BEHAVE is involved in organising a session on “The Lab in the Model, the Model in the Lab” at the Social Simulation Fest 2021, which will take place on Tuesday 16 March 2021 at 1:30-4:45 PM (CET).
Below you find the programme.
For any info, including the registration (free), please visit the event website here.
SocSimFest2021 “The Lab in the Model, the Model in the Lab: The New Frontiers of Experimental and Computational Research on Social Behaviour”
Tuesday 16 March 2021, 1:30-4.45 PM (CET Time)
Chairs: Andreas Flache, University of Groningen, NL (firstname.lastname@example.org), William Rand, NC State University, USA (email@example.com) & Flaminio Squazzoni, University of Milan, Italy (firstname.lastname@example.org)
This workshop aims to present research that integrates experimental and computational approaches to examine the emergence of large-scale, complex social patterns from agent interaction. Social norms, collective opinions and institutions are emergent features of agent interaction, and the analysis of these patterns requires in-depth integration of different methods, observation scales and disciplines. While perspectives and findings from different disciplines are essential to understand social behaviour, a unique combination of analytical instruments of detection is necessary to connect composite and plural approaches. The workshop is addressed to scholars in the experimental and social simulation communities who are interested in exploring this integration and discussing different lines for cross-methodological developments such as: a) agent-based models calibrated with experimental data to perform what-if analysis, robustness and sensitivity tests; b) experimental tests of simulation findings to corroborate causal explanations and dissect alternative behavioural explanations; c) use of agent-based models on experimental data to explore counterfactual analysis and optimization analysis; d) full-cycle analysis of lab experiments to computational models and back to lab experiments; e) integration of artificial and human agents in lab/online experiments.
The lab in the model, the model in the lab: An introduction, Andreas Flache (University of Groningen, NL), William Rand (NC State University, USA) & Flaminio Squazzoni (University of Milan, Italy)
This talk will introduce the section by providing an overview of the advantages and main problems when integrating experiments and ABM.
Data-driven reactive and cognitive agent models replicate differently a long-term experiment on social norms and cooperation under risk, Giulia Andrighetto & Mario Paolucci (CNR, Rome, Italy), Guillaume Deffuant & Omid Roozmand (Université Clermont Auvergne, France)
In this talk, we present a long term experiment on the formation and change of social norms and their effect in promoting human cooperation in situations of collective risk and agent-based simulations based on the experimental data. We consider two data-driven models: reactive agents, which play only from past data and cognitive agents, which also use expectations and normative beliefs. We closely derive the both types of agents from experimental data. Comparing the models provides information about how crucial expectations and normative beliefs are in the emergence of norms.
From the lab to the model: Exploring the dynamics of a social norm of honesty in a participative budgeting setting, Lucia Bellora-Bienengräber (University of Groningen, NL), Kai G. Mertens & Matthias Meyer (Hamburg University of Technology, Germany)
This talk will present an experimental study on the effect of peer pressure (i.e., being observed by and exposed to another peer) and team identity on the activation of a social norm of honesty in a participative budgeting setting leading to increased reporting. First, we conducted an online experiment with a 3 (peer pressure) x 2 (team identity) between-subjects design in which 110 participants were grouped into teams of two and were each responsible for a project for one period. Both participants acted as managers and had to report the true project costs to the headquarters to get funding. Since the headquarters does not know the true project costs, we allow participants to increase their earnings by overstating the costs that are needed for the project. Our results show that consistent with a social norm activation theory (Bicchieri, 2006), participants tend to have higher empirical and normative expectations and report more honestly when they can be observed and socially sanctioned by their peer. Further analyses indicate that peer pressure affects honesty conditionally on individuals’ personal normative beliefs regarding honesty and that the effect of peer pressure is mediated by empirical expectations. Second, we present our ideas to use the data from our experiment as input for our subsequent agent-based model (ABM) to explore the dynamics of a social norm of honesty in a participative budgeting setting. The experimental findings will be used to build an agent-based model to explore the role of temporality, networks, and heterogeneity.
When laboratory experiment confirms agent based simulations: Positivity bias without self-enhancement, Guillaume Deffuant, T. Roubin, Silvye Huet, A. Nugier & S. Guimond (Université Clermont Auvergne, France)
Social psychologists have established that we all tend to overestimate ourselves; we tend to be overoptimistic about our capacities and qualities. This observation is generally explained by self-enhancement: we tend to dismiss our involvement in negative experiences and to overestimate our involvement in positive ones. Yet, in a recent agent-based model, a positivity bias on self-evaluation emerges without self-enhancement, as soon as the sensitivity of the self-evaluation to the feedback decreases when the self-evaluation increases. This is a simple mechanical effect. It is of limited amplitude in the short term, but simulations suggest that it can produce dramatic effects on collective dynamics in the long run. Hence we designed and performed an experiment that aims at checking if this positivity bias without self-enhancement is present in human subjects. A pilot experiment involving 200 participants has been completed and the main experiment on 1500 participants is currently being processed. The results of the pilot experiment confirm that, in the conditions where self-enhancement is not activated, the sensitivity of self-evaluation to feedback tends to decrease when the self-evaluation increases. Some positivity bias can be detected in these conditions.
Micro-foundations of polarization In online social media: A model In the lab fed back into the model, Marijn A. Keijzer (University of Groningen, NL), Michael Mäs (Karlsruhe Institute of Technology, Germany) & Andreas Flache (University of Groningen, N)
This talk will present a full integration of agent-based modelling and experiments to explore micro-foundations of polarization in online social media. With a survey experiment, we measured the experienced shift in opinion after being exposed to an argument that is either in line, or discordant with participants´ initially held opinion. In the first experiment, differences between liberals and conservatives in receptiveness to arguments built on certain moral foundations were tested. In the second experiment, we manipulated social distance to account for preferences of belief alignment with similar alters. Bayesian models for censored outcome data were estimated to acknowledge the fact that opinion scales are bounded, and may thus hide individual’s tendency to indicate the same or a less extreme version of their opinion once negative influence would push them to an extreme end of the opinion scale. Results confirm that opinion distance is related to the receptiveness to novel arguments. Social distance and moral framing of argumentation can amplify those effects and provide indicative evidence that—under certain circumstances— negative influence may occur. Finally, we feed the results of our empirical study back into an agent-based model for opinion formation in online social media and show that the context of social networking websites has profound impact on the opinion dynamics compared to offline interaction. The results from this study are discussed in light of the prominent hypothesis that personalization promotes polarization. Our study suggests instead that filter bubbles created by personalization may actually prevent polarization.