And You Are?

An agent-based model of interaction at a cocktail party

by Yann Bramoullé and Charlie Williams

SFI Graduate Workshop in Computational Economics

June 20, 2000

What if you threw a party, but everyone just talked to people they already knew? How do people’s initial networks affect the breadth of contacts they have with new people at a party? Can a few party animals tip the balance to get people mixing?

We developed a model of interaction between pairs at a party, which examines how patterns of interaction depend on

• initial social networks
• social attributes,
• and the proportion of extroverts at the party.

We began with the thought that agents that enter with denser initial networks might actually talk with fewer people over the course of the party because they will be more likely to remain within their own network. Our model did exhibit this phenomenon, though under specific conditions.

Agents attend a party to talk with others. They are looking for people they would get along with. They commence interactions at random with other agents, but they interact longer with agents to whom they are closer in social characteristics. In addition, when agents come into contact with a new potential partner, they are more likely to commence a conversation with someone they already know. The level of extroversion determines the probability of interacting with a stranger.

Every round the agents test other agents at random to initiate an interaction. For ongoing interactions, each round each agent tests whether to continue the interaction based on their social affinity (closeness in a 2 dimensional space of social characteristics). The network of dyadic interactions is summed each round to give a final frequency table of each agent’s interactions with others and rounds spent alone.

At any party, agents must spend some time alone. As with unemployment, agents must float between interactions before meeting someone new to talk with. In our model, this appears to converge to a “natural” rate of just under 40% as the initial networks become denser. This is true for all proportions of extroverts, but the curve becomes flatter as more extroverts are introduced.

We are interested in whether agents will meet most other agents at the party, if only briefly. We measured the proportion of interactions between agents over the course of the party. When there are no extroverts, people don’t meet strangers and only spend time with people of their initial network that they like. Thus for introverts, denser initial networks lead to higher levels of interaction at the party. When there are only extroverts, denser initial networks actually reduce the amount of mixing at the party. This results from the low probability of interacting with strangers, even for extroverts: in our model extroverts were assigned a .4 probability of interacting with strangers, and introverts a 0 probability.

The simulation was written in Mathematica 4.0. The code for the simulation is available here.