Research: Stella Models

This paper is in process, is forthcoming as a book chapter. A draft of the paper will be posted soon, when it is finished. This current posting of Stella models is to aid communication with my coauthor in the revision process.

Three Stella files were used to generate the models in the network paper. The archive (about 550 KB zipped) contains three Stella files and two Excel spreadsheets.

Download a save-disabled demo of Stella (necessary for viewing & interacting with these files).

A few notes on reading the files.

Note: All models contain the “useseed” switch. When turned on, the same seed is always used for the random numbers. When turned off, you can see the effect of random variations.

1) Network model revised. This file has three separate models with a wide variety of switches and parameters. All three run at once unless you choose sector runs. a) The simplest model (individual) generates random actions at a fixed rate. If the “feedback” switch is turned on, actors change their probability in response to the sum of all actors’ actions. There are two algorithms, either responding to the total, or the difference between the recent total and the lag total. These are chosen by the weights assigned, in general, one of them should be 0. (A later version uses switches instead.) b) The network model uses the same feedback algorithms, but these are controlled by switches as well as weights. Actors are affected only by people to whom they are connected. Switches permit choices among network structures. c) Media model. Again the same logic of feedback algorithms, but here communication is through news coverage. Three choices are available for how the probability of news coverage is affected.

2) Influence model. This file contains two separate models, one in which network ties are input as 0,1 entries in a who-to-whom matrix, and the other in which the strength of ties are input as probabilities ranging from 0-100 (and the 0,1 matrix at each trial is randomly generated from these probabilities). Again, unless you select sectors, these both run independently at the same time. One of the network options in this file is to accept an input matrix. An Excel file is set up to provide the input in the proper format. This model assumes that influence is through a weighted average of the actor’s opinion level and the average of all those the actor is connected to.

3) Organizing model. This file has one core model which generates a “big action” by increasing the probabilities until a given time period. Influence is a function of the number of organizing contacts. A small submodel generates the random “noise” for the graphic. This can accept spreadsheet input into the network matrix.