Search for content and authors
 

Conformity in numbers: How to validate social agent-based models

Piotr Nyczka 1Katarzyna Byrka 2Paul Neil 3Katarzyna B. Sznajd-Weron 4

1. Jacobs Univiersity Bremen, Bremen 00-768, Germany
2. University of Social Sciences and Humanities, Institute of Social Psychology (SWPS), Aleksandra Ostrowskiego 30b, Wrocław 50-001, Poland
3. University of Central Arkansas,, Arkansas AR 72035, United States
4. Politechnika Wrocławska (PWR), Wybrzeże Wyspiańskiego, Wrocław 50-370, Poland

Abstract

Ample empirical evidence shows that descriptive norms, understood as "how most people behave in a given situation", are a powerful means to influence compliance. The problem is what does it mean "most" and this is far from being clear. Is it just an absolute majority, unanimity or maybe a certain threshold like 75%? Although effectiveness of normative messages have been studied thoroughly, researchers operationalize norms and select their numerical representations according to undefinedrules. In most studies, numerical representations of norms come from real data. When the access to real data is limited, the decision about numbers representing norms is taken arbitrary. The question that naturally arose here is the following: Is there anything like the critical threshold for the majority and if yes, what the value of this critical threshold?

Usually, ABMs are defined as computer simulations of social interaction between agents (e.g., individuals, firms, or states), embedded in social structures. Such models are aimed to observe, analyze and understand the emergence of aggregate outcomes, such as public opinion, results of voting, diffusion of innovation, cultural or political revolutions, international terrorism, social inequality, urban ethnic segregation etc. Although, the origins of ABMs in sociology can be traced back to the 1960s, it was only from the 1990s that ABM applications reached a critical mass. Almost simultaneously, yet somehow independently a new field of sociophysics, i.e. applications of statistical physics in social sciences, emerged.

The essence of statistical physics is to understand collective phenomena (macroscopic level) on the basis of interactions between basic elements of the system (microscopic level). Therefore, all models of statistical physics can be viewed as ABMs, although in physics this type of approach has been known for years under the name of microscopic modeling. Because physicists have gained much experience and developed many computational and analytical tools to study collective phenomena, the temptation arose to apply methods of statistical physics outside of physics.

As in physics, most of sociophysics models are relatively simple and often can be viewed rather as toys than tools. Although the role of the toy models in the development of statistical physics cannot be overestimated, yet the main challenge that persists with social ABMs is the describing properly of complex social systems in terms of a relatively simple approach. In other words, our main goal is not to draw the precise portraits of real social systems but rather caricatures. However, these caricatures should be as good as possible and therefore there is a strong need of validation not only of the model's assumptions but also outcomes.

During this lecture we show how ABM's can be validate on the basis of social experiments. Asan example we will discuss a model of opinion dynamics with three different types of social response: conformity, anticonformity and independence. We show how to build a model in agreement with social theories and experiments, but first of all we will show how to validate results given by the model. Finally, we show how ABM’s can help to answer the questions pose by social psychologists, such as one on the existence of the critical threshold for the majority.

 

Legal notice
  • Legal notice:
 

Related papers

Presentation: Invited oral at Econophysics Colloquium 2017, Symposium A, by Katarzyna B. Sznajd-Weron
See On-line Journal of Econophysics Colloquium 2017

Submitted: 2017-03-30 20:28
Revised:   2017-04-09 20:11