ithiel de Sola Pool, Robert Abelson and Samuel Popkin (1965) addressed mainly the same issue: the persuasive influence of the campaign communication and the role played by the networks of interpersonal relations on individuals’ voting predisposition, choice and behaviour. Other projects, like the Simulmatics project (de Sola Pool and Abelson, 1961) and Community Referendum (Abelson and Bernstein, 1963), emphasized and deepened the differences from McPhee’s type of approach and identified new dimensions of research on the political attitude phenomena.

Abelson and Bernstein (1963) developed a computer simulation model on similar issues, namely the attitude change of individuals in a small community under the influence of (a) campaign media communication and (b) individual interactions. Though concerned with the same research questions regarding the individual interactions and their effects in the aggregate, their model focused more on the interactions’ characteristics, like the rules which should govern the communication with the citizens, and the behavioural response of each citizen to the influence exercised by means of campaign messages. The study of these rules finally resulted in a heavy simulation model which has never been tested against empirical data due to its overwhelming, sophisticated structure.

McPhee and collaborators modelled the aggregate phenomena of the changes induced in the individuals’ political preferences and choices by the campaign media and the networks of close ties. This modelling approach is based more on the contents of the political predispositions, and not on formal aspects of the individuals’ social actions, like the number of individuals or their interactions’ characteristics.

This approach has inspired further development of two areas of research: one is concerned with the methods and techniques of modelling and simulating the social systems as complex collections of individual agents, which has increased the interest in agent‐based modelling of social phenomena. The other one is mainly concerned with the micro‐level individual interactions and the macro‐level phenomena they might generate.

Much later, Axelrod (1997) developed a simulation model of culture dissemination which builds upon the idea that attitudes (as culture elements) change due to individual interactions, which are viewed as a generative engine: the emergence of macro‐level phenomena is explained by the micro‐level individual interactions (methodological individualism).

Both Abelson and Bernstein’s model and Axelrod’s model are extensively investigating the idea of generative individual interactions at the social micro level. Their models are rather interaction models and accomplish, in many ways, the claims of the early programme of McPhee and Coleman (1958) on the study of mass behaviour and social action. Their model is briefly introduced in the subsequent chapter in order to describe its main characteristics and performances.

Philosophy of Computational Modelling and Simulation

McPhee’s voting choice model was the first computer simulation model of voting behaviour in the history of computational approaches in political science.

It could be that this model arrived too soon in a research community which was not yet prepared for the intricate claims of a ‘simulation program’. Computers were still too big and too slow, whereas the political science research of the 1940s had been almost ‘swept’ by research on military issues and war propaganda. Columbia’s well‐known studies were meant to bring about the behavioural revolution in political science. In the underground of this tardy revolution, however, the computer simulation era started almost unnoticed, as a silent companion. William McPhee was remarkably ingenious and subtle. And so is his Voting Choice Simulation Model.

