21/07/2009 Meeting With Christina

Following are minutes of the meeting held on July 21 2009 on Agent Based Modeling of Social Capital:
Venue: Residence, Christina Prell

Next meeting C.Prell, M. Pasha 27/07/09

• Pasha to run model for all agents(coding issue) and introduce randomness in tie selection, and further develop the system.
• Christina to send a document on simulation methodologies (Monte Carlo simulation). (Done)



Present
Christina Prell
Mudassar Pasha

Agenda
• Demonstration of implementation of the Model as of now.
• Discussion of strategies used in implementation currently and further ahead.
Discussion and decisions:
Sequential selection of actors during simulation:

The logic of the system developed so far is such that, the OF is calculated for all actor in the first state of the agent. Based on the OF the actors chose a new network option and put a message on the message board notifying of their selected option.
In the next, the actors read these messages sequentially. If the message is targeted at an actor, the recipient picks up the message and updates its tie status with the sending actor.

A concern was raised here: Does this result in a preferential tie formation, i.e by always executing the functions for the actors in a sequence?
For example, if actor one is asked to execute the OF, then actor 2 , then actor 3.. will this result in arising of a preferential network?

o Modeler’s Analysis:

In the first state of the Xagents, the send_tie() function is executed.
In the second state of the Xagent, the function receive_tie() is executed.

In the first state, The OF is calculated for each actor for the current network.
The network remains the same until all actors have chosen their options and put them on the message board.
Agent message: The message contains sent_id(senders id), target id(the intended recipient), changed tie(the new ties status between these two ,that the sender selected).

In the second state, all actors read these messages. If a message is directed towards a particular actor, the recipient picks this message and updates its tie(with the sender) in its network.

Hence it was suggested that the selection of ties is not fully preferential, i.e everyone gets a relatively fair chance to select an options.

Randomness vs. Probabilistic approach:
The Actor selects an option based on the max value of the OF for each network optioin.
This must be randomized.

Solution suggested:

Consider an actor has 5 options with OF values 0.3, 0.1, 0.4, 0.6, 0.7. If a random number X has a value 0.45, which is closest to third option 0.4, then 0.4 is selected as the new network option.

Arguments:
• What if the random number is 0.2? It lies between 0.3 and 0.4. Which option must be selected in this case? Should there be a secondary criteria? Or both options be given equal preference and a selection made at random?
• What if two options have the same values. Say option 1 =0.4, option 2=0.4. Random number = 0.2. Again, it’s a similar situation where I secondary criteria for selection might be required.


Model on Net logo:
• Further meetings would involve discussions on previous attempts on Netlogo.
• It was suggested that the project could be carried out on Netlogo and Flame in parallel, which could help analyse the advantages and disadvantages of both approaches.

Simulation with large number of agents:
• The model must be capable to accommodated a large number of agents against the current scenario which handles 5 agents.
• A pre recommended solution was that the model be run with a small number of agents. The simulation would be allowed to dynamically create new agents. For example, at the end of 100 iterations, there could be 500 agents created. Then, a new run could use the 100.xml file as the 0.xml file which now has 500 agents. This is an easy solution, but a the concern of preferential networks arises here. Would there be many ties clustered around the initial 5 agents?



Tentative agenda for the next meeting
• Implement suggested corrections.
• Run model for multiple iterations and analyze the results.
Please suggest any additions or corrections.