From Beliefs to Attitudes: Polias, a Model of Attitude Dynamics Based on Cognitive Modeling and Field Data
Journal of Artificial Societies and Social Simulation
Attitude is a key concept in social psychology. The paper presents a novel agent-based model to simulate attitude formation by combining a rational and an emotional components based on cognitive, psychological and social theories. Individuals of the artificial population perceive actions taken by actors such as government or brands, they form an attitude toward them and also communicate the events through a social network. The model outputs are first studied through a functional analysis in
... nal analysis in which some unique macroscopic behaviors have emerged such as the impact of social groups, the resistance of the population toward disinformation campaigns or the social pressure. We then applied our model on a real world scenario depicting the e ort of French Forces in their stabilization operations in Kapisa (Afghanistan) between and . We calibrated the model parameters based on this scenario and the results of opinion polls that were conducted in the area during the same period about the sentiment of the population toward the Forces. Our model was able to reproduce polls results with a global error under %. Based on these results, we show the di erent dynamics tendencies that emerged among the population by applying a non-supervised classification algorithm. . This paper introduces Polias, a new agent-based model of attitude dynamics, rooted in sociopsychological theories. Attitude is a central concept to study human behavior. As many constructs in psychology, there are several ways to define it. Broadly speaking, it is "an overall evaluation of an object that is based on cognitive, a ective and behavioral information"(Maio & Haddock , p. ). Allport also suggested that an attitude is a predisposition to act, being "a mental and neural state of readiness organized through experience, exerting a directive or dynamic influence upon the individual's response to all objects and situations with which it is related" (Allport ). Thus, an attitude is an evaluative judgment, and it has a valence to express a positive (in favor), neutral or negative (disfavor) predisposition toward this object. It has also a valence, where one could slightly dislike spinach while another really hates it. Moreover, when several people are interested in an object, and exert a social behavior on this object (share it, exchange opinions about it, etc.), this object will be called (according to social psychology) a social object. . According to scientists in social psychology, the concept of attitude plays a major role in various mechanisms such as the construction of mental representation (e.g. Fazio ), self-maintenance (e.g. Steele ) or behavior (e.g. Fishbein & Ajzen ). This is the reason why Attitude dynamics is one of the trending topics in the field of social simulation (e.g. Castellano et al. ; Xia et al. ; Edmonds ). . However, Chattoe-Brown ( ) points out two critical shortcomings in existing works. First, at the microscopic level, most of the models are not grounded on actual social science theories on attitudes. Indeed, the majority of works are based on the bounded-confidence model (De uant et al. ; that sees attitude formation process as a black box. Second, on the macroscopic level, the studied sociological phenomenon are not confronted to any empirical data such as opinion polls results. These two points will be further discussed in next sections below. . In this paper, ( ) we propose a model anchored on major theories in social psychology that articulates a cognitive and an emotional dimensions of attitudes, ( ) we perform a functional analysis of the model's behavior and ( ) we calibrate the model's parameters in order to reproduce real world opinion polls results given a reconstituted military scenario that took place in Kapisa (Afghanistan) between and . Due to this application context, many examples given in the paper are from the military domain. However, the proposed model is general enough to be used in other context. . In the next section, we will first present attitude dynamics models in social simulation and review some related works in social psychology that constitute the foundation of our model. We will then present our model (Section ) and its functional properties (Section ). This model will be then investigated through a functional analysis and calibrated on a real world study case that we present in Section . Related Works . In this section, we discuss related works on attitude dynamics. We first present existing models in social simulation and we show how the definition of micro-founded models and validation based on empirical studies has developed recently. We then discuss existing models in social psychology that serve as the foundation of our own model. Attitudes in Social Simulation . Many agent-based models of attitude treat the attitude formation process as a black box, and focus on how one individual's attitude is influenced by others. The first models were inspired by statistical physics, with binary valued attitudes, and applied to voting (Galam ). They usually view opinions and attitudes as the same thing. Then models appeared with attitudes of continuous values, like the well-known bounded confidence model (De uant et al. ; Hegselmann & Krause ), where two individuals (selected randomly) have attitude values close to each other (with a fixed threshold), each one modifying its attitude so that it gets closer to its peer's (see e.g. for a review on these models Castellano et al. ( )). This is not a suitable approach for two reasons: ) due to its black-box nature, bounded confidence models the consequences of attitude formation, not the formation process itself (the causes); ) such models are not rooted in any socio-psychological theory, raising the question of their empirical plausibility (Chattoe-Brown ). fist series of action impacting SGA, SGB's and SGC's attitude means are also impacted). This is enabled by the communication network that ties all the groups, the information about an action is propagated amidst the social network. . Also, we can notice that the attitude dynamics progress correspondingly to the inter-group attitudes: when SGA is rewarded, SGA and SGB's attitude increase, in contrast, SGC decreases because they are against SGA. . When agents face events a ecting other social groups (than their own), weakly emotional agents (and balanced ones, to a lesser extent) react less steeply and less rapidly than highly emotional agent. This is because, when the emotion weight is low, the surprise weights a lot, and interest decreases with the social distance between the agent and the one a ected by the event (cf. Appendix). Thus, lesser individuals get to know the information, mitigating in this way the attitude change amplitude.