References

1. Abelson, R.P. and Bernstein, A. (1963) A computer simulation model of community referendum controversies. Public Opinion Quarterly , 27(1), 93–122.
2. Allport, G.W. (1935) Attitudes. In C. Murchison (ed.), Handbook of Social Psychology . Clark University Press, Worcester, MA; pp. 798–844.
3. Axelrod, R. (1997) The dissemination of culture: a model with local convergence and global polarization. Journal of Conflict Resolution , 41(2), 203–226.
4. Berelson, B., Lazarsfeld, P. and McPhee, W.N. (1954) Voting: A Study of Opinion Formation in a Presidential Campaign . Chicago University Press, Chicago, IL.
5. Bogardus, E.S. (1931) Fundamentals of Social Psychology , 2nd edn. Century, New York.
6. Campbell, D.T. (1963) Social attitudes and other acquired behavioral dispositions. In S. Koch (ed.), Psychology: A Study of a Science , vol. 6. McGraw‐Hill, New York; pp. 94–172.
7. Casti, J. (1997) Would‐Be Worlds: How Simulation is Changing the Face of Science . John Wiley & Sons, Inc., New York.
8. Cederman, L.‐E. (2005) Computational models of social forms: advancing generative process theory. American Journal of Sociology , 110(4), 864–893.
9. De Sola Pool, I. and Abelson, R.P. (1961) The Simulmatics project. Public Opinion Quarterly , 25, 167–183.
10. De Sola Pool, I., Abelson, R.P. and Popkin, S. (1965) Candidates, Issues and Strategies . MIT Press, Cambridge, MA.
11. Dennet, D. (1994) Artificial life as philosophy. Artificial Life , 1(3), 291–292.
12. Estes, W.K. (1950) Toward a statistical theory of learning. Psychological Review , 57, 94–101.
13. Festinger, L. (1950) Informal social communication. Psychological Review , 57, 271–282.
14. Festinger, L. (1957) A Theory of Cognitive Dissonance, Stanford University Press.
15. Festinger, L., Schachter, S. and Back, K. (1950) Social Pressures in Informal Groups . Stanford University Press.
16. Guthrie, E.R. (1946) Psychological facts and psychological theory. Psychological Bulletin , 43, 1–20.
17. Huckfeldt, R., Johnson, P.E. and Sprague, J. (2004) Political Disagreement. The Survival of Diverse Opinions within Communication Networks , Cambridge Studies in Political Opinion and Political Psychology. Cambridge University Press.
18. Janowitz, M. and Miller, W.E. (1952) The index of political predispositions in the 1948 election. Journal of Politics , 14(7), 710–727.
19. Johnson, P.E. (1989) Formal theories of politics: the scope of mathematical modeling in political science. Mathematical and Computer Modeling , 12(4–5), 397–404.
20. Johnson, P.E. (1999) Simulation modeling in political science. American Behavioral Scientist , 42(1), 1509–1530.
21. Kahn, H. (1962) Thinking about the Unthinkable . Horizon Press.
22. Katz, E. and Lazarsfeld, P.F. (1955) Personal Influence: The Part Played by People in the Flow of Mass Communications . Free Press, New York.
23. Lazarsfeld, P.F. (1934) The Psychological Aspect of Market Research, Harvard Business Review , 13(1), 54–71.
24. Lazarsfeld, P.F. and Merton, R.K. (1954) Friendship as a social process: a substantive and methodological analysis. In M. Berger, T. Abel and C.H. Page (eds), Freedom and Control in Modern Society . Van Nostrand, New York; pp. 18–66.
25. Lazarsfeld, P.F, Berelson, B. and Gaudet, H. (1944) The People’s Choice: How the Voter Makes Up His Mind in a Presidential Campaign . Columbia University Press, New York.
26. Lazarsfeld, P.F. (1950a) The Logical and Mathematical Foundations of Latent Structure Analysis, (Chapter 10) in: Samuel A. Stouffer, Louis Guttman, Edward A. Suchman, Paul F. Lazarsfeld, Shirley A. Star, John A. Clausen (Eds.), The American Soldier. Studies in Social Psychology in World War II , Vol. IV: Measurement and Prediction, Vol. IV: Measurement and Prediction, Princeton University Press. Reprinted 1973 by Peter Smith, Gloucester MA.
