July 28: Building your model

Helge Klapper, Assistant Professor in Strategy, Krannert School of Management, Purdue University

Julien Clement, Assistant Professor of Organizational Behavior, Stanford Graduate School of Business

Link for synchronous sessions: https://hbs.zoom.us/j/97071216335?pwd=MUZRMFB0L2xXUDRxdU9nSHVMNWtYQT09

Meeting ID: 970 7121 6335

Password: tom

Building your model (of social influence)

This session will build upon the “information aggregation” session taught by Helge and Thorbjorn. We will work with one specific type of model (models of social influence), but the goal will be more generally to think about how to build upon prior work to create your own models and use them to understand problems you find interesting.

I am adding a few references at the end of this manual in case you want to have a look at them, but I don’t think you need to. The only two things you will need to look at before the session are:

a)  Videos

Watch this short Youtube video first, and then watch the (arguably less entertaining) video about models of social influence found in this Dropbox module.

b) Colab notebook

As a starting point for thinking about how to extend a model, we will use this Colab notebook featuring a simple model of social diffusion (it actually includes two different models, really!). Please have a quick look through the code to have a general idea of how it works, and answer the following questions:

1. First, let's set leave the diffusion_mode parameter as "contagion" and compare different scenarios. 

1. • Compare the results when we set the contagion_nb parameter to 1 versus 2. What happens?

   • With contagion_nb set to 2, compare the results when setting the density parameter equal to 0.1 vs. 0.2. What happens?

2. Now set the diffusion_mode parameter to "degroot" (for DeGroot social learning).

2. • How do these results compare to the "contagion" results?

   • Now compare these results when setting the density parameter equal to 0.1 vs. 0.2. Does increasing density have the same effect as in the "contagion" case? Why?

3. Based on these results, what types of social diffusion processes are better modeled as a contagion process vs. a DeGroot learning process?