27. Lazarsfeld, P.F. (1950b) Some Latent Structures, (Chapter 11), in: Samuel A. Stouffer, Louis Guttman, Edward A. Suchman, Paul F. Lazarsfeld, Shirley A. Star, John A. Clausen (Eds.), Studies in Social Psychology in World War II , Vol. IV: Measurement and Prediction, Vol. IV: Measurement and Prediction, Princeton University Press. Reprinted 1973 by Peter Smith, Gloucester MA.
28. Lazarsfeld, P.F. (1959) Latent Structure Analysis, in Psychology: A Study of a Science, Vol. 3, S. Koch (ed.). New York: McGraw‐Hill.
29. Lewin, K. (1947a) Frontiers in group dynamics, Human Relations , 1(1), 5–41. Available online at url: <http://hum.sagepub.com/content/1/1/5>. Accessed: February 16, 2016.
30. Lewin, K. (1947b) Group decisions and social change. In T.M. Newcomb and E.L. Hartley (eds), Readings in Social Psychology . Henry Holt and Company: New York, pp. 340–344.
31. McKuen, M. and Brown C. (1987) Political context and attitude change. American Political Science Review , 81(2), 471–490.
32. McPhee, W.N. (1961) Note on a campaign simulator. Public Opinion Quarterly , 25(2), 184–193.
33. McPhee, W.N. (1963) Note on a campaign simulator. In W.N. McPhee (ed.), Formal Theories of Mass Behavior . Free Press, Glencoe, IL; Chapter 4.
34. McPhee, W.N and Coleman, J. (1958) A program of research in mass dynamics. Political Research, Organization and Design , 1(4), 6–10.
35. McPhee, W.N. and Ferguson, J. (1962) Political immunization. In W.N. McPhee and W. A. Glaser (eds), Public Opinion and Congressional Elections . Free Press, New York; pp. 123–154.
36. McPhee, W.N. and Smith, R.B. (1962) A model for analyzing voting systems. In W.N. McPhee and W.A. Glaser (eds), Public Opinion and Congressional Elections , Free Press. New York; pp. 123–154.
37. McPhee, W.N., Smith, R.B. and Ferguson, J. (1963) A theory of informal social influence. In W.N. McPhee (ed.), Formal Theories of Mass Behavior . Free Press, New York. (Chapter 2, pp. 74–99)
38. Merton, R. (1948) Patterns of influence: a study of interpersonal influence and of communications behavior in a local community. In P.F. Lazarsfeld and F. Stanton (eds), Communication Research . Harper and Brothers, New York; pp. 180–215.
39. Orcutt, G.H. (1957) A new type of socio‐economic system. Review of Economics and Statistics , 39(2), 116–123. (Reproduced in Orcutt, G.H. (2007) International Journal of Microsimulation , 1(1), 3–9.)
40. Polya, G. ([1957]1973) How To Solve It . Princeton University Press, Princeton, NJ.
41. Smith, R.B. (1985) Aspects of the perspective and method. In R.B. Smith (ed.), A Handbook of Social Science Methods, Volume 3, Quantitative Methods, Focused Survey Research and Causal Modeling . Praeger, New York; Chapter 1, pp. 59–79.
42. Troitzsch, K.G. (2013) Historical introduction. In B. Edmonds and R. Meyer (eds), Simulating Social Complexity, Understanding Complex Systems . Springer‐Verlag, Berlin; pp. 13–21.
43. Troitzsch, K.G., Mueller, U., Gilbert, N. and Doran, J.E. (eds) (1996) Social Science Microsimulation . Springer, Berlin.
44. Visser, M. (1996) Voting: A Behavioral Analysis. Behavior and Social Issues , 6(1), 23–34.

Notes

1 Details provided by Dr Robert B. Smith in a personal letter transmitted to the author by e‐mail on 4 October 2014, concerning the history of the computer simulation voting choice model elaborated together with William McPhee. Helpful details, precious support and encouragement from Dr Robert B. Smith are hereby gratefully acknowledged.

2 See Figure 1 in McPhee (1963: 171).

3 Surveys fully described in the appendix sections of both Lazarsfeld et al. (1944) and Lazarsfeld and Merton (1954).

4 The technical details in this section are mainly based on: Smith, R.B. (1985) Aspects of the perspective and method. In R. B. Smith (ed.), A Handbook of Social Science Methods, Volume 3, Quantitative Methods, Focused Survey Research and Causal Modeling . Praeger, New York; Chapter 1, pp. 59–79. I gratefully acknowledge the support offered to me by Dr Robert B. Smith and his precious guidance.

5 See Figures 4 and 5 in McPhee (1963: 181–182).